5 Ways Claude Tag Could Change How You Use AI

The AI Daily Brief: Artificial Intelligence News 首頁 每日列表 原始影片
實際影片長度：23:05.000。原文、繁中、雙語可點擊句子跳轉影片。
0:00.560–0:07.120
today on the ai daily brief why clawed tag and approaches like it might change how you use ai
0:08.240–0:13.440
before that in the headlines an anthropic customer has sued the us over losing fable
0:13.440–0:18.400
but can it possibly help the ai daily brief is a daily podcast and video about the most
0:18.400–0:24.080
important news and discussions in ai all right friends quick announcements before we dive in
0:24.080–0:28.800
i am continuing to remind folks about the executive agent leadership program which is the
0:28.800–0:33.680
enterprise grade descendant of enterprise claw now offered in conjunction with super intelligent
0:33.680–0:39.920
you can find out about that at training.bsuper.ai the next cohort begins on monday june 29th so if
0:39.920–0:46.480
you're interested in that check it out well here we are heading towards the back half of this week
0:46.480–0:54.240
with a return of fable 5 no closer insight in fact the rumor mill suggests that we are not in line for
0:54.240–0:59.520
pretty much any models for some time later this week i'm going to do a show about what you can
0:59.520–1:04.880
do to make the most of this forced ai model pause but for now let's look at some of the conversations
1:04.880–1:09.920
that are actually shaping the discourse now first up i flagged this before but i want to take this
1:09.920–1:15.120
chance to flag it again right now emotions are running at an 11 when it comes to this anthropic news
1:15.760–1:21.200
i saw a post go viral on x about the significance of this bipartisan letter asking our lutnik if
1:21.200–1:25.440
anthropic had been singled out but friends this is at this point about a week old and these sort of
1:25.440–1:30.480
open congressional letters don't actually do anything they don't usually get responses this is
1:30.480–1:35.280
in a word a nothing burger one thing that also probably amounts to nothing but is at least a little
1:35.280–1:40.240
bit more interesting is that one anthropic customer is taking it upon themselves to get the model
1:40.240–1:46.080
reinstated by suing the government in a lawsuit filed on tuesday a legal tech firm called legion
1:46.080–1:50.640
claimed the ban was illegal and damaging their business the harm to legion is immediate
1:50.640–1:55.600
irreparable and existential the lawsuit stated the pace of frontier ai advancement is blistering and
1:55.600–2:01.280
competitive ground lost during a suspension cannot be regained after the fact certainly regardless of
2:01.280–2:05.600
the ultimate success of this lawsuit i think that is a sentiment that many of you listening now will
2:05.600–2:10.720
agree with the filing claimed that export control laws do not cover shutting down access to cloud
2:10.720–2:15.440
software a point that many other legal commentators made as soon as the ban was announced serving
2:15.440–2:20.480
computer code or text outputs from a computer isn't typically seen as an export that can be proper
2:20.480–2:26.720
subject of export controls in the alternative if the ban relied on ieepa powers similar to sanctions
2:26.720–2:31.200
the lawsuit notes that these powers do not apply to information and no prerequisite national security
2:31.200–2:36.080
emergency was declared further the lawsuit argues the move was overbroad and possibly retaliatory
2:36.080–2:40.560
towards anthropic it also notes the decision to block distribution of a model directly contradicts
2:40.560–2:44.800
the executive order sign at the beginning of june which ruled out the idea of a model licensing
2:44.800–2:49.280
scheme in making their argument legion centers the facts on their canadian development team losing
2:49.280–2:53.280
access to the model pointing out that targeting allied nations with export controls is highly
2:53.280–2:57.440
unusual anthropic for their part remains silent on the matter and there is certainly no indication
2:57.440–3:01.600
this lawsuit is being filed on their behalf by a third party in comments to the press they referred
3:01.600–3:05.280
to previous statements indicating they're grateful to the administration for working quickly to resolve
3:05.280–3:10.560
the issue now in terms of the potential success of this at this point most legal analysts that i've
3:10.560–3:14.160
seen seem to think that the administration has gone outside the bounds of the law on this one
3:14.160–3:18.640
but also that it's unlikely to be a clear-cut decision for a judge and frankly none of that is going to
3:18.640–3:23.200
actually matter because as long as anthropic is taking it seriously which they clearly are presumably
3:23.200–3:27.840
and oh my gosh hopefully it'll be resolved through negotiation before any legal process could actually
3:27.840–3:33.040
make it through the courts now the administration's friskiness when it comes to controlling the flow
3:33.040–3:37.600
of ai is doing nothing but increasing the most recent news is that the administration is now
3:37.600–3:41.840
pressuring meta to agree to voluntary model review sources told the new york times that the white
3:41.840–3:46.480
house is pressing meta to submit their ai models for safety and performance testing prior to release
3:46.480–3:51.520
meta is currently the lone holdout among major ai labs after microsoft xai and google signed
3:51.520–3:55.760
agreements shortly before the recent executive order on ai and cyber security now reportedly the
3:55.760–3:59.440
administration's requests seem to only relate to testing at the center for ai standards and
3:59.440–4:03.520
innovation at the commerce department not whatever additional security testing has been going on at
4:03.520–4:08.320
the nsa and meta does seem to be leaning towards an agreement with a spokesperson commenting
4:08.320–4:12.960
we share the administration's goal of advancing us leadership on robust and secure frontier ai
4:12.960–4:17.680
while we are working through the details we hope to sign the agreement soon still the whole idea of
4:17.680–4:22.000
pressuring a company to submit their technology for government testing really highlights just how
