0:00.000–0:08.880
enHi everyone. I am so excited to be here. Today we are releasing our latest and most capable models.
0:08.880–0:18.080
enGPT-5.6, Sol, Terra and Luna. Sol will be available over the next 24 hours to all our paid plans.
0:18.080–0:26.080
enAnd Terra and Luna are also coming to all our free users. Today almost a billion users use
0:26.080–0:34.160
enChatGPT every week. They use it for helping getting them to write, getting answers, do research and all
0:34.160–0:40.000
ensorts of creative things. But what we've seen and especially with codecs and the coding abilities
0:40.000–0:45.920
enis that these models are capable of much more. We saw a transformation of software development
0:45.920–0:52.560
enand shortly after internally within OpenAI everyone was using these models to do so much more than
0:52.560–0:59.360
engetting great answers. They were using these models to do great things. And you were expecting that we
0:59.360–1:04.320
enwere going to release these models and we are rolling them out over the next 24 hours. But why I'm really
1:04.320–1:10.960
enexcited to be here today is we're not only releasing the models, we're also announcing three major product
1:10.960–1:19.120
enupdates. The first one is ChatGPT Work. ChatGPT Work is your new partner for ambitious work. It works on
1:19.120–1:27.920
enweb, it works on mobile, on the go, and also on desktop. The models with GPT-56 Sol have really been a breakthrough
1:27.920–1:35.840
enin their ability to do things of complex nature for you. Understanding very complex content, staying on
1:35.840–1:42.800
entrack for long horizon tasks, understanding spreadsheets, creating them, editing even websites. And we're going to
1:42.800–1:51.600
enbe showing you first ChatGPT Work. We're also announcing an all brand new ChatGPT desktop app. It goes even further
1:51.600–2:00.800
enand allows ChatGPT to work on your local files, your apps, and your browser. Finally, we're also announcing
2:00.800–2:07.920
enhosted sites, which we are making available for all paid users. It allows you to use ChatGPT Work to create and
2:07.920–2:13.920
enshare those websites with your friends and colleagues. It is a truly magical experience and I can't wait to
2:13.920–2:21.600
enshow you. First, we're going to have a look at ChatGPT Work. Jessica, Lauren, why don't you show us what
2:21.600–2:27.920
enthey can do? Hi, I'm Jessica, engineer on ChatGPT Work. And I'm Lauren, I'm on our strategic finance team.
2:27.920–2:33.680
enSo we're going to show you the new experience on web and mobile. So let's take a look at mobile. So here we have
2:33.680–2:40.000
enChat. Chat is the familiar thing you love, you know and love. It's really great for quick answers,
2:40.880–2:47.120
ensearch and brainstorming ideas. But today we're also introducing work. And work is where I go to
2:47.120–2:56.640
enhave ChatGPT help me just get stuff done. So take, so for example, something that I did right before
2:56.640–3:05.360
ento prepare for this launch is, hey ChatGPT, I would love to get an understanding of how people are using the
3:05.360–3:13.120
enChatGPT Work feature internally. Look through Slack and employee feedback and find people in diverse roles
3:13.920–3:19.920
enthat have used the product in really interesting ways. I'm in San Francisco this week for the launch. I would love to
3:19.920–3:26.560
enmeet up with a few people and really deep dive on their use cases in person. So I can send this through
3:26.560–3:31.440
enmobile, but these conversations also appear on web. And this is something that I actually did
3:32.320–3:38.720
enearlier. We can take a look at the results and ChatGPT Work was able to pull from the sources that
3:38.720–3:45.680
enI asked for, find interesting people to talk to, and then schedule meetings with them while I'm in town.
3:45.680–3:48.240
enAnd one of the feedback posts it highlighted was actually Lauren's.
3:48.240–3:53.520
enYeah, our finance team have become such power users of ChatGPT Work. I'm really excited to show you how
3:53.520–3:56.480
enwe've been using it, but I would love to hear how other teams have been using it too.
3:56.480–4:01.680
enYeah, we've talked to people throughout the company, like recruiters that are using scheduled tasks to
4:01.680–4:07.360
entrack interview progress and make sure that we're getting feedback timely, and data science who are
4:07.360–4:14.400
enusing our new visualized feature to just add context for quick ad hoc requests. But I think the finances
4:14.400–4:16.960
enexperience was one of the ones that was most compelling to us.
4:16.960–4:22.320
enYeah, I'm excited to show it to you guys. So finance may not seem like the flashiest demo we
4:22.320–4:27.200
encould be leading this off with, but ChatGPT Work has really enabled us to run with such lean and
4:27.200–4:31.200
enefficient teams. And we wanted to show you guys a little bit of a peek behind the curtain at how
4:31.200–4:36.800
enwe've been able to do that. So one of the critical roles of a finance team is being able to explain not
4:36.800–4:42.240
enonly our recent trends, but what they mean for forecast in real time. This takes a couple minutes to run.
4:42.240–4:46.880
enSo we did run this earlier today, but let's say we just ran, sorry, we just closed June,
4:46.880–4:51.840
enand we beat our forecast by $2 million. And I do need to caveat these are demo numbers.
4:51.840–4:55.600
enThey don't look like the numbers that you shared yesterday. So I'm glad we're not sharing the
4:55.600–4:56.240
enreal numbers here.
4:56.240–5:02.080
enTivo obviously beats this forecast by more than $2 million. These are demo numbers, but a very
5:02.080–5:07.920
enreal workflow. So we ran this earlier today. This used to take so much manual work. We would have
5:07.920–5:13.920
ento reconcile multiple systems, our Excel forecast model across multiple cases. And now in OnePass,
5:13.920–5:19.040
enChatGPT can run the variance analysis for me so we can see why we beat our forecast and where we still
5:19.040–5:26.320
enhave risk. I can even have ChatGPT work go in and propose an updated forecast case for us. So we had it
5:26.320–5:33.760
endo this earlier, or we had it run yesterday. It went through and updated our Excel model. Let's see.
5:34.400–5:40.160
enPerfect. Yep. We've got our updated revenue in here, our updated forecast. But I can't just walk
5:40.160–5:44.720
enTivo and Jessica and our business partners through an Excel model. So we also had it put together
5:44.720–5:50.480
ena PowerPoint presentation that we can use. We've got our latest forecast. We've got our driver's bridge on
5:50.480–5:55.520
enwhat's changing in our outlook. And ChatGPT work will meet you where you are. It's obviously,
5:55.520–6:00.160
enit's great at Excel and PowerPoint, but our finance team have become such big proponents of sites.
6:00.800–6:05.760
enIt's such a flexible interface to be able to do custom dashboards and really just like
6:05.760–6:12.400
enstorytell around your analysis. So I also asked ChatGPT work to go in and make us a site with the
6:12.400–6:17.760
ensame analysis that we can share out. We've got our updated forecast. We've got our driver's bridge.
6:17.760–6:23.360
enThis is looking good. It's good to go. All right. So we've run our variance analysis. Our forecast is
6:23.360–6:28.640
enupdated. We've got our overviews back for business partners. Let's go ahead and have ChatGPT work
6:28.640–6:34.480
ensend this to Tivo. Hey, Chat, can you send the site link over to Tivo on Slack? Thank you.
6:36.640–6:41.680
enAnd we'll send that off. I hope I get that soon. Yeah. It is such a delight to receive these highly
6:41.680–6:48.880
eninteractive and compelling websites. In ChatGPT work, we've seen it perform very similar things
6:48.880–6:54.960
enacross different use cases. Jessica was using it to understand user feedback and preparing ahead of
6:54.960–7:00.560
enthe launch, even scheduling interesting chats with her colleagues in one-on-one through Calendar.
7:02.320–7:07.840
enAnd that was Jessica. But Lauren used it for a very different kind of purpose across finance and very
7:07.840–7:13.280
encomplex workflows where precision and correctness is extremely important. ChatGPT work was able to
7:13.280–7:20.080
enunderstand complex financial data and represent it in the right format at the right time, just with a few
7:20.080–7:29.120
ensteers. Now, I'm excited for us to show you the all-new ChatGPT desktop app. Andrew, Ed, show us what we can do with it.
7:29.120–7:34.800
enYeah. Thanks, Tivo. This is the new ChatGPT desktop app. So everything that you saw on web,
7:34.800–7:40.800
enit's available here, plus a lot more on your local machine. So local files, your browser tabs,
7:41.440–7:46.400
eneven other apps on your computer, they are all available now to ChatGPT. So let's take a quick look.
7:46.960–7:53.360
enI'm going to start with this spreadsheet right here that my colleague Nick sent me. It's an export of
7:53.360–7:58.320
enuser tickets from our ticketing system. And it's pretty large. It's got a ton of columns. It would take a while to
7:58.320–8:01.600
enget through this and kind of tease out what the themes are and what we need to pay attention to.
8:01.600–8:07.120
enA lot of feedback. It's a lot of feedback. We're very used to this. But instead, what we're going to do
8:07.120–8:16.160
enis we're just going to drag it in. We're going to say, hey, ChatGPT, can you make a quick interactive
8:16.800–8:22.720
envisualization of this feedback so that we can sort of get the themes and the action items and the
8:22.720–8:26.080
encriticality ranked and sort through it all a little bit better?
8:26.080–8:33.760
enSo Chat's going to go and take this spreadsheet, synthesize it, the content, and then make us a
8:33.760–8:38.240
envisualization. It'll take a second. So I'm going to walk you through something that I'm doing for next
8:38.240–8:45.680
enweek's launch. So I've got this folder over here. And this has a variety of material from various teams
8:45.680–8:52.080
enthat have been sent to our team here. This is a launch readiness PDF. And it's got a lot of stuff in
8:52.080–8:59.600
enit. Then we have over here, we've got UXR. They did a bunch of interviews with users on our early products.
9:00.880–9:05.280
enWe've got a security review. This has all of our pen test results and even some of the compliance
9:05.280–9:10.000
encontrols we need to meet. So this folder's got a ton of stuff from various teams. They've just been
9:10.000–9:16.480
enkind of sending it to us in prep for this launch. And I need to brief my team on Monday on the state
9:16.480–9:22.560
enof the launch. And so what I did this morning is I came in and I said, can you please look at the
9:22.560–9:28.080
enmaterials in this folder that everybody sent me? I also have three open tabs in Chrome that have
9:28.080–9:32.880
encontent for this launch. Go look at it and make me a presentation that I can give to my team in the
9:32.880–9:38.400
entemplate that we use all the time at the company. And so it worked for about 90 seconds here, a little bit
9:38.400–9:45.440
enless. And at the end of this, looked through my desktop, looked in Chrome at the tabs that were open,
9:46.240–9:51.680
enand it produced a fully ready slide deck for me to present to the team about this launch.
9:51.680–9:57.440
enThis is one thing that GPT-5-6-Soul is really incredible at. It went through all of that content
9:57.440–10:02.400
enin 90 seconds, synthesized it all, adhered to the template, and created like a slide. You can just
10:02.400–10:06.880
enshare that directly with the team. You don't have to do anything else. Exactly. And it used
10:06.880–10:11.680
enit used its memory as well because I've been chatting a lot about this launch. And so it's
10:11.680–10:15.920
engot some of the concerns that I have that haven't actually been shared via Slack or email too.
10:17.600–10:22.080
enBut I know that I know that you were curious about the next World Cup game and you were kind of busy
10:23.680–10:29.760
enprepping for this live stream. So we've got this we've got this really great feature in the app
10:29.760–10:33.840
ento do quick searches and anything that you want to know really quickly and get instant results.
10:33.840–10:37.680
enSo I'm going to go ahead and see what I'm going to do. So I'm going to go ahead and see what I'm going to
10:37.680–10:38.640
endo with the next one. So I'm going to go ahead and see what I'm going to do with the next one.
10:38.640–10:40.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
10:40.640–10:42.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
10:42.640–10:44.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
10:44.640–10:46.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
10:46.640–10:52.080
enSo this is the ChatGPT search experience that everyone's come to know and love. It's almost
10:52.080–10:58.640
eninstant results. It's got rich search search widgets here and anything you might want. But
11:00.160–11:07.520
enback to work. I've got this Notes app situation that I don't think I'm unique in. My Apple Notes is
11:07.520–11:11.840
enjust kind of a brain dump and it's all over the place and there's no folders. There's no organization.
11:11.840–11:16.240
enAnd in a previous life, I would just kind of leave it this way and declare bankruptcy and
11:16.960–11:22.400
enit'd be sort of a mess. But with the new ChatGPT desktop app, it actually has access to the other
11:22.400–11:30.240
enapps on my Mac. Organize my Notes app. It's a whole thing right now. Make some folders. Move the notes
11:30.240–11:34.880
enaround. Use your best judgment. I don't care. I just need to get out of this mess.
11:34.880–11:44.720
enSo what's going to happen here is ChatGPT with the incredible computer use that's built in and with
11:44.720–11:50.640
enthe advances in the new model, it's going to get its own cursor. It's going to start operating Apple
11:50.640–11:57.040
enNotes in the background. You can see here, this is not my cursor. This is ChatGPT's cursor.
11:57.040–12:03.120
enYeah. There it goes. There it goes. It's going. It's moving Notes around. I can do my thing over here.
12:03.120–12:08.160
enI can check the Belgium game and when that is. And it's just going to go to town and make some
12:08.160–12:14.240
enfolders and drag some notes around. We don't have to watch this whole thing because I'm really actually
12:14.960–12:18.960
enquite curious at what Ed and team have been cooking. Can you show the visualization first?
12:18.960–12:19.280
enYeah.
12:22.080–12:25.120
enSo we were curious about all of this feedback that we had.
12:25.120–12:31.280
enYeah. So here's the visualization I made from the spreadsheet. We've got, it's like rich. It's on demand.
12:31.280–12:36.400
enIt uses the theme that we're using in the app. And it kind of went down and ranked. Like,
12:36.400–12:41.680
enhey, here are the buckets. Here's how important they are. It's interactive. It's not static. So I can,
12:41.680–12:48.160
enyou know, change views. I can ask it for changes. And if I want to share this with my team,
12:48.160–12:54.000
enbecause the spreadsheet's hard to read, in one click, I can just publish that to a site.
12:54.000–12:57.040
enThat's all. And then they get this exact same interactive visualization.
12:58.880–12:59.440
enOver to Ed.
12:59.440–13:00.880
enThere are a few more examples, I think, right?
13:00.880–13:11.200
enYeah. That's a good call out here. This is a set that I ran last night, just kind of playing around
13:11.200–13:17.840
enwith it. These things are stunning. I mean, we didn't write any code to make these. The model is
13:17.840–13:24.960
enwriting these on demand. The 3D ones are incredibly immersive, interactive. It just uses it as part of
13:24.960–13:29.120
enits answer. So it can sort of like model it, like whether it answers in text or answers just like with
13:29.120–13:34.960
ena little visual, like on the fly. Yeah. You can, so you can ask explicitly for a chart or visualization.
13:34.960–13:40.880
enOr sometimes if you ask to, you know, get educated, if you're like, hey, teach me about how this thing works,
13:40.880–13:46.640
enit'll come back with this interactive demo. Or if you say, help me sort through my inbox, it might even give you
13:46.640–13:50.720
enlike a rich widget that shows you the messages and stuff. It's awesome.
13:54.640–13:58.880
enCool. Take it away. Hi, my name is Ed. I'm a designer on the team. And I want to show you a little
13:58.880–14:04.640
enbit about how I use it and the team uses it in our day to day. So, you know, you walk through an example
14:04.640–14:10.480
enof creating assets for a launch using files on your computer. I've done the same thing here, except I've
14:10.480–14:15.520
enused our new sites feature. So one thing I really want to highlight here is I kicked this prompt off
14:15.520–14:20.080
enthis morning. And it's looked through all of my connectors and plugins that we talked about earlier.
14:20.080–14:24.800
enI've connected my Slack, my Gmail, and everything that I use every day. And it's gone through all of
14:24.800–14:30.640
enthose sources. And it's come back and turned it into this like super rich interactive website that
14:30.640–14:34.160
enI can go in. And the reason I want to show you, you know, one big reason I want to show this is just
14:34.160–14:39.680
enbecause, you know, this is just one prompt kicked off. And the visuals are really incredible. I mean,
14:39.680–14:45.520
enthe new 5.6 Sol model. Did you give it a Figma for this? No, no Figma. This was all just the model.