4:22.000–4:28.240
air quotes voluntary the program actually is and having contained advanced ai models the commerce
4:28.240–4:32.880
department is reportedly turning to robots as their next technology to ban politico reports that
4:32.880–4:37.280
commerce secretary howard lutnik held a closed door meeting on monday to discuss the risk of chinese-made
4:37.280–4:40.480
models according to notes provided to politico lutnik said
4:41.200–4:46.160
we don't want state subsidized robotics attacking us in america this is the arms race that is coming
4:46.160–4:49.920
robotic arms are coming we need to make sure they're produced in america so we're going to study
4:49.920–4:54.320
those right now commerce department officials reportedly view state-backed chinese robot companies
4:54.320–4:59.120
as a massive threat fearing they could proliferate through global markets before u.s alternatives hit scale
5:00.000–5:03.920
chinese robots are already subject to steep tariffs but the commerce department is considering
5:03.920–5:08.320
additional action the monday meeting included more than a dozen executives from major firms including
5:08.320–5:13.440
boston dynamics spacex siemens and goldman sachs the major topic of the meeting was how the u.s can
5:13.440–5:18.240
best rebuild its industrial base but apparently the risks of chinese robots dominated the discussion
5:18.240–5:22.400
one source said the whole idea that what we're going to end up with is an american brain with a
5:22.400–5:28.400
chinese body is a very very bad strategic plan now it's unclear from the reporting how fast we should
5:28.400–5:34.720
expect to crack down but the sabers are already rattling elsewhere in washington on monday the x account for
5:34.720–5:39.280
the congressional select committee on china posted a screenshot of a unitary humanoid robot for sale
5:39.280–5:43.680
through amazon it's priced at seventeen thousand nine hundred ninety dollars and can ship to washington
5:43.680–5:49.200
in a month the select committee added unitary was recently designated as a chinese military company
5:49.200–5:53.680
and its products are a threat to our national security yet here is amazon selling a unitary robot
5:53.680–5:58.560
in america we need chairman mulnar's guard act to stop this threat and support american robotics
5:58.560–6:02.800
now several chinese robotics firms have received the same designation which many see as a precursor to
6:02.800–6:08.720
a huawei style ban now on the topic of american-made robots elon musk recently said that they are in the
6:08.720–6:13.520
final stages of completion of optimus 3 which he says is really going to be by far the most advanced
6:13.520–6:18.720
robot in the world nothing's even close but people are i would say fairly skeptical of those claims
6:19.600–6:26.160
speaking of elon though xai has followed suit to openai and anthropic introducing the goals primitive
6:26.160–6:31.840
in an update to their coding harness grok build as with the goal command in codex this allows users to
6:31.840–6:36.640
define an outcome and set grok to work on a long horizon task grok will continue working across
6:36.640–6:40.960
multiple steps deploying sub-agents to complete individual tasks and orchestrating the entire
6:40.960–6:45.360
process without user intervention the orchestration process includes planning implementation and
6:45.360–6:50.560
verification to ensure a high quality output so a couple things to make this worthy of inclusion
6:50.560–6:55.920
here first of all it reinforces the idea that slash goal is an actual new ai ux primitive not just a
6:55.920–7:01.040
generic feature but second this is our first glimpse of what the xai cursor partnership will look like
7:01.040–7:05.760
after the acquisition was finalized last week the goal command natively uses a fine tune of grok
7:05.760–7:11.920
called grok build 0.1 as well as cursor's composer 2.5 to complete its tasks and the combination has at
7:11.920–7:16.320
least the potential to be more than the sum of its parts composer is known for cost-effective high
7:16.320–7:20.560
performance making it a natural fit for coding sub-agents and while the latest version of the
7:20.560–7:25.760
flagship grok model didn't grab that much attention on release one big point of differentiation it tried to
7:25.760–7:31.200
make was a sophisticated orchestration process built into the model when used as a chatbot this mostly
7:31.200–7:35.360
just resulted in sub-agents arguing over how to interpret information but that same process applied
7:35.360–7:40.320
to coding tasks could make grok really good at that sort of orchestration we'll see what comes next and
7:40.320–7:45.040
whether this makes any waves but certainly the update to build reinforces the idea that xai at this
7:45.040–7:50.080
point as much as spacex is becoming a neocloud isn't quite done releasing ai products just yet
7:50.640–7:56.240
speaking of ai products one that is getting a lot of attention is byte dance's seed dance 2.5 we've
7:56.240–8:01.120
started to get a ton of previews of the new model which doubles maximum clip length from 15 to 30 seconds
8:01.120–8:06.400
and adds 4k resolution the new version also supports up to 50 input references to use as objects or
8:06.400–8:12.560
characters within scenes by comparison seed dance 2.0 could handle 12 references and google's vo 3.1
8:12.560–8:18.000
released back in january maxed out at 3. c dance 2.5 is also able to make use of image video and audio
8:18.000–8:22.880
files as references the first time anyone has moved beyond images now so far it doesn't look
8:22.880–8:26.560
like anyone has actually got their hands on the model so we're just going off of a demo clip
8:26.560–8:30.560
provided by byte dance yet that's enough for people to suggest that this will be the mythos
8:30.560–8:36.160
moment of video models simon smith who i'm fairly sure isn't getting paid by byte dance by the way
8:36.160–8:40.160
that's a joke if you've heard simon smith quoted here he was very very much not getting paid by byte
8:40.160–8:45.600
dance wondered how they managed to gain such a big lead in video he asked seed dance seemed to be
8:45.600–8:50.480
pulling away in the video generation race is there a recursive feedback loop at work here if so what
8:50.480–8:54.640
is it are they generating training data with each generation to train the next if so what do they do
8:54.640–8:58.720
to increase quality with each iteration whatever it is the corpus of training content they have access
8:58.720–9:03.760
to through tick tock seems to be paying off lastly today i don't want to get into it deeply but one
9:03.760–9:07.840
story that i am keeping an eye on is some of the market moves that i had previously discussed with
9:07.840–9:12.560
regard to google specifically moving to a broader ai sell-off to the extent that this becomes a more
9:12.560–9:16.640
significant narrative trend as opposed to just a particular market moment we'll come back and
9:16.640–9:21.200
explore it more for now though that's going to do it for today's headlines next up the main episode
9:22.400–9:28.640
welcome back to the ai daily brief sometimes big stories in ai are extremely obvious for example the
9:28.640–9:33.920
u.s government sending a note telling anthropic that they're using export controls to deny access to the
9:33.920–9:38.800
latest model to everyone who's not american and anthropic responding by taking down the model that they
9:38.800–9:43.840
had released just a few days ago for everyone that's a big story in a very obvious way today's
9:43.840–9:47.760
story is big not because of some geopolitical implications but because of the way that it
9:47.760–9:53.360
potentially changes at somewhat of a core level how we think about interacting with ai at work now
9:53.360–9:58.080
the feature we're discussing is called claude tag which claude announced on tuesday as a new way for
9:58.080–10:03.840
teams to work with claude and to put it really simply it's basically claude in slack but not just as a
10:03.840–10:09.600
different channel to prompt claude but instead as a full team member that has access to all of the
10:09.600–10:13.840
context from all the different channels as well as the different tools that your team is already using
10:13.840–10:19.520
anthropic writes claude tag is an evolution of claude code made more proactive and built to work with a
10:19.520–10:24.800
full team it's now one of the main ways we get things done at anthropic 65 of our product team's code
10:24.800–10:30.160
now comes from our internal version tag claude with a request and it breaks the task into stages works
10:30.160–10:34.480
through them with the tools it has access to and responds in the thread with what it creates it'll
10:34.480–10:39.920
write or merge pull requests run data analysis or help resolve an incident in a channel that's a
10:39.920–10:44.640
channel on slack there's one claude that interacts with everyone so a teammate can pick up exactly where
10:44.640–10:49.120
you left off it builds more context about the work as it follows the channel so you don't need to
10:49.120–10:54.480
explain things from scratch turn ambient behavior on and claude takes initiative it follows up on threads
10:54.480–10:59.360
that have gone quiet and flags what's relevant from across its channels and tools so for someone who's
10:59.360–11:04.080
moving fast you could be forgiven for thinking it's just another agent in slack sort of announcement
11:04.080–11:08.400
of which we've had a fair few but just based on the way that the anthropic team was talking about
11:08.400–11:13.360
it there's clearly something going on here anthropic team member alexander bricken writes
11:13.920–11:18.560
claude tag has completely changed the way we do work at anthropic having the multiplayer experience
11:18.560–11:22.800
of claude within slack with all of the fundamental primitives of claude code like proactivity
11:22.800–11:27.120
subagents and long horizon task management is an entirely new way to do work asynchronously
11:27.120–11:33.680
tobin south writes claude tag is how i do 90 plus of my work i even had people from anthropic texting
11:33.680–11:38.800
me specifically when i didn't cover it on yesterday's show basically arguing behind the scenes that this
11:38.800–11:43.200
is a big deal now i had already begun planning this episode when those texts started coming in
11:43.200–11:48.000
but still that should tell you how the team behind it views its significance now this is not the only
11:48.000–11:52.640
feature of this type that's available now anthropic is not the first company to release something like
11:52.640–11:58.000
this the semi-analysis team wrote perplexity's computer slack co-worker has been wildly useful
11:58.000–12:02.960
internally at semi-analysis which is why anthropic launched a computing product the team is configuring
12:02.960–12:07.920
claude tag ai co-workers and is excited to compare and contrast them with perplexity computer click health
12:07.920–12:14.000
simon smith wrote immediately asked our cto to activate claude tag we already use chat gpt workspace
12:14.000–12:18.160
agents in slack they're great for a bunch of use cases like providing support based on prior resolve
12:18.160–12:23.280
slack posts and summarizing slack activity to create a weekly brief the difference with claude tag is
12:23.280–12:27.200
that this level of configuration isn't even necessary you just drop claude into a slack
12:27.200–12:31.520
channel and it picks up the context and becomes useful we currently create one slack channel per
12:31.520–12:36.080
project we've historically also created chat gpt projects and notebook lm notebooks along with those
12:36.080–12:40.800
slack channels to work with ai duplicating or triplicating context across all those surfaces
12:40.800–12:45.200
now we can just create the slack channel as we've always done and drop claude in i also think this will
12:45.200–12:49.600
massively help with ai diffusion people don't have to install claude desktop to access claude's most
12:49.600–12:53.600
powerful capabilities in co-work and code they can just interact with claude as if it's another
12:53.600–13:00.160
person in slack no less than andre carpathy argued that this is indeed a bigger deal than it seems at first
13:00.160–13:06.720