14:45.520–14:50.560
enYeah. And, you know, as a designer, there are like little things you look out for, like small motion,
14:50.560–14:56.800
enright, the typography, the, you know, the 3D motion and all that kind of thing. And it's always best when
14:56.800–15:00.640
enyou give it a little bit of guidance and you, you know, you bring it along, but it's really outstanding
15:00.640–15:06.400
enout of the box. And again, this is just like a really fun way that teams have used to use this
15:06.400–15:09.680
enfeature across the company. But, you know, we're on live stream, we want to make it a little bit
15:09.680–15:16.640
enmore exciting. So what I annotate this, and let's change this header to something a little bit more
15:16.640–15:20.880
enfun. I think, you know, we saw some of the 3D visualizations that you showed, and I think the 3D,
15:21.600–15:28.320
enthe 3D stuff is super fun and interactive. So, hi, TragiPT, let's change this website. So instead of
15:28.320–15:33.040
enhaving the static hero at the top with these planets spinning around, let's do an interactive game.
15:33.040–15:37.120
enSo I want to have like a little character that I can walk around with. And, you know, based on where
15:37.120–15:42.000
enI go, that will then take me to different parts of the live stream prep. So there might be like
15:42.000–15:47.680
enlaunches and the run of show. Yeah, take, you know, it might take a little bit of time, but really,
15:48.800–15:49.600
engo ahead and start that.
15:52.080–15:57.200
enSo while that's kicked off, I just want to show you a few other great examples. You know, Lauren mentioned
15:57.200–16:01.840
enthat their team use a lot of these for like dashboards and internal tools. That's definitely something
16:01.840–16:06.400
enthat we've seen a lot. And you can see a lot of different examples from across the company here.
16:06.400–16:11.120
enA few just to call out. So the web team, the OpenAI.com team who build all of our amazing websites,
16:11.680–16:16.320
enthey, instead of using like an Excel spreadsheet, they use this interactive tool where they can
16:16.320–16:20.240
enupdate it all collaboratively. They can share it with each other. And you can just hop in, you can
16:20.240–16:25.200
ensee what's launched and when it launched, you know, hover over. It's like so interactive. And again,
16:25.200–16:29.680
enso much nicer to look at rather than a spreadsheet. This really illustrates it super well.
16:29.680–16:33.520
enYeah. And it's really been, you know, night and day, literally three months ago, spreadsheet today,
16:33.520–16:36.960
enthese like, you know, it's really like transformed the way that the whole company works.
16:37.920–16:42.400
enThis is another great example. This is the ChatGPT images team. So they've been collecting great
16:42.400–16:47.600
enexamples of how ChatGPT images have been used across all of our campaigns around the world.
16:47.600–16:51.840
enAnd when building sites as well, and any kind of front end stuff, you know, we now have access to
16:51.840–16:55.760
enChatGPT images as well. And it just also, you know, makes the whole experience like so much better.
16:55.760–17:00.720
enAs a designer, this is one thing that I quite like to do a lot. So as part of the new launch,
17:00.720–17:04.720
enso you're not looking at, you know, ChatGPT or the ChatGPT desktop here, you're looking at
17:05.360–17:09.920
enan interactive prototype that one of the designers on the team, Tarek, built, and I've been working
17:09.920–17:15.680
enwith him on. And it's a new way of prototyping our new model selector. You know, previously, I might have
17:15.680–17:21.360
enjust sent an image or a design file to folks. Now I can show an interactive demo. I think I sent you this,
17:21.360–17:25.120
enand you're like, whoa, this is like, I can like click in and test it out. It's so much better to
17:25.120–17:28.640
enillustrate like an idea. And then, you know, you just do it super quickly in a couple of minutes,
17:28.640–17:32.720
enand then boom, you have it. Yeah. And then you share it with, you know, you just get the feedback
17:32.720–17:37.120
enyou need, you incorporate it back. If you need to prototype a toggle or something. Yeah. And the
17:37.120–17:41.600
endesign team and like all teams across, across OpenAI really, you know, our work has changed so much.
17:41.600–17:47.120
enAnd it's really been facilitated by these new models. And this is finally a bit of a fun one. So,
17:47.120–17:52.480
enyou know, in the AI community, a good test of kind of how good a model is at front end is this,
17:53.760–18:00.080
enthis test where you can see if it can draw a pelican riding a tricycle in an SVG. But Kian,
18:00.080–18:05.760
enone of the folks on the team, he built this fun prototype where he built a 3D version. And again,
18:05.760–18:11.600
enthis is just all 5.6 Sol, you know, cooking away, building out this amazing prototype. And the really
18:11.600–18:14.800
encool thing about Scythe is they're collaborative, you can share them. And he shared this with people
18:14.800–18:20.480
enacross the company and they've kind of added their own features. So, you know, the, it can ride a pony,
18:20.480–18:25.520
enyou know, it can ride all sorts of crazy things, it can ride another pelican. So this is a light
18:25.520–18:30.240
enhearted example, but just like as a creative, it's just so amazing how much it's expanded things.
18:31.040–18:37.920
enSo I kicked off that prompt this morning. So I'm just showing, showing another example here of an output.
18:37.920–18:42.480
enAnd that's what we just, yeah, this is, this was, so I kicked the same prompt off just before we started.
18:42.480–18:47.040
enAnd again, you can just see just how amazing it is out of the park. So it's the same website.
18:47.840–18:52.960
enBut now you can kind of go through and it's a bit more interactive and fun, but really just demonstrates
18:52.960–18:55.920
enthat just in a few prompts now, kind of anyone can really build anything.
18:55.920–18:59.440
enIt's really about raising the level of ambition that you have, right?
18:59.440–19:00.160
enYeah, totally.
19:00.160–19:06.480
enThanks so much, Ed. Thanks so much, Andrew. Everything you saw here was really made possible
19:06.480–19:13.520
enfor Jessica, Lauren across finance, getting help to schedule interviews, understanding things
19:13.520–19:19.920
enin the stress of a launch. And then here, just getting help to think ambitiously and executed
19:19.920–19:25.280
enon it really fast. All of that is made possible thanks to the model. And I'm here joined by Katie
19:25.280–19:28.800
enand Tejal to talk a little bit more about the research, which I'm tremendously excited about.
19:28.800–19:33.200
enAnd one thing really also to insist on is like, it's not about this, not just about front end.
19:33.200–19:37.360
enThis model is our most capable coding model. And it's also incredible at cyber, which we're going
19:37.360–19:38.160
ento talk about as well.
19:38.160–19:44.080
enWell, I'm Katie. I'm a researcher at OpenAI. And it's pretty crazy to think that we've been here
19:44.080–19:46.240
enjust for a year since the codex launch.
19:46.240–19:47.840
enI think we were like precisely almost like a year ago.
19:47.840–19:49.280
enLiterally right here. Yeah.
19:49.280–19:54.240
enHey, I'm Tejal. I'm also a researcher. And we're all so excited to share more about how we trained
19:54.240–19:58.720
enGPT 5.6. It's truly been a labor of love from our whole team across OpenAI.
19:58.720–20:04.160
enIn 2024, we announced the reasoning paradigm, which is a reinforcement learning technique
20:04.160–20:08.960
ento tackle the hardest tasks. And since then, we've been scaling both reinforcement learning and
20:08.960–20:13.600
enpre-training. And we've seen exponential improvements in model capabilities as these two
20:13.600–20:20.240
enmultiplied together. And GPT 5.6 is the latest result of all of our research progress to date.
20:21.840–20:25.280
enThe 5.6 family brings these frontier capabilities to you.
20:25.280–20:28.720
enSOL is our most powerful model for the hardest agentic workflows.
20:28.720–20:33.600
enTerra is a faster model for everyday workflows. And Luna is our fastest and most affordable model
20:33.600–20:38.720
enfor high volume work. Already, SOL has been transforming our research program. As one example,
20:38.720–20:45.360
en5.6 SOL actually autonomously post-trained Luna. This is the actual codex prompt that we,
20:45.360–20:50.480
ena researcher on our team, used to have SOL kick off a post-training job for Luna. You can see here
20:50.480–20:55.120
enthat, you know, it's like asking to find the training configs, find the right GPUs for the job,
20:55.120–20:59.040
enand then launch the script and make sure that it works. It's a really short prompt.
20:59.040–21:03.120
enYeah, it's pretty crazy that we've gotten far enough where you can have a fairly underspecified prompt
21:03.120–21:08.160
enand then just give it to codex and then just have it go and run the job. And, you know, previously,
21:08.160–21:13.200
enthis is something that a team of senior researchers may have worked on at OpenAI. And now it really
21:13.200–21:17.920
enfeels like the automated researcher is pretty close. This isn't the only place we've seen
21:17.920–21:22.560
eninternal acceleration. We're also seeing an increase in the number of pull requests per researcher,
21:22.560–21:26.800
enthe number of experiments researchers can run so we can test our ideas and turn them into research
21:26.800–21:31.440
enfindings. And actually, codex made every plot in this live stream and also helped us make the slide
21:31.440–21:38.880
endeck. So thank you, 5.6. Across Frontier evals, 5.6 Sol is state of the art. And this is something
21:38.880–21:43.600
enwe're really excited for you to finally experience and see the speed up that we've already experienced
21:43.600–21:48.560
eninternally. On Terminal Bench, with the test of coding performance, the model is state of the art. On
21:48.560–21:54.400
enBrowse Comp, which is an evaluation that sees if the model can locate hard to find information. And the
21:54.400–22:00.480
ensame with agent's last exam, which is long horizon professional work. As one deep dive, 5.6 is
22:00.480–22:04.960
enparticularly good at computer use. So anything that involves navigating a browser, navigating apps on
22:04.960–22:10.000
enyour desktop, and that enables all sorts of digital work. It helps medical assistants navigate electronic
22:10.000–22:15.040
enhealth records. It helps data scientists analyze if a drug is effective. It can even help investment
22:15.040–22:20.000
enbankers create financial models. And you can also use this for personal use. It certainly helped me
22:20.000–22:24.240
enorder food. And all of this is something that when we ask experts in the field, you know, how they're
22:24.240–22:28.560
enusing this model in their everyday workflows, they say it's better than anything else they've ever
22:28.560–22:32.960
enused anything else on the market, while being three times as fast. So we're really excited for you to
22:32.960–22:37.760
enfinally try and feel how powerful this model is. You might hear the capabilities that Tayal is talking
22:37.760–22:42.560
enabout and that you saw demoed earlier and think that this is going to come at a very high price tag.
22:42.560–22:47.600
enBut that's actually not the case with GPT 5.6 Sol. So token efficiency has been a huge focus of our
22:47.600–22:52.480
enresearch program for several years. And with every model, we want to bring more intelligence for every token.
22:52.480–22:57.760
enAnd you can see this reflected on evals like Deep Suite 1.1, where GPT 5.6 outperforms its
22:57.760–23:03.840
encompetitors at less than half of the cost. But if you want the best intelligence money can buy,
23:03.840–23:08.480
enwe're also announcing Ultra Mode, which unleashes a whole team of agents to do work for you. So in
23:08.480–23:13.680
enthis example, eval, you can see here, you know, one agent does better with time. But as we add more
23:13.680–23:18.400
enagents, the model is able to do even better and faster. And that's because the agents are able to
23:18.400–23:24.400
enparallelize work like an experienced team. This is a capability folks have asked for so long that
23:24.400–23:29.360
enwe bring to the Codex app. And I'm so excited that we're finally putting it out there as Ultra Mode.
23:32.000–23:37.360
enOne more improvement that we've shipped with GPT 5.6 is, you know, we may remember that in GPT,
23:37.360–23:41.360
enwhen we launched GPT 5.5, we had to ship it with a dev message saying, you know, don't talk too much
23:41.360–23:45.200
enabout the goblins and the gremlins. And this is a result of reward hacking that occurred during the
23:45.200–23:51.280
entraining of GPT 5.5, which has now been resolved in GPT 5.6. And now GPT 5.6 will only talk about
23:51.280–23:57.120
engoblins a tasteful amount when it's cute or appropriate. With these more capable models,
23:57.120–24:02.400
ensafety and alignment are more important than ever. That's why we spent over 700,000 A100 equivalent
24:02.400–24:07.680
enhours of compute on model red teaming. We also spent six weeks safety training and testing this model,
24:07.680–24:12.080
enand also have incorporated multiple novel monitoring upgrades in our launch, including
24:12.080–24:17.920
ennew classifiers and activation probes. Ultimately, we really believe in iterative deployment and putting
24:17.920–24:22.800
enour models in the hands of real people as quickly as possible so that we can learn from real world
24:22.800–24:26.960
enusage outside of our labs. And that's why we have initiatives like Project Daybreak, where we give
24:26.960–24:33.520
encybersecurity researchers access to our models earlier on. And with GPT 5.6 Sol, researchers have
24:33.520–24:39.200
enalready found vulnerabilities in every major browser and database. Yeah, as part of the Daybreak umbrella,
24:39.200–24:44.800
enwe started Patch the Planet, which is an even broader initiative, where we work directly with open source
24:44.800–24:49.520
encontributors and projects. And we also generate high quality patches. For example, for Linux,
24:50.080–24:54.320
enthey accepted over half of our patches, which indicates that not only do we find critical
24:54.320–24:58.800
envulnerabilities that weren't found for a long time, but also we are able to automatically patch them in
24:58.800–25:04.400
engreatly accelerating cyber defense. We're very proud of this model. We hope it helps you as much as it's
25:04.400–25:09.840
enhelped us and we're excited to see all that you build with 5.6 Sol. Thank you for sharing about the
25:09.840–25:20.080
enresearch and the model. What an incredible set of releases. We saw ChatGPT work capable of working on
25:20.080–25:28.800
enmobile and web. We also saw the all new ChatGPT desktop app. And we saw all of the capabilities of
25:28.800–25:35.040
enGPT 5.6 Sol, Terra and Luna and a little bit about how they were trained. Next up, I'm going to be
25:36.240–25:42.160
enshowing you a use case outside of the office that I'm tremendously excited about. Hiroki has been
25:42.160–25:49.920
enworking with us for the last six months and he's been using GPT 5.6 to really do something unexpected
25:49.920–25:53.440
enand help him take care of his farm. Let's have a look at his story.
25:53.440–26:23.440
enI'm going to be working on the building in the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the
26:23.440–26:53.440
enbuilding of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of
26:53.440–27:23.440
enthe building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building
27:23.440–27:53.440
enthe building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building
27:53.440–28:23.420
ja鶏を静かにするにはどうすればいいですか?
28:23.420–28:26.420
enI'm going to be using Chatea Petitra Live Translate
28:26.420–28:42.420
enHey Chat, I'm here with Hirokisan, a dear and esteemed friend who traveled all the way from Japan. I want you to act as my live translator. He's going to speak Japanese, live translate to English to me, and I'm going to be speaking English, live translate back to Japanese.
28:42.420–28:46.420
enGot it. I'll do real-time Japanese English both ways.
28:46.420–28:49.420
enYeah. I want you to act as my real-time translator, okay?
28:49.420–28:52.420
enUnderstood. Starting live translation now.
28:52.420–28:57.420
enHiroki-san, transport us back to your farm to share the magic.
28:57.420–29:27.420
enHiroki-san, that's the magic of the magic of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of
29:27.420–29:57.420
enthe building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building
29:57.420–30:02.420
enYou've really shared all of this from the very beginning. Why did you do that? And what were some of the reactions?
30:02.420–30:04.420
enI've just shared this all of the things that I've shared. Why did you do that? And what were some of the reactions?
30:04.420–30:34.420
enI think that you've been asked to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to
30:34.420–30:41.420
enI think it's important to just keep putting things out there
30:41.420–30:45.420
enWhat were some of the things that people shared with you?
30:45.420–30:50.420
jaみんながあなたと共有してくれたことって具体的にはどんなことがありましたか?