he wrote this is a new paradigm for interacting with claude that is significantly more in line with
13:06.720–13:10.880
all the other human activity org wide once you do all of the under the hood engineering work to make
13:10.880–13:16.080
this just work eg across tools integrations compute environments memory security etc claude basically
13:16.080–13:20.000
joins the team in a seamless way you can talk to it as you would talk to a person and it can help with
13:20.000–13:26.880
a very large variety of workloads in my opinion this is the third major redesign of llm ui ux the first
13:26.880–13:31.920
paradigm was that the llm is a website you go to the second was that it is an app you download on your
13:31.920–13:37.280
computer the third one is that it is a self-contained persistent asynchronous entity with org wide tools and
13:37.280–13:42.000
context working alongside teams of humans it really takes a while to wrap your head around it but it
13:42.000–13:48.720
works and it is awesome now for a lot of folks the big eye blinking statistic was that anthropic is
13:48.720–13:56.560
saying that 65 of their product team's code now comes from claude tag ejaz wrote it's blowing my mind
13:56.560–14:02.000
that 65 of product code in anthropic is now written by tagging claude in group chats of staff discussing
14:02.000–14:06.640
what they want to build r.i.p the days of tediously writing long product docs now you can literally go
14:06.640–14:11.760
from slack to a production ready feature cloud code is barely a year old by the way developer nick
14:11.760–14:17.840
dobos wrote anthropic isn't using cloud code anymore major agent ux shift alert and i think this is one
14:17.840–14:21.440
of the sneaky things that's going to take people a little while to fully wrap their heads around
14:21.440–14:27.600
while you might be tagging at claude what you're really calling upon is claude code now certainly
14:27.600–14:32.560
both anthropic and openai have already spent lots of cycles trying to make the coding capabilities of
14:32.560–14:37.040
their tools more accessible for non-coders but this is the biggest move in that direction yet
14:37.040–14:41.760
when claude already exists in the workspace that you're operating in the barrier to using it to
14:41.760–14:46.400
build something is having an idea and dropping it into the slack where you're already working
14:46.400–14:51.040
now while it's early days for non-anthropic users many were quick to share what they had learned from
14:51.040–14:56.560
using claude tag over the past few weeks or months claude code team member tarik pointed out that each claude in
14:56.560–15:01.920
each channel is different one suggestion he had was to introduce air quotes claude to a new channel
15:01.920–15:06.000
with a pin message and a set of instructions to remember like which persona to take when to respond
15:06.000–15:10.480
etc he says you can think of it like it's claude.md because all of this will get added to its memory
15:10.480–15:15.600
tarik also suggests having a personal channel eg tarik-claude where you can tag claude for your own
15:15.600–15:20.800
work and give it instructions specific to you you can then forward other messages to that channel so your
15:20.800–15:25.760
claude can get to work the way you like he also noted that this helps with overwhelm because when things
15:25.760–15:30.160
get going it can be hard to keep track of all the threads he wrote in my personal channel i like to
15:30.160–15:34.800
ask claude to keep a pin message that it updates with the status of everything he even uses a set
15:34.800–15:39.520
of emojis to be able to help him understand where different work streams are in a single glance
15:39.520–15:44.000
chris taylor from fractional who joined the broader anthropic ecosystem to run their version of a
15:44.000–15:48.320
forward deployed engineering group described how they had been using it across their company he said
15:48.320–15:52.320
it kind of works like a co-worker that just picks things up it owns busy work daily status
15:52.320–15:55.600
summaries following up when people are out of office nudging teammates on open to
15:55.600–15:59.920
dues pushing back due dates his plans shift chris said that when one of its own scheduled tasks
15:59.920–16:04.400
started misfiring it caught the bug fixed it and wrote it up without being asked he also said that
16:04.400–16:08.960
it manages itself we needed somewhere to track bugs and ideas for claude tag while we were testing it
16:08.960–16:14.000
so it built that tracker for us and files every new issue and use case on its own finally reinforcing
16:14.000–16:19.680
that this is in fact a fully fledged version of claude code not just claude he says it writes and ships
16:19.680–16:23.680
handed one of our code bases claude read through it wrote up the 10 most important problems to fix
16:23.680–16:29.040
stack ranked then made the changes and opened them for review all in slack the claude dev team shared
16:29.040–16:33.840
that they're using it for incident response bug triage dependent work i.e work that's blocked on
16:33.840–16:39.360
something else background watching and monitoring metrics nityesh from the every team wrote about
16:39.360–16:44.960
why this will feel familiar to many he said claude tag has completely changed the way i do work for the
16:44.960–16:50.160
last four months except it's not claude tag anthropic only announced that a few hours ago and i don't even
16:50.160–16:54.240
have access to it but i did build a version of it for myself which i've been using for months now
16:54.240–16:58.880
basically he said that after openclaw came online he built a harness that allowed him to turn any mac
16:58.880–17:03.280
into an ai employee with the claude code headless mode he then ended up building three such employees
17:03.280–17:08.800
and had them working in slack for the last few months now this idea of agents inside slack is also
17:08.800–17:12.960
something that we've experimented a lot with over here as well one of the reasons that i burned through
17:12.960–17:18.320
a billion tokens in march is that i was experimenting with a series of ai enablement agents that lived in
17:18.320–17:22.800
slack and could do things like figuring out how we were using ai providing coaching on how to better
17:22.800–17:28.480
use ai and developing overall ai strategy now you also might remember me talking about a month ago
17:28.480–17:34.320
about how the team at every had shifted their agent strategy from originally having every employee build
17:34.320–17:39.520
an ai agent version of themselves to instead having a group of agents that worked across employees based
17:39.520–17:44.800
on function and goal and had shared context claude tag in many ways seems to be institutionalizing some
17:44.800–17:50.480
of those shifts as best practice so trying to sum this all up let's talk about the five big shifts
17:50.480–17:56.800
that i think claude tag could represent and keep in mind this is not just about claude tag as a feature
17:56.800–18:01.920
but about it as a leading indicator of how labs and agent app companies are thinking about the future of
18:01.920–18:08.240
ai at work the first big shift is from an app native interface to existing workplace interfaces obviously
18:08.240–18:13.440
this breaks agents out of the controlled settings of the web app or the desktop apps that the labs have
18:13.440–18:18.720
built and instead uses the existing spaces that people do work in now to some extent one has to
18:18.720–18:23.440
think that this was inevitable as you try to shift behavior and get everyone working in a new way
18:23.440–18:27.600
one of the biggest points of friction is the new interfaces that they have to go interact with
18:27.600–18:32.800
the fact that the full power of claude code and everything that it can build is now accessible from
18:32.800–18:37.760
the most default channel where people are working seems likely to me to make a big difference in terms
18:37.760–18:44.080
of who accesses that power likewise this moves ai from a private chatbot to a shared teammate experience
18:44.080–18:47.840
now of course there have been no shortage of sharing based features to break your personal
18:47.840–18:52.480
chats out of their home in fact both openai and anthropic have released versions of better native
18:52.480–18:57.280
websites slash web app builders in order to help make sharing more easily this of course takes that
18:57.280–19:02.480
to the next level which also leads to our third shift which is from single user context i.e what the
19:02.480–19:08.640
ai or agent knows about you to the full team context part of the power of moving the agent from its native
19:08.640–19:13.360
home to your team's native home is that it gets access to the ambient context that exists all around
19:13.360–19:18.960
it without you having to create an entirely different way of accessing all that context now already
19:18.960–19:23.600
claude tag also reinforces a shift which was already very much happening which is the shift from
19:23.600–19:28.720
prompting to delegation as the main mode of interacting with ai already throughout the year as agent
19:28.720–19:32.960
the capacity has come online people have been slowly but surely breaking out of the way that they used to
19:32.960–19:38.080
interact with these tools a big part of that has been in shifting from telling the ai what to do to
19:38.080–19:42.560
telling it what you are trying to accomplish and giving it more freedom and latitude to accomplish an
19:42.560–19:49.360
ever-expanding scale of goals lastly claude tag and more broadly agents that work in your team's workspaces
19:49.360–19:54.480
shift ai's importance from something that becomes personally essential to key workers to an actual
19:54.480–19:59.600
organizational dependency a new way of working not just individually but across the entire team
19:59.600–20:04.000
which is not to say that there aren't challenges first of all while yes this is theoretically just
20:04.000–20:09.680
dropping claude into slack there are a lot of questions around access and tool configuration that
20:09.680–20:14.400
make it set up potentially much more complicated than it at first seems anthropic released an entire
20:14.400–20:19.360
blog post about best practices for setting up agent identity in this new claude tag paradigm
20:19.360–20:24.000
they also pointed out some best practices around permissions which wouldn't necessarily be obvious at first
20:24.000–20:28.960
glance gail wiener points out that while the positive side is the democratization of claude
20:28.960–20:33.040
code capabilities to a broader subset of the organization the flip side of that is that something
20:33.040–20:38.240
like claude tag potentially runs into existing trust barriers and skepticism that could create a challenge
20:38.240–20:42.960
she describes a scenario where there are five people on a team with her as the ai power user two
20:42.960–20:48.000
members of the team not trusting ai one team member thinking that the power user uses ai too much
20:48.000–20:52.320
and another being indifferent gail writes i start tagging claude in the project chat and here's what
20:52.320–20:56.320
happens in that room the moment claude is in the channel every message anyone types is being read by
20:56.320–21:00.720
it it stops looking like a tool the team uses and starts looking like my tool that's now sitting in
21:00.720–21:04.960
everyone's workspace i become the person who brought the surveillance device to the meeting even if that's
21:04.960–21:10.080
not what it is the skeptics position gets validated every time claude produces something good see she's
21:10.080–21:14.560
outsourcing her thinking and every time it produces something mediocre see this is what we were worried
21:14.560–21:19.040
about you can't win that frame basically she's arguing that this creates a whole new challenge around the
21:19.040–21:24.720
human layer that will have to be solved as well one specific thing that seems fairly challenging to me
21:24.720–21:29.360
is that despite the fact that you just use at claude for every instance of where you want to call upon
21:29.360–21:34.560
these capabilities different clauds and different channels have potentially access to different tools
21:34.560–21:39.600