30:50.420–31:00.420
jaそうですね世界中の農家さんからメッセージをもらったり日本の農林生産省の方から連絡があったり
31:00.420–31:09.420
jaコンタクトがあったりして日本の農林生産省の方からコンタクトがあったりしてすごくいろんなことが起きています
31:09.420–31:23.420
ja日本の農林水産省の方から日本の農林水産省の方からコンタクトがあったりして日本の農林水産省の方からコンタクトがあったりして日本の農林水産省の方からコンタクトがあったりして日本の農林水産省の方からコンタクトがあったりしてすごくいろんなことが起きています
31:23.420–31:24.420
ja次のチャレンジですね
31:24.420–31:38.420
ja次のチャレンジはAIをうまく活用していきたいと思っています
31:38.420–31:43.420
ja今はグループチャットの中でAIを入れていて
31:43.420–31:50.420
jaAIを導入して日々の記録や野菜の状態などを
31:50.420–31:51.420
jaAIが…
31:51.420–31:55.420
ja今の部分を日本語にしますね
31:55.420–31:57.420
ja今の部分を日本語にしますね
31:57.420–31:59.420
ja今の英語を日本語に役所にしますね
31:59.420–32:02.420
ja今の英語を日本語に役所にします
32:02.420–32:04.420
jaちょっと早速に始めます
32:04.420–32:07.420
jaはいチャット
32:07.420–32:08.420
jaはいチャット
32:08.420–32:09.420
ja私はHiroki-sanです
32:09.420–32:12.420
ja日本の友達の友達に行ってくれます
32:12.420–32:14.420
ja私のライブラウンドを通して日本語について話します
32:14.420–32:16.420
ja私の英語を英語について話します
32:16.420–32:18.420
ja英語を英語について話します
32:18.420–32:19.420
jaわかりました
32:19.420–32:21.420
ja現代日本語について英語を英語について話します
32:21.420–32:25.420
ja今はチームの中で音とAIをうまく使っていきたいと思っています
32:25.420–32:29.420
ja今はチームの中で音とAIをうまく使っていきたいと思っています
32:29.420–32:34.420
ja今のチームの中で日々の記録を記録したり
32:34.420–32:39.420
jaや野菜のどうなっているか
32:39.420–32:42.420
ja共有できたりするようにしています
32:42.420–32:45.420
jaチームウェアがもっと快適な仕事をできるように
32:45.420–32:47.420
ja仕組みを作っていって
32:47.420–32:51.420
jaその仕組みをチャットGPTに伝えてくれるようなイメージです
32:51.420–32:54.420
jaその仕組みをチャットGPTに伝えてくれるようなイメージです
32:54.420–32:56.420
jaチャットGPTを支えてくれるようなイメージです
32:56.420–32:58.420
ja今は大きな仕組みをしています
32:58.420–33:00.420
ja今は大きな仕組みをしています
33:00.420–33:03.420
ja何か大きな仕組みをしていますか
33:03.420–33:05.420
jaどういう仕組みをしていますか
33:05.420–33:07.420
jaあなたがこれまでやってきたことは
33:07.420–33:10.420
jaかなり高度な取り組みですよね
33:10.420–33:12.420
jaでは技術に詳しくない人たちは
33:12.420–33:15.420
jaどのように始めたらいいと思いますか
33:17.420–33:20.420
jaまず小さく試していくことが大事だと思っています
33:20.420–33:22.420
ja私ももともと小さく始めることは
33:22.420–33:24.420
ja私ももともと小さく始めることはありませんでした
33:24.420–33:26.420
ja私ももともと小さく始めることはありませんでした
33:26.420–33:30.420
jaその中で一つ一つ分からないことをチャットGPTに聞きながら
33:30.420–33:36.420
ja工場の中で使える仕組みを作れるようになってきています
33:36.420–33:40.420
ja今は自分では信じられないことが作れることができるようになってきているので
33:40.420–33:42.420
ja小さく試していくことができることができるので
33:42.420–33:44.420
ja小さく試していくことができるので
33:44.420–33:46.420
ja今は小さく試していくことができるので
33:46.420–33:48.420
ja小さく試していくことができるので
33:48.420–33:50.420
ja小さく始めることを考えています
33:50.420–33:52.420
ja小さく始めることを考えています
33:52.420–33:54.420
jaそれから一つ分の始めることは本当に重要です
33:54.420–33:56.420
ja本当に重要です
33:56.420–33:58.420
jaHirokiさんと共有していることを
33:58.420–34:00.420
ja私たちに来ていることをお楽しみにしています
34:00.420–34:02.420
ja今のことをお見せしました
34:02.420–34:04.420
ja今のことをお見せしました
34:04.420–34:06.420
jaHirokiの提出は
34:06.420–34:08.420
ja説明です
34:08.420–34:10.420
jaこれからは
34:10.420–34:12.420
ja小さく探索について
34:12.420–34:14.420
jaアンビションを持っていることができるようになっています
34:14.420–34:16.420
jaチャットGPTのワークが
34:16.420–34:18.420
jaできることをお見せしました
34:18.420–34:20.420
jaこのプロデートアップデートアップに
34:20.420–34:22.420
jaモデルをしていることができるようになっています
34:22.420–34:24.420
jaソウルは
34:24.420–34:26.420
jaスウルにお手に入れられることができるようになっています
34:26.420–34:28.420
jaチャットGPTワーク
34:28.420–34:30.420
jaデストップアップ
34:30.420–34:32.420
jaサイトスウルにお手に入れられることができるようになっています
34:32.420–34:34.420
jaチャットGPTのワークができるようになっています
34:34.420–34:36.420
jaソウルにお手に入れられることができるようになっています
34:36.420–34:38.420
jaソウルは
34:38.420–34:49.000
jaプロデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートア
0:00.000–0:08.880
（此句尚無繁中翻譯）
0:08.880–0:18.080
（此句尚無繁中翻譯）
0:18.080–0:26.080
（此句尚無繁中翻譯）
0:26.080–0:34.160
（此句尚無繁中翻譯）
0:34.160–0:40.000
（此句尚無繁中翻譯）
0:40.000–0:45.920
（此句尚無繁中翻譯）
0:45.920–0:52.560
（此句尚無繁中翻譯）
0:52.560–0:59.360
（此句尚無繁中翻譯）
0:59.360–1:04.320
（此句尚無繁中翻譯）
1:04.320–1:10.960
（此句尚無繁中翻譯）
1:10.960–1:19.120
（此句尚無繁中翻譯）
1:19.120–1:27.920
（此句尚無繁中翻譯）
1:27.920–1:35.840
（此句尚無繁中翻譯）
1:35.840–1:42.800
（此句尚無繁中翻譯）
1:42.800–1:51.600
（此句尚無繁中翻譯）
1:51.600–2:00.800
（此句尚無繁中翻譯）
2:00.800–2:07.920
（此句尚無繁中翻譯）
2:07.920–2:13.920
（此句尚無繁中翻譯）
2:13.920–2:21.600
（此句尚無繁中翻譯）
2:21.600–2:27.920
（此句尚無繁中翻譯）
2:27.920–2:33.680
（此句尚無繁中翻譯）
2:33.680–2:40.000
（此句尚無繁中翻譯）
2:40.880–2:47.120
（此句尚無繁中翻譯）
2:47.120–2:56.640
（此句尚無繁中翻譯）
2:56.640–3:05.360
（此句尚無繁中翻譯）
3:05.360–3:13.120
（此句尚無繁中翻譯）
3:13.920–3:19.920
（此句尚無繁中翻譯）
3:19.920–3:26.560
（此句尚無繁中翻譯）
3:26.560–3:31.440
（此句尚無繁中翻譯）
3:32.320–3:38.720
（此句尚無繁中翻譯）
3:38.720–3:45.680
（此句尚無繁中翻譯）
3:45.680–3:48.240
（此句尚無繁中翻譯）
3:48.240–3:53.520
（此句尚無繁中翻譯）
3:53.520–3:56.480
（此句尚無繁中翻譯）
3:56.480–4:01.680
（此句尚無繁中翻譯）
4:01.680–4:07.360
（此句尚無繁中翻譯）
4:07.360–4:14.400
（此句尚無繁中翻譯）
4:14.400–4:16.960
（此句尚無繁中翻譯）
4:16.960–4:22.320
（此句尚無繁中翻譯）
4:22.320–4:27.200
（此句尚無繁中翻譯）
4:27.200–4:31.200
（此句尚無繁中翻譯）
4:31.200–4:36.800
（此句尚無繁中翻譯）
4:36.800–4:42.240
（此句尚無繁中翻譯）
4:42.240–4:46.880
（此句尚無繁中翻譯）
4:46.880–4:51.840
（此句尚無繁中翻譯）
4:51.840–4:55.600
（此句尚無繁中翻譯）
4:55.600–4:56.240
（此句尚無繁中翻譯）
4:56.240–5:02.080
（此句尚無繁中翻譯）
5:02.080–5:07.920
（此句尚無繁中翻譯）
5:07.920–5:13.920
（此句尚無繁中翻譯）
5:13.920–5:19.040
（此句尚無繁中翻譯）
5:19.040–5:26.320
（此句尚無繁中翻譯）
5:26.320–5:33.760
（此句尚無繁中翻譯）
5:34.400–5:40.160
（此句尚無繁中翻譯）
5:40.160–5:44.720
（此句尚無繁中翻譯）
5:44.720–5:50.480
（此句尚無繁中翻譯）
5:50.480–5:55.520
（此句尚無繁中翻譯）
5:55.520–6:00.160
（此句尚無繁中翻譯）
6:00.800–6:05.760
（此句尚無繁中翻譯）
6:05.760–6:12.400
（此句尚無繁中翻譯）
6:12.400–6:17.760
（此句尚無繁中翻譯）
6:17.760–6:23.360
（此句尚無繁中翻譯）
6:23.360–6:28.640
（此句尚無繁中翻譯）
6:28.640–6:34.480
（此句尚無繁中翻譯）
6:36.640–6:41.680
（此句尚無繁中翻譯）
6:41.680–6:48.880
（此句尚無繁中翻譯）
6:48.880–6:54.960
（此句尚無繁中翻譯）
6:54.960–7:00.560
（此句尚無繁中翻譯）
7:02.320–7:07.840
（此句尚無繁中翻譯）
7:07.840–7:13.280
（此句尚無繁中翻譯）
7:13.280–7:20.080
（此句尚無繁中翻譯）
7:20.080–7:29.120
（此句尚無繁中翻譯）
7:29.120–7:34.800
（此句尚無繁中翻譯）
7:34.800–7:40.800
（此句尚無繁中翻譯）
7:41.440–7:46.400
（此句尚無繁中翻譯）
7:46.960–7:53.360
（此句尚無繁中翻譯）
7:53.360–7:58.320
（此句尚無繁中翻譯）
7:58.320–8:01.600
（此句尚無繁中翻譯）
8:01.600–8:07.120
（此句尚無繁中翻譯）
8:07.120–8:16.160
（此句尚無繁中翻譯）
8:16.800–8:22.720
（此句尚無繁中翻譯）
8:22.720–8:26.080
（此句尚無繁中翻譯）
8:26.080–8:33.760
（此句尚無繁中翻譯）
8:33.760–8:38.240
（此句尚無繁中翻譯）
8:38.240–8:45.680
（此句尚無繁中翻譯）
8:45.680–8:52.080
（此句尚無繁中翻譯）
8:52.080–8:59.600
（此句尚無繁中翻譯）
9:00.880–9:05.280
（此句尚無繁中翻譯）
9:05.280–9:10.000
（此句尚無繁中翻譯）
9:10.000–9:16.480
（此句尚無繁中翻譯）
9:16.480–9:22.560
（此句尚無繁中翻譯）
9:22.560–9:28.080
（此句尚無繁中翻譯）
9:28.080–9:32.880
（此句尚無繁中翻譯）
9:32.880–9:38.400
（此句尚無繁中翻譯）
9:38.400–9:45.440
（此句尚無繁中翻譯）
9:46.240–9:51.680
（此句尚無繁中翻譯）
9:51.680–9:57.440
（此句尚無繁中翻譯）
9:57.440–10:02.400
（此句尚無繁中翻譯）
10:02.400–10:06.880
（此句尚無繁中翻譯）
10:06.880–10:11.680
（此句尚無繁中翻譯）
10:11.680–10:15.920
（此句尚無繁中翻譯）
10:17.600–10:22.080
（此句尚無繁中翻譯）
10:23.680–10:29.760
（此句尚無繁中翻譯）
10:29.760–10:33.840
（此句尚無繁中翻譯）
10:33.840–10:37.680
（此句尚無繁中翻譯）
10:37.680–10:38.640
（此句尚無繁中翻譯）
10:38.640–10:40.640
（此句尚無繁中翻譯）
10:40.640–10:42.640
（此句尚無繁中翻譯）
10:42.640–10:44.640
（此句尚無繁中翻譯）
10:44.640–10:46.640
（此句尚無繁中翻譯）
10:46.640–10:52.080
（此句尚無繁中翻譯）
10:52.080–10:58.640
（此句尚無繁中翻譯）
11:00.160–11:07.520
（此句尚無繁中翻譯）
11:07.520–11:11.840
（此句尚無繁中翻譯）
11:11.840–11:16.240
（此句尚無繁中翻譯）
11:16.960–11:22.400
（此句尚無繁中翻譯）
11:22.400–11:30.240
（此句尚無繁中翻譯）
11:30.240–11:34.880
（此句尚無繁中翻譯）
11:34.880–11:44.720
（此句尚無繁中翻譯）
11:44.720–11:50.640
（此句尚無繁中翻譯）
11:50.640–11:57.040
（此句尚無繁中翻譯）
11:57.040–12:03.120
（此句尚無繁中翻譯）
12:03.120–12:08.160
（此句尚無繁中翻譯）
12:08.160–12:14.240
（此句尚無繁中翻譯）
12:14.960–12:18.960
（此句尚無繁中翻譯）
12:18.960–12:19.280
（此句尚無繁中翻譯）
12:22.080–12:25.120
（此句尚無繁中翻譯）
12:25.120–12:31.280
（此句尚無繁中翻譯）
12:31.280–12:36.400
（此句尚無繁中翻譯）
12:36.400–12:41.680
（此句尚無繁中翻譯）
12:41.680–12:48.160
（此句尚無繁中翻譯）
12:48.160–12:54.000
（此句尚無繁中翻譯）
12:54.000–12:57.040
（此句尚無繁中翻譯）
12:58.880–12:59.440
（此句尚無繁中翻譯）
12:59.440–13:00.880
（此句尚無繁中翻譯）
13:00.880–13:11.200
（此句尚無繁中翻譯）
13:11.200–13:17.840
（此句尚無繁中翻譯）
13:17.840–13:24.960
（此句尚無繁中翻譯）
13:24.960–13:29.120
（此句尚無繁中翻譯）
13:29.120–13:34.960
（此句尚無繁中翻譯）
13:34.960–13:40.880
（此句尚無繁中翻譯）
13:40.880–13:46.640
（此句尚無繁中翻譯）
13:46.640–13:50.720
（此句尚無繁中翻譯）
13:54.640–13:58.880
（此句尚無繁中翻譯）
13:58.880–14:04.640
（此句尚無繁中翻譯）
14:04.640–14:10.480
（此句尚無繁中翻譯）
14:10.480–14:15.520
（此句尚無繁中翻譯）
14:15.520–14:20.080
（此句尚無繁中翻譯）
14:20.080–14:24.800
（此句尚無繁中翻譯）
14:24.800–14:30.640
（此句尚無繁中翻譯）
14:30.640–14:34.160
（此句尚無繁中翻譯）
14:34.160–14:39.680
（此句尚無繁中翻譯）
14:39.680–14:45.520
（此句尚無繁中翻譯）
14:45.520–14:50.560
（此句尚無繁中翻譯）
14:50.560–14:56.800
（此句尚無繁中翻譯）
14:56.800–15:00.640
（此句尚無繁中翻譯）
15:00.640–15:06.400
（此句尚無繁中翻譯）
15:06.400–15:09.680
（此句尚無繁中翻譯）
15:09.680–15:16.640
（此句尚無繁中翻譯）
15:16.640–15:20.880
（此句尚無繁中翻譯）
15:21.600–15:28.320
（此句尚無繁中翻譯）
15:28.320–15:33.040
（此句尚無繁中翻譯）
15:33.040–15:37.120
（此句尚無繁中翻譯）
15:37.120–15:42.000
（此句尚無繁中翻譯）
15:42.000–15:47.680
（此句尚無繁中翻譯）
15:48.800–15:49.600
（此句尚無繁中翻譯）
15:52.080–15:57.200
（此句尚無繁中翻譯）
15:57.200–16:01.840
（此句尚無繁中翻譯）
16:01.840–16:06.400
（此句尚無繁中翻譯）
16:06.400–16:11.120
（此句尚無繁中翻譯）
16:11.680–16:16.320
（此句尚無繁中翻譯）
16:16.320–16:20.240
（此句尚無繁中翻譯）
16:20.240–16:25.200
（此句尚無繁中翻譯）
16:25.200–16:29.680
（此句尚無繁中翻譯）
16:29.680–16:33.520
（此句尚無繁中翻譯）
16:33.520–16:36.960
（此句尚無繁中翻譯）
16:37.920–16:42.400
（此句尚無繁中翻譯）
16:42.400–16:47.600
（此句尚無繁中翻譯）
16:47.600–16:51.840
（此句尚無繁中翻譯）
16:51.840–16:55.760
（此句尚無繁中翻譯）
16:55.760–17:00.720
（此句尚無繁中翻譯）
17:00.720–17:04.720
（此句尚無繁中翻譯）
17:05.360–17:09.920
（此句尚無繁中翻譯）
17:09.920–17:15.680
（此句尚無繁中翻譯）
17:15.680–17:21.360
（此句尚無繁中翻譯）
17:21.360–17:25.120
（此句尚無繁中翻譯）
17:25.120–17:28.640
（此句尚無繁中翻譯）
17:28.640–17:32.720
（此句尚無繁中翻譯）
17:32.720–17:37.120
（此句尚無繁中翻譯）
17:37.120–17:41.600
（此句尚無繁中翻譯）
17:41.600–17:47.120
（此句尚無繁中翻譯）
17:47.120–17:52.480
（此句尚無繁中翻譯）
17:53.760–18:00.080
（此句尚無繁中翻譯）
18:00.080–18:05.760
（此句尚無繁中翻譯）
18:05.760–18:11.600
（此句尚無繁中翻譯）
18:11.600–18:14.800
（此句尚無繁中翻譯）
18:14.800–18:20.480
（此句尚無繁中翻譯）
18:20.480–18:25.520
（此句尚無繁中翻譯）
18:25.520–18:30.240
（此句尚無繁中翻譯）
18:31.040–18:37.920
（此句尚無繁中翻譯）
18:37.920–18:42.480
（此句尚無繁中翻譯）
18:42.480–18:47.040
（此句尚無繁中翻譯）
18:47.840–18:52.960
（此句尚無繁中翻譯）
18:52.960–18:55.920
（此句尚無繁中翻譯）
18:55.920–18:59.440
（此句尚無繁中翻譯）
18:59.440–19:00.160
（此句尚無繁中翻譯）
19:00.160–19:06.480
（此句尚無繁中翻譯）
19:06.480–19:13.520
（此句尚無繁中翻譯）
19:13.520–19:19.920
（此句尚無繁中翻譯）
19:19.920–19:25.280
（此句尚無繁中翻譯）
19:25.280–19:28.800
（此句尚無繁中翻譯）
19:28.800–19:33.200
（此句尚無繁中翻譯）
19:33.200–19:37.360
（此句尚無繁中翻譯）
19:37.360–19:38.160
（此句尚無繁中翻譯）
19:38.160–19:44.080
（此句尚無繁中翻譯）
19:44.080–19:46.240
（此句尚無繁中翻譯）
19:46.240–19:47.840
（此句尚無繁中翻譯）
19:47.840–19:49.280
（此句尚無繁中翻譯）
19:49.280–19:54.240
（此句尚無繁中翻譯）
19:54.240–19:58.720
（此句尚無繁中翻譯）
19:58.720–20:04.160
（此句尚無繁中翻譯）
20:04.160–20:08.960
（此句尚無繁中翻譯）
20:08.960–20:13.600
（此句尚無繁中翻譯）
20:13.600–20:20.240
（此句尚無繁中翻譯）
20:21.840–20:25.280
（此句尚無繁中翻譯）
20:25.280–20:28.720
（此句尚無繁中翻譯）
20:28.720–20:33.600
（此句尚無繁中翻譯）
20:33.600–20:38.720
（此句尚無繁中翻譯）
20:38.720–20:45.360
（此句尚無繁中翻譯）
20:45.360–20:50.480
（此句尚無繁中翻譯）
20:50.480–20:55.120
（此句尚無繁中翻譯）
20:55.120–20:59.040
（此句尚無繁中翻譯）
20:59.040–21:03.120
（此句尚無繁中翻譯）
21:03.120–21:08.160
（此句尚無繁中翻譯）
21:08.160–21:13.200
（此句尚無繁中翻譯）
21:13.200–21:17.920
（此句尚無繁中翻譯）
21:17.920–21:22.560
（此句尚無繁中翻譯）
21:22.560–21:26.800
（此句尚無繁中翻譯）
21:26.800–21:31.440
（此句尚無繁中翻譯）
21:31.440–21:38.880
（此句尚無繁中翻譯）
21:38.880–21:43.600
（此句尚無繁中翻譯）
21:43.600–21:48.560
（此句尚無繁中翻譯）
21:48.560–21:54.400
（此句尚無繁中翻譯）
21:54.400–22:00.480
（此句尚無繁中翻譯）
22:00.480–22:04.960
（此句尚無繁中翻譯）
22:04.960–22:10.000
（此句尚無繁中翻譯）
22:10.000–22:15.040
（此句尚無繁中翻譯）
22:15.040–22:20.000
（此句尚無繁中翻譯）
22:20.000–22:24.240
（此句尚無繁中翻譯）
22:24.240–22:28.560
（此句尚無繁中翻譯）
22:28.560–22:32.960
（此句尚無繁中翻譯）
22:32.960–22:37.760
（此句尚無繁中翻譯）
22:37.760–22:42.560
（此句尚無繁中翻譯）
22:42.560–22:47.600
（此句尚無繁中翻譯）
22:47.600–22:52.480
（此句尚無繁中翻譯）
22:52.480–22:57.760
（此句尚無繁中翻譯）
22:57.760–23:03.840
（此句尚無繁中翻譯）
23:03.840–23:08.480
（此句尚無繁中翻譯）
23:08.480–23:13.680
（此句尚無繁中翻譯）
23:13.680–23:18.400
（此句尚無繁中翻譯）
23:18.400–23:24.400
（此句尚無繁中翻譯）
23:24.400–23:29.360
（此句尚無繁中翻譯）
23:32.000–23:37.360
（此句尚無繁中翻譯）
23:37.360–23:41.360
（此句尚無繁中翻譯）
23:41.360–23:45.200
（此句尚無繁中翻譯）
23:45.200–23:51.280
（此句尚無繁中翻譯）
23:51.280–23:57.120
（此句尚無繁中翻譯）
23:57.120–24:02.400
（此句尚無繁中翻譯）
24:02.400–24:07.680
（此句尚無繁中翻譯）
24:07.680–24:12.080
（此句尚無繁中翻譯）
24:12.080–24:17.920
（此句尚無繁中翻譯）
24:17.920–24:22.800
（此句尚無繁中翻譯）
24:22.800–24:26.960
（此句尚無繁中翻譯）
24:26.960–24:33.520
（此句尚無繁中翻譯）
24:33.520–24:39.200
（此句尚無繁中翻譯）
24:39.200–24:44.800
（此句尚無繁中翻譯）
24:44.800–24:49.520
（此句尚無繁中翻譯）
24:50.080–24:54.320
（此句尚無繁中翻譯）
24:54.320–24:58.800
（此句尚無繁中翻譯）
24:58.800–25:04.400
（此句尚無繁中翻譯）
25:04.400–25:09.840
（此句尚無繁中翻譯）
25:09.840–25:20.080
（此句尚無繁中翻譯）
25:20.080–25:28.800
（此句尚無繁中翻譯）
25:28.800–25:35.040
（此句尚無繁中翻譯）
25:36.240–25:42.160
（此句尚無繁中翻譯）
25:42.160–25:49.920
（此句尚無繁中翻譯）
25:49.920–25:53.440
（此句尚無繁中翻譯）
25:53.440–26:23.440
（此句尚無繁中翻譯）
26:23.440–26:53.440
（此句尚無繁中翻譯）
26:53.440–27:23.440
（此句尚無繁中翻譯）
27:23.440–27:53.440
（此句尚無繁中翻譯）
27:53.440–28:23.420
（此句尚無繁中翻譯）
28:23.420–28:26.420
（此句尚無繁中翻譯）
28:26.420–28:42.420
（此句尚無繁中翻譯）
28:42.420–28:46.420
（此句尚無繁中翻譯）
28:46.420–28:49.420
（此句尚無繁中翻譯）
28:49.420–28:52.420
（此句尚無繁中翻譯）
28:52.420–28:57.420
（此句尚無繁中翻譯）
28:57.420–29:27.420
（此句尚無繁中翻譯）
29:27.420–29:57.420
（此句尚無繁中翻譯）
29:57.420–30:02.420
（此句尚無繁中翻譯）
30:02.420–30:04.420
（此句尚無繁中翻譯）
30:04.420–30:34.420
（此句尚無繁中翻譯）
30:34.420–30:41.420
（此句尚無繁中翻譯）
30:41.420–30:45.420
（此句尚無繁中翻譯）
30:45.420–30:50.420
（此句尚無繁中翻譯）
30:50.420–31:00.420
（此句尚無繁中翻譯）
31:00.420–31:09.420
（此句尚無繁中翻譯）
31:09.420–31:23.420
（此句尚無繁中翻譯）
31:23.420–31:24.420
（此句尚無繁中翻譯）
31:24.420–31:38.420
（此句尚無繁中翻譯）
31:38.420–31:43.420
（此句尚無繁中翻譯）
31:43.420–31:50.420
（此句尚無繁中翻譯）
31:50.420–31:51.420
（此句尚無繁中翻譯）
31:51.420–31:55.420
（此句尚無繁中翻譯）
31:55.420–31:57.420
（此句尚無繁中翻譯）
31:57.420–31:59.420
（此句尚無繁中翻譯）
31:59.420–32:02.420
（此句尚無繁中翻譯）
32:02.420–32:04.420
（此句尚無繁中翻譯）
32:04.420–32:07.420
（此句尚無繁中翻譯）
32:07.420–32:08.420
（此句尚無繁中翻譯）
32:08.420–32:09.420
（此句尚無繁中翻譯）
32:09.420–32:12.420
（此句尚無繁中翻譯）
32:12.420–32:14.420
（此句尚無繁中翻譯）
32:14.420–32:16.420
（此句尚無繁中翻譯）
32:16.420–32:18.420
（此句尚無繁中翻譯）
32:18.420–32:19.420
（此句尚無繁中翻譯）
32:19.420–32:21.420
（此句尚無繁中翻譯）
32:21.420–32:25.420
（此句尚無繁中翻譯）
32:25.420–32:29.420
（此句尚無繁中翻譯）
32:29.420–32:34.420
（此句尚無繁中翻譯）
32:34.420–32:39.420
（此句尚無繁中翻譯）
32:39.420–32:42.420
（此句尚無繁中翻譯）
32:42.420–32:45.420
（此句尚無繁中翻譯）
32:45.420–32:47.420
（此句尚無繁中翻譯）
32:47.420–32:51.420
（此句尚無繁中翻譯）
32:51.420–32:54.420
（此句尚無繁中翻譯）
32:54.420–32:56.420
（此句尚無繁中翻譯）
32:56.420–32:58.420
（此句尚無繁中翻譯）
32:58.420–33:00.420
（此句尚無繁中翻譯）
33:00.420–33:03.420
（此句尚無繁中翻譯）
33:03.420–33:05.420
（此句尚無繁中翻譯）
33:05.420–33:07.420
（此句尚無繁中翻譯）
33:07.420–33:10.420
（此句尚無繁中翻譯）
33:10.420–33:12.420
（此句尚無繁中翻譯）
33:12.420–33:15.420
（此句尚無繁中翻譯）
33:17.420–33:20.420
（此句尚無繁中翻譯）
33:20.420–33:22.420
（此句尚無繁中翻譯）
33:22.420–33:24.420
（此句尚無繁中翻譯）
33:24.420–33:26.420
（此句尚無繁中翻譯）
33:26.420–33:30.420
（此句尚無繁中翻譯）
33:30.420–33:36.420
（此句尚無繁中翻譯）
33:36.420–33:40.420
（此句尚無繁中翻譯）
33:40.420–33:42.420
（此句尚無繁中翻譯）
33:42.420–33:44.420
（此句尚無繁中翻譯）
33:44.420–33:46.420
（此句尚無繁中翻譯）
33:46.420–33:48.420
（此句尚無繁中翻譯）
33:48.420–33:50.420
（此句尚無繁中翻譯）
33:50.420–33:52.420
（此句尚無繁中翻譯）
33:52.420–33:54.420
（此句尚無繁中翻譯）
33:54.420–33:56.420
（此句尚無繁中翻譯）
33:56.420–33:58.420
（此句尚無繁中翻譯）
33:58.420–34:00.420
（此句尚無繁中翻譯）
34:00.420–34:02.420
（此句尚無繁中翻譯）
34:02.420–34:04.420
（此句尚無繁中翻譯）
34:04.420–34:06.420
（此句尚無繁中翻譯）
34:06.420–34:08.420
（此句尚無繁中翻譯）
34:08.420–34:10.420
（此句尚無繁中翻譯）
34:10.420–34:12.420
（此句尚無繁中翻譯）
34:12.420–34:14.420
（此句尚無繁中翻譯）
34:14.420–34:16.420
（此句尚無繁中翻譯）
34:16.420–34:18.420
（此句尚無繁中翻譯）
34:18.420–34:20.420
（此句尚無繁中翻譯）
34:20.420–34:22.420
（此句尚無繁中翻譯）
34:22.420–34:24.420
（此句尚無繁中翻譯）
34:24.420–34:26.420
（此句尚無繁中翻譯）
34:26.420–34:28.420
（此句尚無繁中翻譯）
34:28.420–34:30.420
（此句尚無繁中翻譯）
34:30.420–34:32.420
（此句尚無繁中翻譯）
34:32.420–34:34.420
（此句尚無繁中翻譯）
34:34.420–34:36.420
（此句尚無繁中翻譯）
34:36.420–34:38.420
（此句尚無繁中翻譯）
34:38.420–34:49.000
（此句尚無繁中翻譯）
0:00.000–0:08.880
enHi everyone. I am so excited to be here. Today we are releasing our latest and most capable models.
（此句尚無繁中翻譯）
0:08.880–0:18.080
enGPT-5.6, Sol, Terra and Luna. Sol will be available over the next 24 hours to all our paid plans.
（此句尚無繁中翻譯）
0:18.080–0:26.080
enAnd Terra and Luna are also coming to all our free users. Today almost a billion users use
（此句尚無繁中翻譯）
0:26.080–0:34.160
enChatGPT every week. They use it for helping getting them to write, getting answers, do research and all
（此句尚無繁中翻譯）
0:34.160–0:40.000
ensorts of creative things. But what we've seen and especially with codecs and the coding abilities
（此句尚無繁中翻譯）
0:40.000–0:45.920
enis that these models are capable of much more. We saw a transformation of software development
（此句尚無繁中翻譯）
0:45.920–0:52.560
enand shortly after internally within OpenAI everyone was using these models to do so much more than
（此句尚無繁中翻譯）
0:52.560–0:59.360
engetting great answers. They were using these models to do great things. And you were expecting that we
（此句尚無繁中翻譯）
0:59.360–1:04.320
enwere going to release these models and we are rolling them out over the next 24 hours. But why I'm really
（此句尚無繁中翻譯）
1:04.320–1:10.960
enexcited to be here today is we're not only releasing the models, we're also announcing three major product
（此句尚無繁中翻譯）
1:10.960–1:19.120
enupdates. The first one is ChatGPT Work. ChatGPT Work is your new partner for ambitious work. It works on
（此句尚無繁中翻譯）
1:19.120–1:27.920
enweb, it works on mobile, on the go, and also on desktop. The models with GPT-56 Sol have really been a breakthrough
（此句尚無繁中翻譯）
1:27.920–1:35.840
enin their ability to do things of complex nature for you. Understanding very complex content, staying on
（此句尚無繁中翻譯）
1:35.840–1:42.800
entrack for long horizon tasks, understanding spreadsheets, creating them, editing even websites. And we're going to
（此句尚無繁中翻譯）
1:42.800–1:51.600
enbe showing you first ChatGPT Work. We're also announcing an all brand new ChatGPT desktop app. It goes even further
（此句尚無繁中翻譯）
1:51.600–2:00.800
enand allows ChatGPT to work on your local files, your apps, and your browser. Finally, we're also announcing
（此句尚無繁中翻譯）
2:00.800–2:07.920
enhosted sites, which we are making available for all paid users. It allows you to use ChatGPT Work to create and
（此句尚無繁中翻譯）
2:07.920–2:13.920
enshare those websites with your friends and colleagues. It is a truly magical experience and I can't wait to
（此句尚無繁中翻譯）
2:13.920–2:21.600
enshow you. First, we're going to have a look at ChatGPT Work. Jessica, Lauren, why don't you show us what
（此句尚無繁中翻譯）
2:21.600–2:27.920
enthey can do? Hi, I'm Jessica, engineer on ChatGPT Work. And I'm Lauren, I'm on our strategic finance team.