and different permissions and different contexts and that feels like it'll get confusing super super
21:39.600–21:46.160
fast in fact in simon smith's first impressions post later on he did write claude being many clauds is a bit
21:46.160–21:50.720
disorienting i've come to think of claude as my claude the one in the app that knows me and connects to
21:50.720–21:55.600
things i've connected it to but there is no single claude in slack it's channel dependent and admin
21:55.600–22:00.480
configured and none of the clauds are mine this seems to be a common feeling one of the first things i'm
22:00.480–22:04.640
seeing is people tagging claude in channels to ask what it can do what it remembers how it maintains
22:04.640–22:10.240
context in the channel indeed how this creates new challenges and opportunities for context is a big topic of
22:10.240–22:14.480
conversation and for some in light of the fable challenge in light of everything that's happened with
22:14.480–22:19.280
fable there's also increasingly good reasons to not want to outsource this sort of process to a single
22:19.280–22:24.400
closed provider hugging faces head of product victor writes at hugging face we've been building our own
22:24.400–22:28.720
agent that we use via slack honestly building your own is quite simple and you'll be happy you did
22:28.720–22:33.600
any model you want including self-hosted if needed fully customizable to your stack and of course no
22:33.600–22:38.880
lock-in and no waitlist and not overpriced look ultimately it is very easy to overestimate the
22:38.880–22:44.000
significance of any one new feature and yet if for no other reason than these significant
22:44.000–22:48.800
shifts that the anthropic team themselves are reporting it seems like claude tag is one paradigm
22:48.800–22:53.600
shift that will at least be worth exploring i'll report back as we try it out and see what we learn
22:53.600–22:57.440
for now though that's going to do it for today's ai daily brief appreciate you listening or watching
22:57.440–23:05.000
as always and until next time peace
0:00.560–0:07.120
（此句尚無繁中翻譯）
0:08.240–0:13.440
（此句尚無繁中翻譯）
0:13.440–0:18.400
（此句尚無繁中翻譯）
0:18.400–0:24.080
（此句尚無繁中翻譯）
0:24.080–0:28.800
（此句尚無繁中翻譯）
0:28.800–0:33.680
（此句尚無繁中翻譯）
0:33.680–0:39.920
（此句尚無繁中翻譯）
0:39.920–0:46.480
（此句尚無繁中翻譯）
0:46.480–0:54.240
（此句尚無繁中翻譯）
0:54.240–0:59.520
（此句尚無繁中翻譯）
0:59.520–1:04.880
（此句尚無繁中翻譯）
1:04.880–1:09.920
（此句尚無繁中翻譯）
1:09.920–1:15.120
（此句尚無繁中翻譯）
1:15.760–1:21.200
（此句尚無繁中翻譯）
1:21.200–1:25.440
（此句尚無繁中翻譯）
1:25.440–1:30.480
（此句尚無繁中翻譯）
1:30.480–1:35.280
（此句尚無繁中翻譯）
1:35.280–1:40.240
（此句尚無繁中翻譯）
1:40.240–1:46.080
（此句尚無繁中翻譯）
1:46.080–1:50.640
（此句尚無繁中翻譯）
1:50.640–1:55.600
（此句尚無繁中翻譯）
1:55.600–2:01.280
（此句尚無繁中翻譯）
2:01.280–2:05.600
（此句尚無繁中翻譯）
2:05.600–2:10.720
（此句尚無繁中翻譯）
2:10.720–2:15.440
（此句尚無繁中翻譯）
2:15.440–2:20.480
（此句尚無繁中翻譯）
2:20.480–2:26.720
（此句尚無繁中翻譯）
2:26.720–2:31.200
（此句尚無繁中翻譯）
2:31.200–2:36.080
（此句尚無繁中翻譯）
2:36.080–2:40.560
（此句尚無繁中翻譯）
2:40.560–2:44.800
（此句尚無繁中翻譯）
2:44.800–2:49.280
（此句尚無繁中翻譯）
2:49.280–2:53.280
（此句尚無繁中翻譯）
2:53.280–2:57.440
（此句尚無繁中翻譯）
2:57.440–3:01.600
（此句尚無繁中翻譯）
3:01.600–3:05.280
（此句尚無繁中翻譯）
3:05.280–3:10.560
（此句尚無繁中翻譯）
3:10.560–3:14.160
（此句尚無繁中翻譯）
3:14.160–3:18.640
（此句尚無繁中翻譯）
3:18.640–3:23.200
（此句尚無繁中翻譯）
3:23.200–3:27.840
（此句尚無繁中翻譯）
3:27.840–3:33.040
（此句尚無繁中翻譯）
3:33.040–3:37.600
（此句尚無繁中翻譯）
3:37.600–3:41.840
（此句尚無繁中翻譯）
3:41.840–3:46.480
（此句尚無繁中翻譯）
3:46.480–3:51.520
（此句尚無繁中翻譯）
3:51.520–3:55.760
（此句尚無繁中翻譯）
3:55.760–3:59.440
（此句尚無繁中翻譯）
3:59.440–4:03.520
（此句尚無繁中翻譯）
4:03.520–4:08.320
（此句尚無繁中翻譯）
4:08.320–4:12.960
（此句尚無繁中翻譯）
4:12.960–4:17.680
（此句尚無繁中翻譯）
4:17.680–4:22.000
（此句尚無繁中翻譯）
4:22.000–4:28.240
（此句尚無繁中翻譯）
4:28.240–4:32.880
（此句尚無繁中翻譯）
4:32.880–4:37.280
（此句尚無繁中翻譯）
4:37.280–4:40.480
（此句尚無繁中翻譯）
4:41.200–4:46.160
（此句尚無繁中翻譯）
4:46.160–4:49.920
（此句尚無繁中翻譯）
4:49.920–4:54.320
（此句尚無繁中翻譯）
4:54.320–4:59.120
（此句尚無繁中翻譯）
5:00.000–5:03.920
（此句尚無繁中翻譯）
5:03.920–5:08.320
（此句尚無繁中翻譯）
5:08.320–5:13.440
（此句尚無繁中翻譯）
5:13.440–5:18.240
（此句尚無繁中翻譯）
5:18.240–5:22.400
（此句尚無繁中翻譯）
5:22.400–5:28.400
（此句尚無繁中翻譯）
5:28.400–5:34.720
（此句尚無繁中翻譯）
5:34.720–5:39.280
（此句尚無繁中翻譯）
5:39.280–5:43.680
（此句尚無繁中翻譯）
5:43.680–5:49.200
（此句尚無繁中翻譯）
5:49.200–5:53.680
（此句尚無繁中翻譯）
5:53.680–5:58.560
（此句尚無繁中翻譯）
5:58.560–6:02.800
（此句尚無繁中翻譯）
6:02.800–6:08.720
（此句尚無繁中翻譯）
6:08.720–6:13.520
（此句尚無繁中翻譯）
6:13.520–6:18.720
（此句尚無繁中翻譯）
6:19.600–6:26.160
（此句尚無繁中翻譯）
6:26.160–6:31.840
（此句尚無繁中翻譯）
6:31.840–6:36.640
（此句尚無繁中翻譯）
6:36.640–6:40.960
（此句尚無繁中翻譯）
6:40.960–6:45.360
（此句尚無繁中翻譯）
6:45.360–6:50.560
（此句尚無繁中翻譯）
6:50.560–6:55.920
（此句尚無繁中翻譯）
6:55.920–7:01.040
（此句尚無繁中翻譯）
7:01.040–7:05.760
（此句尚無繁中翻譯）
7:05.760–7:11.920
（此句尚無繁中翻譯）
7:11.920–7:16.320
（此句尚無繁中翻譯）
7:16.320–7:20.560
（此句尚無繁中翻譯）
7:20.560–7:25.760
（此句尚無繁中翻譯）
7:25.760–7:31.200
（此句尚無繁中翻譯）
7:31.200–7:35.360
（此句尚無繁中翻譯）
7:35.360–7:40.320
（此句尚無繁中翻譯）
7:40.320–7:45.040
（此句尚無繁中翻譯）
7:45.040–7:50.080
（此句尚無繁中翻譯）
7:50.640–7:56.240
（此句尚無繁中翻譯）
7:56.240–8:01.120
（此句尚無繁中翻譯）
8:01.120–8:06.400
（此句尚無繁中翻譯）
8:06.400–8:12.560
（此句尚無繁中翻譯）
8:12.560–8:18.000
（此句尚無繁中翻譯）
8:18.000–8:22.880
（此句尚無繁中翻譯）
8:22.880–8:26.560
（此句尚無繁中翻譯）
8:26.560–8:30.560
（此句尚無繁中翻譯）
8:30.560–8:36.160
（此句尚無繁中翻譯）
8:36.160–8:40.160
（此句尚無繁中翻譯）
8:40.160–8:45.600
（此句尚無繁中翻譯）
8:45.600–8:50.480
（此句尚無繁中翻譯）
8:50.480–8:54.640
（此句尚無繁中翻譯）
8:54.640–8:58.720
（此句尚無繁中翻譯）
8:58.720–9:03.760
（此句尚無繁中翻譯）
9:03.760–9:07.840
（此句尚無繁中翻譯）
9:07.840–9:12.560
（此句尚無繁中翻譯）
9:12.560–9:16.640
（此句尚無繁中翻譯）
9:16.640–9:21.200
（此句尚無繁中翻譯）
9:22.400–9:28.640
（此句尚無繁中翻譯）
9:28.640–9:33.920
（此句尚無繁中翻譯）
9:33.920–9:38.800
（此句尚無繁中翻譯）
9:38.800–9:43.840
（此句尚無繁中翻譯）
9:43.840–9:47.760
（此句尚無繁中翻譯）
9:47.760–9:53.360
（此句尚無繁中翻譯）
9:53.360–9:58.080
（此句尚無繁中翻譯）
9:58.080–10:03.840
（此句尚無繁中翻譯）
10:03.840–10:09.600
（此句尚無繁中翻譯）
10:09.600–10:13.840
（此句尚無繁中翻譯）
10:13.840–10:19.520
（此句尚無繁中翻譯）
10:19.520–10:24.800
（此句尚無繁中翻譯）
10:24.800–10:30.160
（此句尚無繁中翻譯）
10:30.160–10:34.480
（此句尚無繁中翻譯）
10:34.480–10:39.920
（此句尚無繁中翻譯）
10:39.920–10:44.640
（此句尚無繁中翻譯）
10:44.640–10:49.120
（此句尚無繁中翻譯）
10:49.120–10:54.480
（此句尚無繁中翻譯）
10:54.480–10:59.360
（此句尚無繁中翻譯）
10:59.360–11:04.080
（此句尚無繁中翻譯）
11:04.080–11:08.400
（此句尚無繁中翻譯）
11:08.400–11:13.360
（此句尚無繁中翻譯）
11:13.920–11:18.560
（此句尚無繁中翻譯）
11:18.560–11:22.800
（此句尚無繁中翻譯）
11:22.800–11:27.120
（此句尚無繁中翻譯）
11:27.120–11:33.680
（此句尚無繁中翻譯）
11:33.680–11:38.800
（此句尚無繁中翻譯）
11:38.800–11:43.200