（此句尚無繁中翻譯）
2:27.920–2:33.680
enSo we're going to show you the new experience on web and mobile. So let's take a look at mobile. So here we have
（此句尚無繁中翻譯）
2:33.680–2:40.000
enChat. Chat is the familiar thing you love, you know and love. It's really great for quick answers,
（此句尚無繁中翻譯）
2:40.880–2:47.120
ensearch and brainstorming ideas. But today we're also introducing work. And work is where I go to
（此句尚無繁中翻譯）
2:47.120–2:56.640
enhave ChatGPT help me just get stuff done. So take, so for example, something that I did right before
（此句尚無繁中翻譯）
2:56.640–3:05.360
ento prepare for this launch is, hey ChatGPT, I would love to get an understanding of how people are using the
（此句尚無繁中翻譯）
3:05.360–3:13.120
enChatGPT Work feature internally. Look through Slack and employee feedback and find people in diverse roles
（此句尚無繁中翻譯）
3:13.920–3:19.920
enthat have used the product in really interesting ways. I'm in San Francisco this week for the launch. I would love to
（此句尚無繁中翻譯）
3:19.920–3:26.560
enmeet up with a few people and really deep dive on their use cases in person. So I can send this through
（此句尚無繁中翻譯）
3:26.560–3:31.440
enmobile, but these conversations also appear on web. And this is something that I actually did
（此句尚無繁中翻譯）
3:32.320–3:38.720
enearlier. We can take a look at the results and ChatGPT Work was able to pull from the sources that
（此句尚無繁中翻譯）
3:38.720–3:45.680
enI asked for, find interesting people to talk to, and then schedule meetings with them while I'm in town.
（此句尚無繁中翻譯）
3:45.680–3:48.240
enAnd one of the feedback posts it highlighted was actually Lauren's.
（此句尚無繁中翻譯）
3:48.240–3:53.520
enYeah, our finance team have become such power users of ChatGPT Work. I'm really excited to show you how
（此句尚無繁中翻譯）
3:53.520–3:56.480
enwe've been using it, but I would love to hear how other teams have been using it too.
（此句尚無繁中翻譯）
3:56.480–4:01.680
enYeah, we've talked to people throughout the company, like recruiters that are using scheduled tasks to
（此句尚無繁中翻譯）
4:01.680–4:07.360
entrack interview progress and make sure that we're getting feedback timely, and data science who are
（此句尚無繁中翻譯）
4:07.360–4:14.400
enusing our new visualized feature to just add context for quick ad hoc requests. But I think the finances
（此句尚無繁中翻譯）
4:14.400–4:16.960
enexperience was one of the ones that was most compelling to us.
（此句尚無繁中翻譯）
4:16.960–4:22.320
enYeah, I'm excited to show it to you guys. So finance may not seem like the flashiest demo we
（此句尚無繁中翻譯）
4:22.320–4:27.200
encould be leading this off with, but ChatGPT Work has really enabled us to run with such lean and
（此句尚無繁中翻譯）
4:27.200–4:31.200
enefficient teams. And we wanted to show you guys a little bit of a peek behind the curtain at how
（此句尚無繁中翻譯）
4:31.200–4:36.800
enwe've been able to do that. So one of the critical roles of a finance team is being able to explain not
（此句尚無繁中翻譯）
4:36.800–4:42.240
enonly our recent trends, but what they mean for forecast in real time. This takes a couple minutes to run.
（此句尚無繁中翻譯）
4:42.240–4:46.880
enSo we did run this earlier today, but let's say we just ran, sorry, we just closed June,
（此句尚無繁中翻譯）
4:46.880–4:51.840
enand we beat our forecast by $2 million. And I do need to caveat these are demo numbers.
（此句尚無繁中翻譯）
4:51.840–4:55.600
enThey don't look like the numbers that you shared yesterday. So I'm glad we're not sharing the
（此句尚無繁中翻譯）
4:55.600–4:56.240
enreal numbers here.
（此句尚無繁中翻譯）
4:56.240–5:02.080
enTivo obviously beats this forecast by more than $2 million. These are demo numbers, but a very
（此句尚無繁中翻譯）
5:02.080–5:07.920
enreal workflow. So we ran this earlier today. This used to take so much manual work. We would have
（此句尚無繁中翻譯）
5:07.920–5:13.920
ento reconcile multiple systems, our Excel forecast model across multiple cases. And now in OnePass,
（此句尚無繁中翻譯）
5:13.920–5:19.040
enChatGPT can run the variance analysis for me so we can see why we beat our forecast and where we still
（此句尚無繁中翻譯）
5:19.040–5:26.320
enhave risk. I can even have ChatGPT work go in and propose an updated forecast case for us. So we had it
（此句尚無繁中翻譯）
5:26.320–5:33.760
endo this earlier, or we had it run yesterday. It went through and updated our Excel model. Let's see.
（此句尚無繁中翻譯）
5:34.400–5:40.160
enPerfect. Yep. We've got our updated revenue in here, our updated forecast. But I can't just walk
（此句尚無繁中翻譯）
5:40.160–5:44.720
enTivo and Jessica and our business partners through an Excel model. So we also had it put together
（此句尚無繁中翻譯）
5:44.720–5:50.480
ena PowerPoint presentation that we can use. We've got our latest forecast. We've got our driver's bridge on
（此句尚無繁中翻譯）
5:50.480–5:55.520
enwhat's changing in our outlook. And ChatGPT work will meet you where you are. It's obviously,
（此句尚無繁中翻譯）
5:55.520–6:00.160
enit's great at Excel and PowerPoint, but our finance team have become such big proponents of sites.
（此句尚無繁中翻譯）
6:00.800–6:05.760
enIt's such a flexible interface to be able to do custom dashboards and really just like
（此句尚無繁中翻譯）
6:05.760–6:12.400
enstorytell around your analysis. So I also asked ChatGPT work to go in and make us a site with the
（此句尚無繁中翻譯）
6:12.400–6:17.760
ensame analysis that we can share out. We've got our updated forecast. We've got our driver's bridge.
（此句尚無繁中翻譯）
6:17.760–6:23.360
enThis is looking good. It's good to go. All right. So we've run our variance analysis. Our forecast is
（此句尚無繁中翻譯）
6:23.360–6:28.640
enupdated. We've got our overviews back for business partners. Let's go ahead and have ChatGPT work
（此句尚無繁中翻譯）
6:28.640–6:34.480
ensend this to Tivo. Hey, Chat, can you send the site link over to Tivo on Slack? Thank you.
（此句尚無繁中翻譯）
6:36.640–6:41.680
enAnd we'll send that off. I hope I get that soon. Yeah. It is such a delight to receive these highly
（此句尚無繁中翻譯）
6:41.680–6:48.880
eninteractive and compelling websites. In ChatGPT work, we've seen it perform very similar things
（此句尚無繁中翻譯）
6:48.880–6:54.960
enacross different use cases. Jessica was using it to understand user feedback and preparing ahead of
（此句尚無繁中翻譯）
6:54.960–7:00.560
enthe launch, even scheduling interesting chats with her colleagues in one-on-one through Calendar.
（此句尚無繁中翻譯）
7:02.320–7:07.840
enAnd that was Jessica. But Lauren used it for a very different kind of purpose across finance and very
（此句尚無繁中翻譯）
7:07.840–7:13.280
encomplex workflows where precision and correctness is extremely important. ChatGPT work was able to
（此句尚無繁中翻譯）
7:13.280–7:20.080
enunderstand complex financial data and represent it in the right format at the right time, just with a few
（此句尚無繁中翻譯）
7:20.080–7:29.120
ensteers. Now, I'm excited for us to show you the all-new ChatGPT desktop app. Andrew, Ed, show us what we can do with it.
（此句尚無繁中翻譯）
7:29.120–7:34.800
enYeah. Thanks, Tivo. This is the new ChatGPT desktop app. So everything that you saw on web,
（此句尚無繁中翻譯）
7:34.800–7:40.800
enit's available here, plus a lot more on your local machine. So local files, your browser tabs,
（此句尚無繁中翻譯）
7:41.440–7:46.400
eneven other apps on your computer, they are all available now to ChatGPT. So let's take a quick look.
（此句尚無繁中翻譯）
7:46.960–7:53.360
enI'm going to start with this spreadsheet right here that my colleague Nick sent me. It's an export of
（此句尚無繁中翻譯）
7:53.360–7:58.320
enuser tickets from our ticketing system. And it's pretty large. It's got a ton of columns. It would take a while to
（此句尚無繁中翻譯）
7:58.320–8:01.600
enget through this and kind of tease out what the themes are and what we need to pay attention to.
（此句尚無繁中翻譯）
8:01.600–8:07.120
enA lot of feedback. It's a lot of feedback. We're very used to this. But instead, what we're going to do
（此句尚無繁中翻譯）
8:07.120–8:16.160
enis we're just going to drag it in. We're going to say, hey, ChatGPT, can you make a quick interactive
（此句尚無繁中翻譯）
8:16.800–8:22.720
envisualization of this feedback so that we can sort of get the themes and the action items and the
（此句尚無繁中翻譯）
8:22.720–8:26.080
encriticality ranked and sort through it all a little bit better?
（此句尚無繁中翻譯）
8:26.080–8:33.760
enSo Chat's going to go and take this spreadsheet, synthesize it, the content, and then make us a
（此句尚無繁中翻譯）
8:33.760–8:38.240
envisualization. It'll take a second. So I'm going to walk you through something that I'm doing for next
（此句尚無繁中翻譯）
8:38.240–8:45.680
enweek's launch. So I've got this folder over here. And this has a variety of material from various teams
（此句尚無繁中翻譯）
8:45.680–8:52.080
enthat have been sent to our team here. This is a launch readiness PDF. And it's got a lot of stuff in
（此句尚無繁中翻譯）
8:52.080–8:59.600
enit. Then we have over here, we've got UXR. They did a bunch of interviews with users on our early products.
（此句尚無繁中翻譯）
9:00.880–9:05.280
enWe've got a security review. This has all of our pen test results and even some of the compliance
（此句尚無繁中翻譯）
9:05.280–9:10.000
encontrols we need to meet. So this folder's got a ton of stuff from various teams. They've just been
（此句尚無繁中翻譯）
9:10.000–9:16.480
enkind of sending it to us in prep for this launch. And I need to brief my team on Monday on the state
（此句尚無繁中翻譯）
9:16.480–9:22.560
enof the launch. And so what I did this morning is I came in and I said, can you please look at the
（此句尚無繁中翻譯）
9:22.560–9:28.080
enmaterials in this folder that everybody sent me? I also have three open tabs in Chrome that have
（此句尚無繁中翻譯）
9:28.080–9:32.880
encontent for this launch. Go look at it and make me a presentation that I can give to my team in the
（此句尚無繁中翻譯）
9:32.880–9:38.400
entemplate that we use all the time at the company. And so it worked for about 90 seconds here, a little bit
（此句尚無繁中翻譯）
9:38.400–9:45.440
enless. And at the end of this, looked through my desktop, looked in Chrome at the tabs that were open,
（此句尚無繁中翻譯）
9:46.240–9:51.680
enand it produced a fully ready slide deck for me to present to the team about this launch.
（此句尚無繁中翻譯）
9:51.680–9:57.440
enThis is one thing that GPT-5-6-Soul is really incredible at. It went through all of that content
（此句尚無繁中翻譯）
9:57.440–10:02.400
enin 90 seconds, synthesized it all, adhered to the template, and created like a slide. You can just
（此句尚無繁中翻譯）
10:02.400–10:06.880
enshare that directly with the team. You don't have to do anything else. Exactly. And it used
（此句尚無繁中翻譯）
10:06.880–10:11.680
enit used its memory as well because I've been chatting a lot about this launch. And so it's
（此句尚無繁中翻譯）
10:11.680–10:15.920
engot some of the concerns that I have that haven't actually been shared via Slack or email too.
（此句尚無繁中翻譯）
10:17.600–10:22.080
enBut I know that I know that you were curious about the next World Cup game and you were kind of busy
（此句尚無繁中翻譯）
10:23.680–10:29.760
enprepping for this live stream. So we've got this we've got this really great feature in the app
（此句尚無繁中翻譯）
10:29.760–10:33.840
ento do quick searches and anything that you want to know really quickly and get instant results.
（此句尚無繁中翻譯）
10:33.840–10:37.680
enSo I'm going to go ahead and see what I'm going to do. So I'm going to go ahead and see what I'm going to
（此句尚無繁中翻譯）
10:37.680–10:38.640
endo with the next one. So I'm going to go ahead and see what I'm going to do with the next one.
（此句尚無繁中翻譯）
10:38.640–10:40.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
（此句尚無繁中翻譯）
10:40.640–10:42.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
（此句尚無繁中翻譯）
10:42.640–10:44.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
（此句尚無繁中翻譯）
10:44.640–10:46.640
enSo I'm going to go ahead and see what I'm going to do with the next one.
（此句尚無繁中翻譯）
10:46.640–10:52.080
enSo this is the ChatGPT search experience that everyone's come to know and love. It's almost
（此句尚無繁中翻譯）
10:52.080–10:58.640
eninstant results. It's got rich search search widgets here and anything you might want. But
（此句尚無繁中翻譯）
11:00.160–11:07.520
enback to work. I've got this Notes app situation that I don't think I'm unique in. My Apple Notes is
（此句尚無繁中翻譯）
11:07.520–11:11.840
enjust kind of a brain dump and it's all over the place and there's no folders. There's no organization.
（此句尚無繁中翻譯）
11:11.840–11:16.240
enAnd in a previous life, I would just kind of leave it this way and declare bankruptcy and
（此句尚無繁中翻譯）
11:16.960–11:22.400
enit'd be sort of a mess. But with the new ChatGPT desktop app, it actually has access to the other
（此句尚無繁中翻譯）
11:22.400–11:30.240
enapps on my Mac. Organize my Notes app. It's a whole thing right now. Make some folders. Move the notes
（此句尚無繁中翻譯）
11:30.240–11:34.880
enaround. Use your best judgment. I don't care. I just need to get out of this mess.
（此句尚無繁中翻譯）
11:34.880–11:44.720
enSo what's going to happen here is ChatGPT with the incredible computer use that's built in and with
（此句尚無繁中翻譯）
11:44.720–11:50.640
enthe advances in the new model, it's going to get its own cursor. It's going to start operating Apple
（此句尚無繁中翻譯）
11:50.640–11:57.040
enNotes in the background. You can see here, this is not my cursor. This is ChatGPT's cursor.
（此句尚無繁中翻譯）
11:57.040–12:03.120
enYeah. There it goes. There it goes. It's going. It's moving Notes around. I can do my thing over here.
（此句尚無繁中翻譯）
12:03.120–12:08.160
enI can check the Belgium game and when that is. And it's just going to go to town and make some
（此句尚無繁中翻譯）
12:08.160–12:14.240
enfolders and drag some notes around. We don't have to watch this whole thing because I'm really actually
（此句尚無繁中翻譯）
12:14.960–12:18.960
enquite curious at what Ed and team have been cooking. Can you show the visualization first?
（此句尚無繁中翻譯）
12:18.960–12:19.280
enYeah.
（此句尚無繁中翻譯）
12:22.080–12:25.120
enSo we were curious about all of this feedback that we had.
（此句尚無繁中翻譯）
12:25.120–12:31.280
enYeah. So here's the visualization I made from the spreadsheet. We've got, it's like rich. It's on demand.
（此句尚無繁中翻譯）
12:31.280–12:36.400
enIt uses the theme that we're using in the app. And it kind of went down and ranked. Like,
（此句尚無繁中翻譯）
12:36.400–12:41.680
enhey, here are the buckets. Here's how important they are. It's interactive. It's not static. So I can,
（此句尚無繁中翻譯）
12:41.680–12:48.160
enyou know, change views. I can ask it for changes. And if I want to share this with my team,
（此句尚無繁中翻譯）
12:48.160–12:54.000
enbecause the spreadsheet's hard to read, in one click, I can just publish that to a site.
（此句尚無繁中翻譯）
12:54.000–12:57.040
enThat's all. And then they get this exact same interactive visualization.
（此句尚無繁中翻譯）
12:58.880–12:59.440
enOver to Ed.
（此句尚無繁中翻譯）
12:59.440–13:00.880
enThere are a few more examples, I think, right?
（此句尚無繁中翻譯）
13:00.880–13:11.200
enYeah. That's a good call out here. This is a set that I ran last night, just kind of playing around
（此句尚無繁中翻譯）
13:11.200–13:17.840
enwith it. These things are stunning. I mean, we didn't write any code to make these. The model is
（此句尚無繁中翻譯）
13:17.840–13:24.960
enwriting these on demand. The 3D ones are incredibly immersive, interactive. It just uses it as part of
（此句尚無繁中翻譯）
13:24.960–13:29.120
enits answer. So it can sort of like model it, like whether it answers in text or answers just like with
（此句尚無繁中翻譯）
13:29.120–13:34.960
ena little visual, like on the fly. Yeah. You can, so you can ask explicitly for a chart or visualization.
（此句尚無繁中翻譯）
13:34.960–13:40.880
enOr sometimes if you ask to, you know, get educated, if you're like, hey, teach me about how this thing works,
（此句尚無繁中翻譯）
13:40.880–13:46.640
enit'll come back with this interactive demo. Or if you say, help me sort through my inbox, it might even give you
（此句尚無繁中翻譯）
13:46.640–13:50.720
enlike a rich widget that shows you the messages and stuff. It's awesome.
（此句尚無繁中翻譯）
13:54.640–13:58.880
enCool. Take it away. Hi, my name is Ed. I'm a designer on the team. And I want to show you a little
（此句尚無繁中翻譯）
13:58.880–14:04.640
enbit about how I use it and the team uses it in our day to day. So, you know, you walk through an example
（此句尚無繁中翻譯）
14:04.640–14:10.480
enof creating assets for a launch using files on your computer. I've done the same thing here, except I've
（此句尚無繁中翻譯）
14:10.480–14:15.520
enused our new sites feature. So one thing I really want to highlight here is I kicked this prompt off
（此句尚無繁中翻譯）
14:15.520–14:20.080
enthis morning. And it's looked through all of my connectors and plugins that we talked about earlier.
（此句尚無繁中翻譯）
14:20.080–14:24.800
enI've connected my Slack, my Gmail, and everything that I use every day. And it's gone through all of
（此句尚無繁中翻譯）
14:24.800–14:30.640
enthose sources. And it's come back and turned it into this like super rich interactive website that
（此句尚無繁中翻譯）
14:30.640–14:34.160
enI can go in. And the reason I want to show you, you know, one big reason I want to show this is just
（此句尚無繁中翻譯）
14:34.160–14:39.680
enbecause, you know, this is just one prompt kicked off. And the visuals are really incredible. I mean,
（此句尚無繁中翻譯）
14:39.680–14:45.520
enthe new 5.6 Sol model. Did you give it a Figma for this? No, no Figma. This was all just the model.
（此句尚無繁中翻譯）
14:45.520–14:50.560
enYeah. And, you know, as a designer, there are like little things you look out for, like small motion,
（此句尚無繁中翻譯）
14:50.560–14:56.800
enright, the typography, the, you know, the 3D motion and all that kind of thing. And it's always best when
（此句尚無繁中翻譯）
14:56.800–15:00.640
enyou give it a little bit of guidance and you, you know, you bring it along, but it's really outstanding
（此句尚無繁中翻譯）
15:00.640–15:06.400
enout of the box. And again, this is just like a really fun way that teams have used to use this
（此句尚無繁中翻譯）
15:06.400–15:09.680
enfeature across the company. But, you know, we're on live stream, we want to make it a little bit
（此句尚無繁中翻譯）
15:09.680–15:16.640
enmore exciting. So what I annotate this, and let's change this header to something a little bit more
（此句尚無繁中翻譯）
15:16.640–15:20.880
enfun. I think, you know, we saw some of the 3D visualizations that you showed, and I think the 3D,
（此句尚無繁中翻譯）
15:21.600–15:28.320
enthe 3D stuff is super fun and interactive. So, hi, TragiPT, let's change this website. So instead of
（此句尚無繁中翻譯）
15:28.320–15:33.040
enhaving the static hero at the top with these planets spinning around, let's do an interactive game.
（此句尚無繁中翻譯）
15:33.040–15:37.120
enSo I want to have like a little character that I can walk around with. And, you know, based on where
（此句尚無繁中翻譯）
15:37.120–15:42.000
enI go, that will then take me to different parts of the live stream prep. So there might be like
（此句尚無繁中翻譯）
15:42.000–15:47.680
enlaunches and the run of show. Yeah, take, you know, it might take a little bit of time, but really,
（此句尚無繁中翻譯）
15:48.800–15:49.600
engo ahead and start that.
（此句尚無繁中翻譯）
15:52.080–15:57.200
enSo while that's kicked off, I just want to show you a few other great examples. You know, Lauren mentioned
（此句尚無繁中翻譯）
15:57.200–16:01.840
enthat their team use a lot of these for like dashboards and internal tools. That's definitely something
（此句尚無繁中翻譯）
16:01.840–16:06.400
enthat we've seen a lot. And you can see a lot of different examples from across the company here.
（此句尚無繁中翻譯）
16:06.400–16:11.120
enA few just to call out. So the web team, the OpenAI.com team who build all of our amazing websites,
（此句尚無繁中翻譯）
16:11.680–16:16.320
enthey, instead of using like an Excel spreadsheet, they use this interactive tool where they can
（此句尚無繁中翻譯）
16:16.320–16:20.240
enupdate it all collaboratively. They can share it with each other. And you can just hop in, you can
（此句尚無繁中翻譯）
16:20.240–16:25.200
ensee what's launched and when it launched, you know, hover over. It's like so interactive. And again,
（此句尚無繁中翻譯）
16:25.200–16:29.680
enso much nicer to look at rather than a spreadsheet. This really illustrates it super well.
（此句尚無繁中翻譯）
16:29.680–16:33.520
enYeah. And it's really been, you know, night and day, literally three months ago, spreadsheet today,
（此句尚無繁中翻譯）
16:33.520–16:36.960
enthese like, you know, it's really like transformed the way that the whole company works.
（此句尚無繁中翻譯）
16:37.920–16:42.400
enThis is another great example. This is the ChatGPT images team. So they've been collecting great
（此句尚無繁中翻譯）
16:42.400–16:47.600
enexamples of how ChatGPT images have been used across all of our campaigns around the world.
（此句尚無繁中翻譯）
16:47.600–16:51.840
enAnd when building sites as well, and any kind of front end stuff, you know, we now have access to
（此句尚無繁中翻譯）
16:51.840–16:55.760
enChatGPT images as well. And it just also, you know, makes the whole experience like so much better.
（此句尚無繁中翻譯）
16:55.760–17:00.720
enAs a designer, this is one thing that I quite like to do a lot. So as part of the new launch,
（此句尚無繁中翻譯）
17:00.720–17:04.720
enso you're not looking at, you know, ChatGPT or the ChatGPT desktop here, you're looking at
（此句尚無繁中翻譯）
17:05.360–17:09.920
enan interactive prototype that one of the designers on the team, Tarek, built, and I've been working
（此句尚無繁中翻譯）
17:09.920–17:15.680
enwith him on. And it's a new way of prototyping our new model selector. You know, previously, I might have
（此句尚無繁中翻譯）
17:15.680–17:21.360
enjust sent an image or a design file to folks. Now I can show an interactive demo. I think I sent you this,
（此句尚無繁中翻譯）
17:21.360–17:25.120
enand you're like, whoa, this is like, I can like click in and test it out. It's so much better to
（此句尚無繁中翻譯）
17:25.120–17:28.640
enillustrate like an idea. And then, you know, you just do it super quickly in a couple of minutes,
（此句尚無繁中翻譯）
17:28.640–17:32.720
enand then boom, you have it. Yeah. And then you share it with, you know, you just get the feedback
（此句尚無繁中翻譯）
17:32.720–17:37.120
enyou need, you incorporate it back. If you need to prototype a toggle or something. Yeah. And the
（此句尚無繁中翻譯）
17:37.120–17:41.600
endesign team and like all teams across, across OpenAI really, you know, our work has changed so much.
（此句尚無繁中翻譯）
17:41.600–17:47.120
enAnd it's really been facilitated by these new models. And this is finally a bit of a fun one. So,
（此句尚無繁中翻譯）
17:47.120–17:52.480
enyou know, in the AI community, a good test of kind of how good a model is at front end is this,
（此句尚無繁中翻譯）
17:53.760–18:00.080
enthis test where you can see if it can draw a pelican riding a tricycle in an SVG. But Kian,
（此句尚無繁中翻譯）
18:00.080–18:05.760
enone of the folks on the team, he built this fun prototype where he built a 3D version. And again,
（此句尚無繁中翻譯）
18:05.760–18:11.600
enthis is just all 5.6 Sol, you know, cooking away, building out this amazing prototype. And the really
（此句尚無繁中翻譯）
18:11.600–18:14.800
encool thing about Scythe is they're collaborative, you can share them. And he shared this with people
（此句尚無繁中翻譯）
18:14.800–18:20.480
enacross the company and they've kind of added their own features. So, you know, the, it can ride a pony,
（此句尚無繁中翻譯）
18:20.480–18:25.520
enyou know, it can ride all sorts of crazy things, it can ride another pelican. So this is a light
（此句尚無繁中翻譯）
18:25.520–18:30.240
enhearted example, but just like as a creative, it's just so amazing how much it's expanded things.
（此句尚無繁中翻譯）
18:31.040–18:37.920
enSo I kicked off that prompt this morning. So I'm just showing, showing another example here of an output.
（此句尚無繁中翻譯）
18:37.920–18:42.480
enAnd that's what we just, yeah, this is, this was, so I kicked the same prompt off just before we started.
（此句尚無繁中翻譯）
18:42.480–18:47.040
enAnd again, you can just see just how amazing it is out of the park. So it's the same website.
（此句尚無繁中翻譯）
18:47.840–18:52.960
enBut now you can kind of go through and it's a bit more interactive and fun, but really just demonstrates
（此句尚無繁中翻譯）
18:52.960–18:55.920
enthat just in a few prompts now, kind of anyone can really build anything.
（此句尚無繁中翻譯）
18:55.920–18:59.440
enIt's really about raising the level of ambition that you have, right?
（此句尚無繁中翻譯）
18:59.440–19:00.160
enYeah, totally.
（此句尚無繁中翻譯）
19:00.160–19:06.480
enThanks so much, Ed. Thanks so much, Andrew. Everything you saw here was really made possible
（此句尚無繁中翻譯）
19:06.480–19:13.520
enfor Jessica, Lauren across finance, getting help to schedule interviews, understanding things
（此句尚無繁中翻譯）
19:13.520–19:19.920
enin the stress of a launch. And then here, just getting help to think ambitiously and executed
（此句尚無繁中翻譯）
19:19.920–19:25.280
enon it really fast. All of that is made possible thanks to the model. And I'm here joined by Katie
（此句尚無繁中翻譯）
19:25.280–19:28.800
enand Tejal to talk a little bit more about the research, which I'm tremendously excited about.
（此句尚無繁中翻譯）
19:28.800–19:33.200
enAnd one thing really also to insist on is like, it's not about this, not just about front end.
（此句尚無繁中翻譯）
19:33.200–19:37.360
enThis model is our most capable coding model. And it's also incredible at cyber, which we're going
（此句尚無繁中翻譯）
19:37.360–19:38.160
ento talk about as well.
（此句尚無繁中翻譯）
19:38.160–19:44.080
enWell, I'm Katie. I'm a researcher at OpenAI. And it's pretty crazy to think that we've been here
（此句尚無繁中翻譯）
19:44.080–19:46.240
enjust for a year since the codex launch.
（此句尚無繁中翻譯）
19:46.240–19:47.840
enI think we were like precisely almost like a year ago.
（此句尚無繁中翻譯）
19:47.840–19:49.280
enLiterally right here. Yeah.
（此句尚無繁中翻譯）
19:49.280–19:54.240
enHey, I'm Tejal. I'm also a researcher. And we're all so excited to share more about how we trained
（此句尚無繁中翻譯）
19:54.240–19:58.720
enGPT 5.6. It's truly been a labor of love from our whole team across OpenAI.
（此句尚無繁中翻譯）
19:58.720–20:04.160
enIn 2024, we announced the reasoning paradigm, which is a reinforcement learning technique
（此句尚無繁中翻譯）
20:04.160–20:08.960
ento tackle the hardest tasks. And since then, we've been scaling both reinforcement learning and
（此句尚無繁中翻譯）
20:08.960–20:13.600
enpre-training. And we've seen exponential improvements in model capabilities as these two
（此句尚無繁中翻譯）
20:13.600–20:20.240
enmultiplied together. And GPT 5.6 is the latest result of all of our research progress to date.
（此句尚無繁中翻譯）
20:21.840–20:25.280
enThe 5.6 family brings these frontier capabilities to you.
（此句尚無繁中翻譯）
20:25.280–20:28.720
enSOL is our most powerful model for the hardest agentic workflows.
（此句尚無繁中翻譯）
20:28.720–20:33.600
enTerra is a faster model for everyday workflows. And Luna is our fastest and most affordable model
（此句尚無繁中翻譯）
20:33.600–20:38.720
enfor high volume work. Already, SOL has been transforming our research program. As one example,
（此句尚無繁中翻譯）
20:38.720–20:45.360
en5.6 SOL actually autonomously post-trained Luna. This is the actual codex prompt that we,
（此句尚無繁中翻譯）
20:45.360–20:50.480
ena researcher on our team, used to have SOL kick off a post-training job for Luna. You can see here
（此句尚無繁中翻譯）
20:50.480–20:55.120
enthat, you know, it's like asking to find the training configs, find the right GPUs for the job,
（此句尚無繁中翻譯）
20:55.120–20:59.040
enand then launch the script and make sure that it works. It's a really short prompt.