（此句尚無繁中翻譯）
11:43.200–11:48.000
（此句尚無繁中翻譯）
11:48.000–11:52.640
（此句尚無繁中翻譯）
11:52.640–11:58.000
（此句尚無繁中翻譯）
11:58.000–12:02.960
（此句尚無繁中翻譯）
12:02.960–12:07.920
（此句尚無繁中翻譯）
12:07.920–12:14.000
（此句尚無繁中翻譯）
12:14.000–12:18.160
（此句尚無繁中翻譯）
12:18.160–12:23.280
（此句尚無繁中翻譯）
12:23.280–12:27.200
（此句尚無繁中翻譯）
12:27.200–12:31.520
（此句尚無繁中翻譯）
12:31.520–12:36.080
（此句尚無繁中翻譯）
12:36.080–12:40.800
（此句尚無繁中翻譯）
12:40.800–12:45.200
（此句尚無繁中翻譯）
12:45.200–12:49.600
（此句尚無繁中翻譯）
12:49.600–12:53.600
（此句尚無繁中翻譯）
12:53.600–13:00.160
（此句尚無繁中翻譯）
13:00.160–13:06.720
（此句尚無繁中翻譯）
13:06.720–13:10.880
（此句尚無繁中翻譯）
13:10.880–13:16.080
（此句尚無繁中翻譯）
13:16.080–13:20.000
（此句尚無繁中翻譯）
13:20.000–13:26.880
（此句尚無繁中翻譯）
13:26.880–13:31.920
（此句尚無繁中翻譯）
13:31.920–13:37.280
（此句尚無繁中翻譯）
13:37.280–13:42.000
（此句尚無繁中翻譯）
13:42.000–13:48.720
（此句尚無繁中翻譯）
13:48.720–13:56.560
（此句尚無繁中翻譯）
13:56.560–14:02.000
（此句尚無繁中翻譯）
14:02.000–14:06.640
（此句尚無繁中翻譯）
14:06.640–14:11.760
（此句尚無繁中翻譯）
14:11.760–14:17.840
（此句尚無繁中翻譯）
14:17.840–14:21.440
（此句尚無繁中翻譯）
14:21.440–14:27.600
（此句尚無繁中翻譯）
14:27.600–14:32.560
（此句尚無繁中翻譯）
14:32.560–14:37.040
（此句尚無繁中翻譯）
14:37.040–14:41.760
（此句尚無繁中翻譯）
14:41.760–14:46.400
（此句尚無繁中翻譯）
14:46.400–14:51.040
（此句尚無繁中翻譯）
14:51.040–14:56.560
（此句尚無繁中翻譯）
14:56.560–15:01.920
（此句尚無繁中翻譯）
15:01.920–15:06.000
（此句尚無繁中翻譯）
15:06.000–15:10.480
（此句尚無繁中翻譯）
15:10.480–15:15.600
（此句尚無繁中翻譯）
15:15.600–15:20.800
（此句尚無繁中翻譯）
15:20.800–15:25.760
（此句尚無繁中翻譯）
15:25.760–15:30.160
（此句尚無繁中翻譯）
15:30.160–15:34.800
（此句尚無繁中翻譯）
15:34.800–15:39.520
（此句尚無繁中翻譯）
15:39.520–15:44.000
（此句尚無繁中翻譯）
15:44.000–15:48.320
（此句尚無繁中翻譯）
15:48.320–15:52.320
（此句尚無繁中翻譯）
15:52.320–15:55.600
（此句尚無繁中翻譯）
15:55.600–15:59.920
（此句尚無繁中翻譯）
15:59.920–16:04.400
（此句尚無繁中翻譯）
16:04.400–16:08.960
（此句尚無繁中翻譯）
16:08.960–16:14.000
（此句尚無繁中翻譯）
16:14.000–16:19.680
（此句尚無繁中翻譯）
16:19.680–16:23.680
（此句尚無繁中翻譯）
16:23.680–16:29.040
（此句尚無繁中翻譯）
16:29.040–16:33.840
（此句尚無繁中翻譯）
16:33.840–16:39.360
（此句尚無繁中翻譯）
16:39.360–16:44.960
（此句尚無繁中翻譯）
16:44.960–16:50.160
（此句尚無繁中翻譯）
16:50.160–16:54.240
（此句尚無繁中翻譯）
16:54.240–16:58.880
（此句尚無繁中翻譯）
16:58.880–17:03.280
（此句尚無繁中翻譯）
17:03.280–17:08.800
（此句尚無繁中翻譯）
17:08.800–17:12.960
（此句尚無繁中翻譯）
17:12.960–17:18.320
（此句尚無繁中翻譯）
17:18.320–17:22.800
（此句尚無繁中翻譯）
17:22.800–17:28.480
（此句尚無繁中翻譯）
17:28.480–17:34.320
（此句尚無繁中翻譯）
17:34.320–17:39.520
（此句尚無繁中翻譯）
17:39.520–17:44.800
（此句尚無繁中翻譯）
17:44.800–17:50.480
（此句尚無繁中翻譯）
17:50.480–17:56.800
（此句尚無繁中翻譯）
17:56.800–18:01.920
（此句尚無繁中翻譯）
18:01.920–18:08.240
（此句尚無繁中翻譯）
18:08.240–18:13.440
（此句尚無繁中翻譯）
18:13.440–18:18.720
（此句尚無繁中翻譯）
18:18.720–18:23.440
（此句尚無繁中翻譯）
18:23.440–18:27.600
（此句尚無繁中翻譯）
18:27.600–18:32.800
（此句尚無繁中翻譯）
18:32.800–18:37.760
（此句尚無繁中翻譯）
18:37.760–18:44.080
（此句尚無繁中翻譯）
18:44.080–18:47.840
（此句尚無繁中翻譯）
18:47.840–18:52.480
（此句尚無繁中翻譯）
18:52.480–18:57.280
（此句尚無繁中翻譯）
18:57.280–19:02.480
（此句尚無繁中翻譯）
19:02.480–19:08.640
（此句尚無繁中翻譯）
19:08.640–19:13.360
（此句尚無繁中翻譯）
19:13.360–19:18.960
（此句尚無繁中翻譯）
19:18.960–19:23.600
（此句尚無繁中翻譯）
19:23.600–19:28.720
（此句尚無繁中翻譯）
19:28.720–19:32.960
（此句尚無繁中翻譯）
19:32.960–19:38.080
（此句尚無繁中翻譯）
19:38.080–19:42.560
（此句尚無繁中翻譯）
19:42.560–19:49.360
（此句尚無繁中翻譯）
19:49.360–19:54.480
（此句尚無繁中翻譯）
19:54.480–19:59.600
（此句尚無繁中翻譯）
19:59.600–20:04.000
（此句尚無繁中翻譯）
20:04.000–20:09.680
（此句尚無繁中翻譯）
20:09.680–20:14.400
（此句尚無繁中翻譯）
20:14.400–20:19.360
（此句尚無繁中翻譯）
20:19.360–20:24.000
（此句尚無繁中翻譯）
20:24.000–20:28.960
（此句尚無繁中翻譯）
20:28.960–20:33.040
（此句尚無繁中翻譯）
20:33.040–20:38.240
（此句尚無繁中翻譯）
20:38.240–20:42.960
（此句尚無繁中翻譯）
20:42.960–20:48.000
（此句尚無繁中翻譯）
20:48.000–20:52.320
（此句尚無繁中翻譯）
20:52.320–20:56.320
（此句尚無繁中翻譯）
20:56.320–21:00.720
（此句尚無繁中翻譯）
21:00.720–21:04.960
（此句尚無繁中翻譯）
21:04.960–21:10.080
（此句尚無繁中翻譯）
21:10.080–21:14.560
（此句尚無繁中翻譯）
21:14.560–21:19.040
（此句尚無繁中翻譯）
21:19.040–21:24.720
（此句尚無繁中翻譯）
21:24.720–21:29.360
（此句尚無繁中翻譯）
21:29.360–21:34.560
（此句尚無繁中翻譯）
21:34.560–21:39.600
（此句尚無繁中翻譯）
21:39.600–21:46.160
（此句尚無繁中翻譯）
21:46.160–21:50.720
（此句尚無繁中翻譯）
21:50.720–21:55.600
（此句尚無繁中翻譯）
21:55.600–22:00.480
（此句尚無繁中翻譯）
22:00.480–22:04.640
（此句尚無繁中翻譯）
22:04.640–22:10.240
（此句尚無繁中翻譯）
22:10.240–22:14.480
（此句尚無繁中翻譯）
22:14.480–22:19.280
（此句尚無繁中翻譯）
22:19.280–22:24.400
（此句尚無繁中翻譯）
22:24.400–22:28.720
（此句尚無繁中翻譯）
22:28.720–22:33.600
（此句尚無繁中翻譯）
22:33.600–22:38.880
（此句尚無繁中翻譯）
22:38.880–22:44.000
（此句尚無繁中翻譯）
22:44.000–22:48.800
（此句尚無繁中翻譯）
22:48.800–22:53.600
（此句尚無繁中翻譯）
22:53.600–22:57.440
（此句尚無繁中翻譯）
22:57.440–23:05.000
（此句尚無繁中翻譯）
0:00.560–0:07.120
today on the ai daily brief why clawed tag and approaches like it might change how you use ai
（此句尚無繁中翻譯）
0:08.240–0:13.440
before that in the headlines an anthropic customer has sued the us over losing fable
（此句尚無繁中翻譯）
0:13.440–0:18.400
but can it possibly help the ai daily brief is a daily podcast and video about the most
（此句尚無繁中翻譯）
0:18.400–0:24.080
important news and discussions in ai all right friends quick announcements before we dive in
（此句尚無繁中翻譯）
0:24.080–0:28.800
i am continuing to remind folks about the executive agent leadership program which is the
（此句尚無繁中翻譯）
0:28.800–0:33.680
enterprise grade descendant of enterprise claw now offered in conjunction with super intelligent
（此句尚無繁中翻譯）
0:33.680–0:39.920
you can find out about that at training.bsuper.ai the next cohort begins on monday june 29th so if
（此句尚無繁中翻譯）
0:39.920–0:46.480
you're interested in that check it out well here we are heading towards the back half of this week
（此句尚無繁中翻譯）
0:46.480–0:54.240
with a return of fable 5 no closer insight in fact the rumor mill suggests that we are not in line for
（此句尚無繁中翻譯）
0:54.240–0:59.520
pretty much any models for some time later this week i'm going to do a show about what you can
（此句尚無繁中翻譯）
0:59.520–1:04.880
do to make the most of this forced ai model pause but for now let's look at some of the conversations
（此句尚無繁中翻譯）
1:04.880–1:09.920
that are actually shaping the discourse now first up i flagged this before but i want to take this
（此句尚無繁中翻譯）
1:09.920–1:15.120
chance to flag it again right now emotions are running at an 11 when it comes to this anthropic news
（此句尚無繁中翻譯）
1:15.760–1:21.200
i saw a post go viral on x about the significance of this bipartisan letter asking our lutnik if
（此句尚無繁中翻譯）
1:21.200–1:25.440
anthropic had been singled out but friends this is at this point about a week old and these sort of
（此句尚無繁中翻譯）
1:25.440–1:30.480
open congressional letters don't actually do anything they don't usually get responses this is