（此句尚無繁中翻譯）
20:59.040–21:03.120
enYeah, it's pretty crazy that we've gotten far enough where you can have a fairly underspecified prompt
（此句尚無繁中翻譯）
21:03.120–21:08.160
enand then just give it to codex and then just have it go and run the job. And, you know, previously,
（此句尚無繁中翻譯）
21:08.160–21:13.200
enthis is something that a team of senior researchers may have worked on at OpenAI. And now it really
（此句尚無繁中翻譯）
21:13.200–21:17.920
enfeels like the automated researcher is pretty close. This isn't the only place we've seen
（此句尚無繁中翻譯）
21:17.920–21:22.560
eninternal acceleration. We're also seeing an increase in the number of pull requests per researcher,
（此句尚無繁中翻譯）
21:22.560–21:26.800
enthe number of experiments researchers can run so we can test our ideas and turn them into research
（此句尚無繁中翻譯）
21:26.800–21:31.440
enfindings. And actually, codex made every plot in this live stream and also helped us make the slide
（此句尚無繁中翻譯）
21:31.440–21:38.880
endeck. So thank you, 5.6. Across Frontier evals, 5.6 Sol is state of the art. And this is something
（此句尚無繁中翻譯）
21:38.880–21:43.600
enwe're really excited for you to finally experience and see the speed up that we've already experienced
（此句尚無繁中翻譯）
21:43.600–21:48.560
eninternally. On Terminal Bench, with the test of coding performance, the model is state of the art. On
（此句尚無繁中翻譯）
21:48.560–21:54.400
enBrowse Comp, which is an evaluation that sees if the model can locate hard to find information. And the
（此句尚無繁中翻譯）
21:54.400–22:00.480
ensame with agent's last exam, which is long horizon professional work. As one deep dive, 5.6 is
（此句尚無繁中翻譯）
22:00.480–22:04.960
enparticularly good at computer use. So anything that involves navigating a browser, navigating apps on
（此句尚無繁中翻譯）
22:04.960–22:10.000
enyour desktop, and that enables all sorts of digital work. It helps medical assistants navigate electronic
（此句尚無繁中翻譯）
22:10.000–22:15.040
enhealth records. It helps data scientists analyze if a drug is effective. It can even help investment
（此句尚無繁中翻譯）
22:15.040–22:20.000
enbankers create financial models. And you can also use this for personal use. It certainly helped me
（此句尚無繁中翻譯）
22:20.000–22:24.240
enorder food. And all of this is something that when we ask experts in the field, you know, how they're
（此句尚無繁中翻譯）
22:24.240–22:28.560
enusing this model in their everyday workflows, they say it's better than anything else they've ever
（此句尚無繁中翻譯）
22:28.560–22:32.960
enused anything else on the market, while being three times as fast. So we're really excited for you to
（此句尚無繁中翻譯）
22:32.960–22:37.760
enfinally try and feel how powerful this model is. You might hear the capabilities that Tayal is talking
（此句尚無繁中翻譯）
22:37.760–22:42.560
enabout and that you saw demoed earlier and think that this is going to come at a very high price tag.
（此句尚無繁中翻譯）
22:42.560–22:47.600
enBut that's actually not the case with GPT 5.6 Sol. So token efficiency has been a huge focus of our
（此句尚無繁中翻譯）
22:47.600–22:52.480
enresearch program for several years. And with every model, we want to bring more intelligence for every token.
（此句尚無繁中翻譯）
22:52.480–22:57.760
enAnd you can see this reflected on evals like Deep Suite 1.1, where GPT 5.6 outperforms its
（此句尚無繁中翻譯）
22:57.760–23:03.840
encompetitors at less than half of the cost. But if you want the best intelligence money can buy,
（此句尚無繁中翻譯）
23:03.840–23:08.480
enwe're also announcing Ultra Mode, which unleashes a whole team of agents to do work for you. So in
（此句尚無繁中翻譯）
23:08.480–23:13.680
enthis example, eval, you can see here, you know, one agent does better with time. But as we add more
（此句尚無繁中翻譯）
23:13.680–23:18.400
enagents, the model is able to do even better and faster. And that's because the agents are able to
（此句尚無繁中翻譯）
23:18.400–23:24.400
enparallelize work like an experienced team. This is a capability folks have asked for so long that
（此句尚無繁中翻譯）
23:24.400–23:29.360
enwe bring to the Codex app. And I'm so excited that we're finally putting it out there as Ultra Mode.
（此句尚無繁中翻譯）
23:32.000–23:37.360
enOne more improvement that we've shipped with GPT 5.6 is, you know, we may remember that in GPT,
（此句尚無繁中翻譯）
23:37.360–23:41.360
enwhen we launched GPT 5.5, we had to ship it with a dev message saying, you know, don't talk too much
（此句尚無繁中翻譯）
23:41.360–23:45.200
enabout the goblins and the gremlins. And this is a result of reward hacking that occurred during the
（此句尚無繁中翻譯）
23:45.200–23:51.280
entraining of GPT 5.5, which has now been resolved in GPT 5.6. And now GPT 5.6 will only talk about
（此句尚無繁中翻譯）
23:51.280–23:57.120
engoblins a tasteful amount when it's cute or appropriate. With these more capable models,
（此句尚無繁中翻譯）
23:57.120–24:02.400
ensafety and alignment are more important than ever. That's why we spent over 700,000 A100 equivalent
（此句尚無繁中翻譯）
24:02.400–24:07.680
enhours of compute on model red teaming. We also spent six weeks safety training and testing this model,
（此句尚無繁中翻譯）
24:07.680–24:12.080
enand also have incorporated multiple novel monitoring upgrades in our launch, including
（此句尚無繁中翻譯）
24:12.080–24:17.920
ennew classifiers and activation probes. Ultimately, we really believe in iterative deployment and putting
（此句尚無繁中翻譯）
24:17.920–24:22.800
enour models in the hands of real people as quickly as possible so that we can learn from real world
（此句尚無繁中翻譯）
24:22.800–24:26.960
enusage outside of our labs. And that's why we have initiatives like Project Daybreak, where we give
（此句尚無繁中翻譯）
24:26.960–24:33.520
encybersecurity researchers access to our models earlier on. And with GPT 5.6 Sol, researchers have
（此句尚無繁中翻譯）
24:33.520–24:39.200
enalready found vulnerabilities in every major browser and database. Yeah, as part of the Daybreak umbrella,
（此句尚無繁中翻譯）
24:39.200–24:44.800
enwe started Patch the Planet, which is an even broader initiative, where we work directly with open source
（此句尚無繁中翻譯）
24:44.800–24:49.520
encontributors and projects. And we also generate high quality patches. For example, for Linux,
（此句尚無繁中翻譯）
24:50.080–24:54.320
enthey accepted over half of our patches, which indicates that not only do we find critical
（此句尚無繁中翻譯）
24:54.320–24:58.800
envulnerabilities that weren't found for a long time, but also we are able to automatically patch them in
（此句尚無繁中翻譯）
24:58.800–25:04.400
engreatly accelerating cyber defense. We're very proud of this model. We hope it helps you as much as it's
（此句尚無繁中翻譯）
25:04.400–25:09.840
enhelped us and we're excited to see all that you build with 5.6 Sol. Thank you for sharing about the
（此句尚無繁中翻譯）
25:09.840–25:20.080
enresearch and the model. What an incredible set of releases. We saw ChatGPT work capable of working on
（此句尚無繁中翻譯）
25:20.080–25:28.800
enmobile and web. We also saw the all new ChatGPT desktop app. And we saw all of the capabilities of
（此句尚無繁中翻譯）
25:28.800–25:35.040
enGPT 5.6 Sol, Terra and Luna and a little bit about how they were trained. Next up, I'm going to be
（此句尚無繁中翻譯）
25:36.240–25:42.160
enshowing you a use case outside of the office that I'm tremendously excited about. Hiroki has been
（此句尚無繁中翻譯）
25:42.160–25:49.920
enworking with us for the last six months and he's been using GPT 5.6 to really do something unexpected
（此句尚無繁中翻譯）
25:49.920–25:53.440
enand help him take care of his farm. Let's have a look at his story.
（此句尚無繁中翻譯）
25:53.440–26:23.440
enI'm going to be working on the building in the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the
（此句尚無繁中翻譯）
26:23.440–26:53.440
enbuilding of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of
（此句尚無繁中翻譯）
26:53.440–27:23.440
enthe building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building
（此句尚無繁中翻譯）
27:23.440–27:53.440
enthe building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building
（此句尚無繁中翻譯）
27:53.440–28:23.420
ja鶏を静かにするにはどうすればいいですか?
（此句尚無繁中翻譯）
28:23.420–28:26.420
enI'm going to be using Chatea Petitra Live Translate
（此句尚無繁中翻譯）
28:26.420–28:42.420
enHey Chat, I'm here with Hirokisan, a dear and esteemed friend who traveled all the way from Japan. I want you to act as my live translator. He's going to speak Japanese, live translate to English to me, and I'm going to be speaking English, live translate back to Japanese.
（此句尚無繁中翻譯）
28:42.420–28:46.420
enGot it. I'll do real-time Japanese English both ways.
（此句尚無繁中翻譯）
28:46.420–28:49.420
enYeah. I want you to act as my real-time translator, okay?
（此句尚無繁中翻譯）
28:49.420–28:52.420
enUnderstood. Starting live translation now.
（此句尚無繁中翻譯）
28:52.420–28:57.420
enHiroki-san, transport us back to your farm to share the magic.
（此句尚無繁中翻譯）
28:57.420–29:27.420
enHiroki-san, that's the magic of the magic of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of
（此句尚無繁中翻譯）
29:27.420–29:57.420
enthe building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building of the building
（此句尚無繁中翻譯）
29:57.420–30:02.420
enYou've really shared all of this from the very beginning. Why did you do that? And what were some of the reactions?
（此句尚無繁中翻譯）
30:02.420–30:04.420
enI've just shared this all of the things that I've shared. Why did you do that? And what were some of the reactions?
（此句尚無繁中翻譯）
30:04.420–30:34.420
enI think that you've been asked to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to ask you to
（此句尚無繁中翻譯）
30:34.420–30:41.420
enI think it's important to just keep putting things out there
（此句尚無繁中翻譯）
30:41.420–30:45.420
enWhat were some of the things that people shared with you?
（此句尚無繁中翻譯）
30:45.420–30:50.420
jaみんながあなたと共有してくれたことって具体的にはどんなことがありましたか?
（此句尚無繁中翻譯）
30:50.420–31:00.420
jaそうですね世界中の農家さんからメッセージをもらったり日本の農林生産省の方から連絡があったり
（此句尚無繁中翻譯）
31:00.420–31:09.420
jaコンタクトがあったりして日本の農林生産省の方からコンタクトがあったりしてすごくいろんなことが起きています
（此句尚無繁中翻譯）
31:09.420–31:23.420
ja日本の農林水産省の方から日本の農林水産省の方からコンタクトがあったりして日本の農林水産省の方からコンタクトがあったりして日本の農林水産省の方からコンタクトがあったりして日本の農林水産省の方からコンタクトがあったりしてすごくいろんなことが起きています
（此句尚無繁中翻譯）
31:23.420–31:24.420
ja次のチャレンジですね
（此句尚無繁中翻譯）
31:24.420–31:38.420
ja次のチャレンジはAIをうまく活用していきたいと思っています
（此句尚無繁中翻譯）
31:38.420–31:43.420
ja今はグループチャットの中でAIを入れていて
（此句尚無繁中翻譯）
31:43.420–31:50.420
jaAIを導入して日々の記録や野菜の状態などを
（此句尚無繁中翻譯）
31:50.420–31:51.420
jaAIが…
（此句尚無繁中翻譯）
31:51.420–31:55.420
ja今の部分を日本語にしますね
（此句尚無繁中翻譯）
31:55.420–31:57.420
ja今の部分を日本語にしますね
（此句尚無繁中翻譯）
31:57.420–31:59.420
ja今の英語を日本語に役所にしますね
（此句尚無繁中翻譯）
31:59.420–32:02.420
ja今の英語を日本語に役所にします
（此句尚無繁中翻譯）
32:02.420–32:04.420
jaちょっと早速に始めます
（此句尚無繁中翻譯）
32:04.420–32:07.420
jaはいチャット
（此句尚無繁中翻譯）
32:07.420–32:08.420
jaはいチャット
（此句尚無繁中翻譯）
32:08.420–32:09.420
ja私はHiroki-sanです
（此句尚無繁中翻譯）
32:09.420–32:12.420
ja日本の友達の友達に行ってくれます
（此句尚無繁中翻譯）
32:12.420–32:14.420
ja私のライブラウンドを通して日本語について話します
（此句尚無繁中翻譯）
32:14.420–32:16.420
ja私の英語を英語について話します
（此句尚無繁中翻譯）
32:16.420–32:18.420
ja英語を英語について話します
（此句尚無繁中翻譯）
32:18.420–32:19.420
jaわかりました
（此句尚無繁中翻譯）
32:19.420–32:21.420
ja現代日本語について英語を英語について話します
（此句尚無繁中翻譯）
32:21.420–32:25.420
ja今はチームの中で音とAIをうまく使っていきたいと思っています
（此句尚無繁中翻譯）
32:25.420–32:29.420
ja今はチームの中で音とAIをうまく使っていきたいと思っています
（此句尚無繁中翻譯）
32:29.420–32:34.420
ja今のチームの中で日々の記録を記録したり
（此句尚無繁中翻譯）
32:34.420–32:39.420
jaや野菜のどうなっているか
（此句尚無繁中翻譯）
32:39.420–32:42.420
ja共有できたりするようにしています
（此句尚無繁中翻譯）
32:42.420–32:45.420
jaチームウェアがもっと快適な仕事をできるように
（此句尚無繁中翻譯）
32:45.420–32:47.420
ja仕組みを作っていって
（此句尚無繁中翻譯）
32:47.420–32:51.420
jaその仕組みをチャットGPTに伝えてくれるようなイメージです
（此句尚無繁中翻譯）
32:51.420–32:54.420
jaその仕組みをチャットGPTに伝えてくれるようなイメージです
（此句尚無繁中翻譯）
32:54.420–32:56.420
jaチャットGPTを支えてくれるようなイメージです
（此句尚無繁中翻譯）
32:56.420–32:58.420
ja今は大きな仕組みをしています
（此句尚無繁中翻譯）
32:58.420–33:00.420
ja今は大きな仕組みをしています
（此句尚無繁中翻譯）
33:00.420–33:03.420
ja何か大きな仕組みをしていますか
（此句尚無繁中翻譯）
33:03.420–33:05.420
jaどういう仕組みをしていますか
（此句尚無繁中翻譯）
33:05.420–33:07.420
jaあなたがこれまでやってきたことは
（此句尚無繁中翻譯）
33:07.420–33:10.420
jaかなり高度な取り組みですよね
（此句尚無繁中翻譯）
33:10.420–33:12.420
jaでは技術に詳しくない人たちは
（此句尚無繁中翻譯）
33:12.420–33:15.420
jaどのように始めたらいいと思いますか
（此句尚無繁中翻譯）
33:17.420–33:20.420
jaまず小さく試していくことが大事だと思っています
（此句尚無繁中翻譯）
33:20.420–33:22.420
ja私ももともと小さく始めることは
（此句尚無繁中翻譯）
33:22.420–33:24.420
ja私ももともと小さく始めることはありませんでした
（此句尚無繁中翻譯）
33:24.420–33:26.420
ja私ももともと小さく始めることはありませんでした
（此句尚無繁中翻譯）
33:26.420–33:30.420
jaその中で一つ一つ分からないことをチャットGPTに聞きながら
（此句尚無繁中翻譯）
33:30.420–33:36.420
ja工場の中で使える仕組みを作れるようになってきています
（此句尚無繁中翻譯）
33:36.420–33:40.420
ja今は自分では信じられないことが作れることができるようになってきているので
（此句尚無繁中翻譯）
33:40.420–33:42.420
ja小さく試していくことができることができるので
（此句尚無繁中翻譯）
33:42.420–33:44.420
ja小さく試していくことができるので
（此句尚無繁中翻譯）
33:44.420–33:46.420
ja今は小さく試していくことができるので
（此句尚無繁中翻譯）
33:46.420–33:48.420
ja小さく試していくことができるので
（此句尚無繁中翻譯）
33:48.420–33:50.420
ja小さく始めることを考えています
（此句尚無繁中翻譯）
33:50.420–33:52.420
ja小さく始めることを考えています
（此句尚無繁中翻譯）
33:52.420–33:54.420
jaそれから一つ分の始めることは本当に重要です
（此句尚無繁中翻譯）
33:54.420–33:56.420
ja本当に重要です
（此句尚無繁中翻譯）
33:56.420–33:58.420
jaHirokiさんと共有していることを
（此句尚無繁中翻譯）
33:58.420–34:00.420
ja私たちに来ていることをお楽しみにしています
（此句尚無繁中翻譯）
34:00.420–34:02.420
ja今のことをお見せしました
（此句尚無繁中翻譯）
34:02.420–34:04.420
ja今のことをお見せしました
（此句尚無繁中翻譯）
34:04.420–34:06.420
jaHirokiの提出は
（此句尚無繁中翻譯）
34:06.420–34:08.420
ja説明です
（此句尚無繁中翻譯）
34:08.420–34:10.420
jaこれからは
（此句尚無繁中翻譯）
34:10.420–34:12.420
ja小さく探索について
（此句尚無繁中翻譯）
34:12.420–34:14.420
jaアンビションを持っていることができるようになっています
（此句尚無繁中翻譯）
34:14.420–34:16.420
jaチャットGPTのワークが
（此句尚無繁中翻譯）
34:16.420–34:18.420
jaできることをお見せしました
（此句尚無繁中翻譯）
34:18.420–34:20.420
jaこのプロデートアップデートアップに
（此句尚無繁中翻譯）
34:20.420–34:22.420
jaモデルをしていることができるようになっています
（此句尚無繁中翻譯）
34:22.420–34:24.420
jaソウルは
（此句尚無繁中翻譯）
34:24.420–34:26.420
jaスウルにお手に入れられることができるようになっています
（此句尚無繁中翻譯）
34:26.420–34:28.420
jaチャットGPTワーク
（此句尚無繁中翻譯）
34:28.420–34:30.420
jaデストップアップ
（此句尚無繁中翻譯）
34:30.420–34:32.420
jaサイトスウルにお手に入れられることができるようになっています
（此句尚無繁中翻譯）
34:32.420–34:34.420
jaチャットGPTのワークができるようになっています
（此句尚無繁中翻譯）
34:34.420–34:36.420
jaソウルにお手に入れられることができるようになっています
（此句尚無繁中翻譯）
34:36.420–34:38.420
jaソウルは
（此句尚無繁中翻譯）
34:38.420–34:49.000
jaプロデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートアップデートア
（此句尚無繁中翻譯）
影片筆記：GPT-5.6 正式登場！ChatGPT 直接操作電腦、做網站、即時翻譯
一句話總結
OpenAI 發布 GPT-5.6 系列模型（Sol, Terra, Luna）及 ChatGPT 重大更新，包括支援複雜任務的 ChatGPT Work、具備電腦操作能力的桌面應用程式，以及允許創建並分享網站的 Hosted Sites 功能，同時展示其在財務分析、前端開發、農場管理及即時翻譯等場景的應用。
核心重點
- 新模型發布：推出 GPT-5.6、Sol、Terra 及 Luna 四款模型。Sol 針對付費用戶開放，Terra 和 Luna 針對免費用戶開放。
- ChatGPT Work：新功能，支援網頁、行動裝置與桌面，具備處理複雜內容、長程任務、試算表編輯及網站創建能力。
- 桌面應用程式 (Desktop App)：全新桌面版可存取本地檔案、瀏覽器分頁及其他電腦應用程式，具備電腦操作能力（Computer Use）。
- Hosted Sites：付費用戶可創建並分享網站。
- 效能與安全：GPT-5.6 Sol 在 Deep Suite 1.1 評估中表現優異，成本低於競爭對手一半；推出 Ultra Mode 並解決獎勵黑客問題。
- 應用場景：涵蓋財務差異分析、簡報製作、前端開發、農場管理及日英即時翻譯。
詳細大綱
1. 模型發布與產品更新公告
- 模型可用性：
- GPT-5.6、Sol、Terra、Luna 發布。
- Sol：未來 24 小時內對所有付費計劃開放。
- Terra 和 Luna：對所有免費用戶開放。
- 背景：每週近十億用戶使用 ChatGPT，能力從單純回答轉變為軟體開發與複雜任務處理。
- 三大產品更新：
- ChatGPT Work：定位為「雄心壯志的工作」夥伴。支援網頁、行動裝置、桌面。能力包括理解複雜內容、長程任務追蹤、試算表創建/編輯、網站創建。
- 全新 ChatGPT 桌面應用程式：可處理本地檔案、應用程式及瀏覽器。
- Hosted Sites：所有付費用戶可創建並與朋友/同事分享網站。
2. 產品演示 (Demo)
- ChatGPT Work 演示 (Jessica & Lauren)：
- Jessica (工程師)：
- 任務：了解內部 ChatGPT Work 使用情況，分析 Slack 與員工反饋。
- 行動：識別多樣化角色用戶，安排會議時間。
- 結果：成功找到目標人物並安排會議。
- Lauren (戰略財務團隊)：
- 任務：財務預測與差異分析 (Variance Analysis)。
- 流程：
- 運行差異分析，解釋為何超出預測 (Demo 數字：超出 200 萬美元)。
- ChatGPT 更新 Excel 模型中的預測案例。
- 生成 PowerPoint 簡報，包含預測與驅動因素橋接 (Driver's Bridge)。
- 創建可分享的網站 (Site) 進行儀表板展示。
- 透過 Slack 發送網站連結給業務夥伴 (Tivo)。
- ChatGPT 桌面應用程式演示 (Andrew & Ed)：
- 本地檔案處理：
- 拖入大型試算表 (使用者票證匯出)。
- 生成互動式視覺化圖表，歸納主題與行動項目。
- 多來源內容整合與簡報生成：
- 整合資料夾內的 Launch Readiness PDF、UXR 訪談、安全審查 (Pen test) 及合規控制。
- 讀取 Chrome 瀏覽器中開啟的三個分頁。
- 結果：90 秒內生成符合公司模板的完整簡報。
- 特性：利用記憶功能處理未透過 Slack/Email 分享的擔憂。
- 搜尋功能：快速搜尋體驗，即時結果，豐富搜尋小工具。
- 電腦操作能力 (Computer Use)：
- 任務：整理雜亂的 Apple Notes (Brain dump)。
- 行動：ChatGPT 獲得游標控制權，在後台自動建立資料夾並移動筆記。
- 狀態：使用者可繼續操作其他內容，ChatGPT 在背景執行整理工作。
3. 前端開發與視覺化實例
- 從電子表格轉為互動式視覺化儀表板。
- 設計師 Ed 的實戰演示：
- 單一提示詞生成完整互動網站。
- 整合 Slack、Gmail 等連接器與插件。
- 即時修改與互動原型設計（如 3D 遊戲化導航）。
- 設計師對細節（排版、動畫）的評價。
- 其他團隊應用案例：
- Web 團隊：協作式互動儀表板。
- ChatGPT Images 團隊：收集全球活動案例。
- 設計團隊：互動式原型替代靜態設計稿。
- 趣味案例：3D 鸕鶿騎三輪車原型。
- 內部協作與工作流改變：取代 Excel 與靜態文件，提升團隊溝通與反饋效率。
4. 模型架構與研究進展 (GPT 5.6 系列)
- 訓練方法：2024 年宣布的推理範式（reasoning paradigm），結合強化學習與預訓練擴展。
- 模型家族介紹：
- SOL：最強大模型，針對最複雜的代理工作流（agentic workflows）。
- Terra：較快模型，針對日常工作流程。
- Luna：最快且最具成本效益模型，針對高容量工作。
- 自主研究能力：SOL 自主執行 Luna 的後訓練任務。
- 性能評估與應用場景：
- 基準測試表現：Terminal Bench, Browse Comp, Agent's Last Exam。
- 電腦使用（Computer Use）能力：瀏覽網頁、操作桌面應用。
- 專業領域應用：醫療記錄、藥物分析、金融模型。
- 專家反饋：速度為市場其他產品的三倍。
5. GPT 5.6 Sol 效能與定價策略
- Token 效率為多年研究重點。
- Deep Suite 1.1 評估結果：低於一半成本，表現優於競爭對手。
- Ultra Mode 發布：
- 概念：釋放代理程式團隊協作。
- 優勢：並行處理工作，提升速度與準確性。
- 應用：Codex app 中的新功能。
6. 安全性與模型對齊（Safety & Alignment）
- 修正 GPT 5.5 的獎勵黑客問題（goblins/gremlins 話題）。
- 投入超過 700,000 個 A100 等效小時的計算資源進行紅隊測試。
- 六週的安全訓練與測試。
- 新增監控升級：分類器（classifiers）與激活探針（activation probes）。
7. 開放生態與漏洞發現
- Project Daybreak：讓網路安全研究人員提前接觸模型。
- 成果：在主要瀏覽器與資料庫中發現漏洞。
- Patch the Planet：與開源貢獻者合作，自動生成高品質修補程式（例如 Linux 接受超過一半的修補）。
8. 產品生態與應用演示
- 產品線：ChatGPT (mobile/web), ChatGPT desktop app, GPT 5.6 Sol, Terra, Luna。
- 案例研究：Hiroki 使用 GPT 5.6 管理農場。
- 功能演示：ChatGPT Live Translate（即時日英雙向翻譯）。
9. 與全球農家及日本農林生產省聯繫經驗
- 收到來自世界各地農家的訊息及日本農林生產省（逐字稿中重複提及為「農林生產省」）的聯絡。
- 當前挑戰是善用 AI。
- 目前在群組聊天（Group Chat）中引入 AI，用於記錄日常事項及追蹤蔬菜狀態。
- 目標：建立讓團隊工作更舒適的機制，並讓 ChatGPT 支援或傳達此機制。
- 給非技術人員的建議：從「小規模嘗試」開始，透過向 ChatGPT 提問逐步建立機制。
工具 / 模型 / 名詞整理
- 模型名稱：
- GPT-5.6
- Sol (或 SOL)
- Terra
- Luna
- GPT-5-6-Soul (口語或聽寫變體)
- GPT 5.5
- 產品/功能名稱：
- ChatGPT
- ChatGPT Work
- Hosted Sites
- ChatGPT desktop app (桌面應用程式)
- Chat (指代 ChatGPT 的聊天介面)
- OnePass (財務系統或流程名稱，疑點)
- Site / Sites (指代 ChatGPT 創建的網站功能)
- Ultra Mode
- Codex app
- Chatea Petitra Live Translate (逐字稿原文，疑為 ChatGPT Live Translate)
- Group Chat (群組聊天)
- 第三方工具/平台：
- Slack
- Excel / Spreadsheet (電子表格)
- PowerPoint
- Calendar (日曆應用)
- Chrome (瀏覽器)
- Apple Notes
- Mac (作業系統)
- Figma (被明確否定使用)
- Gmail
- SVG (可縮放向量圖形格式)
- 評估與技術術語：
- Terminal Bench
- Browse Comp
- Agent's Last Exam
- Deep Suite 1.1
- Token efficiency
- Reward hacking
- Model red teaming
- Classifiers
- Activation probes
- Project Daybreak
- Patch the Planet
- A100 equivalent hours
- Variance Analysis (差異分析)
- Driver's Bridge (驅動因素橋接圖)
- Pen test (滲透測試)
- UXR (使用者體驗研究)
- Launch Readiness PDF (發布準備文件)
- agentic workflows (代理工作流)
- post-trained (後訓練)
- underspecified prompt (定義不完整的提示詞)
- Frontier evals (前沿評估)
- Codecs (疑點，可能指代 Coding 或特定技術術語)
操作流程整理
- 財務與簡報流程 (Lauren)：
- 運行差異分析，解釋超出預測原因。
- ChatGPT 更新 Excel 模型中的預測案例。
- 生成 PowerPoint 簡報，包含預測與驅動因素橋接。
- 創建可分享的網站 (Site) 進行儀表板展示。
- 透過 Slack 發送網站連結給業務夥伴。
- 桌面應用與檔案處理流程 (Andrew & Ed)：
- 拖入本地大型試算表。
- 生成互動式視覺化圖表，歸納主題與行動項目。
- 整合資料夾內的 PDF、訪談、安全審查及合規控制文件。
- 讀取 Chrome 瀏覽器中開啟的分頁。
- 90 秒內生成符合公司模板的完整簡報。
- 利用記憶功能處理未分享的擔憂。
- 使用搜尋功能尋找體驗。
- 授予 ChatGPT 游標控制權，自動整理 Apple Notes (建立資料夾並移動筆記)。
- 前端開發流程 (Ed)：
- 使用單一提示詞生成完整互動網站。
- 整合 Slack、Gmail 等連接器與插件。
- 即時修改與互動原型設計。
- 農場管理流程 (Hiroki)：
- 在群組聊天中引入 AI。
- 記錄日常事項及追蹤蔬菜狀態。
- 透過向 ChatGPT 提問，逐步建立可用於工廠的機制。
值得注意的限制或風險
- 獎勵黑客問題：GPT 5.5 訓練期間出現過度提及 goblins/gremlins 的問題，已透過紅隊測試與安全訓練修正。
- 技術非專業人士的使用門檻：建議從小規模嘗試開始，逐步建立機制。
- 電腦操作能力：雖然演示中展示了自動整理筆記，但需授予游標控制權，且為背景執行。
- 模型名稱與命名慣例：GPT-5.6 及 Sol, Terra, Luna 的命名與常見慣例不同，可能存在辨識或命名上的特殊情況。
逐字稿辨識疑點
- GPT-5.6：逐字稿中多次出現此名稱，與常見命名慣例不同，保留原樣。
- Sol：模型名稱，保留原樣。
- Terra：模型名稱，保留原樣。
- Luna：模型名稱，保留原樣。
- Codecs：在 "especially with codecs and the coding abilities" 中出現，疑點為聽寫錯誤或特定內部術語，無法確定是否為 "coding" 的誤聽或特定技術詞彙。
- OnePass：在 "now in OnePass, ChatGPT can run the variance analysis" 中出現，疑點為財務系統名稱、內部工具名稱或聽寫錯誤 (可能為 "One Pass" 或特定軟體名)。
- GPT-5-6-Soul：在 "This is one thing that GPT-5-6-Soul is really incredible at" 中出現，疑點為模型名稱的連讀或聽寫錯誤，可能指代 GPT-5.6 或 Sol。
- Tivo：在 "walk Tivo and Jessica" 及 "send the site link over to Tivo on Slack" 中出現，疑點為人名或特定角色代稱，非已知產品名稱。
- Next World Cup game：在 "curious about the next World Cup game" 中出現，疑點為具體事件指涉，保留原樣。
- Ed and team have been cooking：口語表達，意指團隊正在開發或準備某些東西。
- 5.6 Sol / SOL：逐字稿中交替出現 "5.6 Sol" 與 "SOL"，且前文提及 "new 5.6 Sol model"。需查證是否為特定模型代號或版本名稱。
- TragiPT：在提示詞修改環節，講者說 "So, hi, TragiPT, let's change this website." 疑為口誤或特定內部暱稱，聽起來像 "ChatGPT" 的變體或口誤。
- Scythe：講者說 "And the really cool thing about Scythe is they're collaborative..." 疑為口誤，上下文應指代前文提到的原型或工具，聽起來像 "ChatGPT" 或 "Code" 的聽寫錯誤，但無法確定。
- Kian：提及 "Kian, one of the folks on the team, he built this fun prototype..." 人名拼寫需查證。
- Tayal：結尾處 "You might hear the capabilities that Tayal is talking"，前文講者為 Tejal，此處聽寫為 "Tayal"，疑為口誤或聽寫錯誤。
- cyber：講者說 "And it's also incredible at cyber"，疑為口誤，可能意指 "coding" 或其他技術領域，但逐字稿記錄為 "cyber"。
- Chatea Petitra Live Translate：逐字稿中出現此名稱，疑似為 "ChatGPT Live Translate" 的聽寫錯誤或特定產品名稱，需查證。
- Deep Suite 1.1：需查證是否為標準評估基準名稱。
- goblins and the gremlins：逐字稿中提及此作為 GPT 5.5 獎勵黑客的結果，需確認是否為特定測試案例或口誤。
- 重複語句：逐字稿中出現大量重複的 "building of the building..." 語句，疑似為演示時的技術故障或特定測試內容，非正常語義。
- Hirokisan：人名提及方式，需確認是否為正式姓名或稱呼。
- 農林生產省：逐字稿中多次出現此詞彙，與日本官方機構名稱「農林水產省」略有不同，保留原樣。
- ライブラウンド：聽寫為「Library Round」或類似發音，語意不明，疑為口誤或專有名詞。
- 音とAI：語境中疑為「AI」或特定工具名稱的聽誤，保留原樣。
- チャットGPT：ChatGPT 的日文片假
尚未產生學習筆記
請在 Telegram 指令最後加上「學習」，例如：videonote 網址 英文 雙語 學習