（此句尚無繁中翻譯）
1:30.480–1:35.280
in a word a nothing burger one thing that also probably amounts to nothing but is at least a little
（此句尚無繁中翻譯）
1:35.280–1:40.240
bit more interesting is that one anthropic customer is taking it upon themselves to get the model
（此句尚無繁中翻譯）
1:40.240–1:46.080
reinstated by suing the government in a lawsuit filed on tuesday a legal tech firm called legion
（此句尚無繁中翻譯）
1:46.080–1:50.640
claimed the ban was illegal and damaging their business the harm to legion is immediate
（此句尚無繁中翻譯）
1:50.640–1:55.600
irreparable and existential the lawsuit stated the pace of frontier ai advancement is blistering and
（此句尚無繁中翻譯）
1:55.600–2:01.280
competitive ground lost during a suspension cannot be regained after the fact certainly regardless of
（此句尚無繁中翻譯）
2:01.280–2:05.600
the ultimate success of this lawsuit i think that is a sentiment that many of you listening now will
（此句尚無繁中翻譯）
2:05.600–2:10.720
agree with the filing claimed that export control laws do not cover shutting down access to cloud
（此句尚無繁中翻譯）
2:10.720–2:15.440
software a point that many other legal commentators made as soon as the ban was announced serving
（此句尚無繁中翻譯）
2:15.440–2:20.480
computer code or text outputs from a computer isn't typically seen as an export that can be proper
（此句尚無繁中翻譯）
2:20.480–2:26.720
subject of export controls in the alternative if the ban relied on ieepa powers similar to sanctions
（此句尚無繁中翻譯）
2:26.720–2:31.200
the lawsuit notes that these powers do not apply to information and no prerequisite national security
（此句尚無繁中翻譯）
2:31.200–2:36.080
emergency was declared further the lawsuit argues the move was overbroad and possibly retaliatory
（此句尚無繁中翻譯）
2:36.080–2:40.560
towards anthropic it also notes the decision to block distribution of a model directly contradicts
（此句尚無繁中翻譯）
2:40.560–2:44.800
the executive order sign at the beginning of june which ruled out the idea of a model licensing
（此句尚無繁中翻譯）
2:44.800–2:49.280
scheme in making their argument legion centers the facts on their canadian development team losing
（此句尚無繁中翻譯）
2:49.280–2:53.280
access to the model pointing out that targeting allied nations with export controls is highly
（此句尚無繁中翻譯）
2:53.280–2:57.440
unusual anthropic for their part remains silent on the matter and there is certainly no indication
（此句尚無繁中翻譯）
2:57.440–3:01.600
this lawsuit is being filed on their behalf by a third party in comments to the press they referred
（此句尚無繁中翻譯）
3:01.600–3:05.280
to previous statements indicating they're grateful to the administration for working quickly to resolve
（此句尚無繁中翻譯）
3:05.280–3:10.560
the issue now in terms of the potential success of this at this point most legal analysts that i've
（此句尚無繁中翻譯）
3:10.560–3:14.160
seen seem to think that the administration has gone outside the bounds of the law on this one
（此句尚無繁中翻譯）
3:14.160–3:18.640
but also that it's unlikely to be a clear-cut decision for a judge and frankly none of that is going to
（此句尚無繁中翻譯）
3:18.640–3:23.200
actually matter because as long as anthropic is taking it seriously which they clearly are presumably
（此句尚無繁中翻譯）
3:23.200–3:27.840
and oh my gosh hopefully it'll be resolved through negotiation before any legal process could actually
（此句尚無繁中翻譯）
3:27.840–3:33.040
make it through the courts now the administration's friskiness when it comes to controlling the flow
（此句尚無繁中翻譯）
3:33.040–3:37.600
of ai is doing nothing but increasing the most recent news is that the administration is now
（此句尚無繁中翻譯）
3:37.600–3:41.840
pressuring meta to agree to voluntary model review sources told the new york times that the white
（此句尚無繁中翻譯）
3:41.840–3:46.480
house is pressing meta to submit their ai models for safety and performance testing prior to release
（此句尚無繁中翻譯）
3:46.480–3:51.520
meta is currently the lone holdout among major ai labs after microsoft xai and google signed
（此句尚無繁中翻譯）
3:51.520–3:55.760
agreements shortly before the recent executive order on ai and cyber security now reportedly the
（此句尚無繁中翻譯）
3:55.760–3:59.440
administration's requests seem to only relate to testing at the center for ai standards and
（此句尚無繁中翻譯）
3:59.440–4:03.520
innovation at the commerce department not whatever additional security testing has been going on at
（此句尚無繁中翻譯）
4:03.520–4:08.320
the nsa and meta does seem to be leaning towards an agreement with a spokesperson commenting
（此句尚無繁中翻譯）
4:08.320–4:12.960
we share the administration's goal of advancing us leadership on robust and secure frontier ai
（此句尚無繁中翻譯）
4:12.960–4:17.680
while we are working through the details we hope to sign the agreement soon still the whole idea of
（此句尚無繁中翻譯）
4:17.680–4:22.000
pressuring a company to submit their technology for government testing really highlights just how
（此句尚無繁中翻譯）
4:22.000–4:28.240
air quotes voluntary the program actually is and having contained advanced ai models the commerce
（此句尚無繁中翻譯）
4:28.240–4:32.880
department is reportedly turning to robots as their next technology to ban politico reports that
（此句尚無繁中翻譯）
4:32.880–4:37.280
commerce secretary howard lutnik held a closed door meeting on monday to discuss the risk of chinese-made
（此句尚無繁中翻譯）
4:37.280–4:40.480
models according to notes provided to politico lutnik said
（此句尚無繁中翻譯）
4:41.200–4:46.160
we don't want state subsidized robotics attacking us in america this is the arms race that is coming
（此句尚無繁中翻譯）
4:46.160–4:49.920
robotic arms are coming we need to make sure they're produced in america so we're going to study
（此句尚無繁中翻譯）
4:49.920–4:54.320
those right now commerce department officials reportedly view state-backed chinese robot companies
（此句尚無繁中翻譯）
4:54.320–4:59.120
as a massive threat fearing they could proliferate through global markets before u.s alternatives hit scale
（此句尚無繁中翻譯）
5:00.000–5:03.920
chinese robots are already subject to steep tariffs but the commerce department is considering
（此句尚無繁中翻譯）
5:03.920–5:08.320
additional action the monday meeting included more than a dozen executives from major firms including
（此句尚無繁中翻譯）
5:08.320–5:13.440
boston dynamics spacex siemens and goldman sachs the major topic of the meeting was how the u.s can
（此句尚無繁中翻譯）
5:13.440–5:18.240
best rebuild its industrial base but apparently the risks of chinese robots dominated the discussion
（此句尚無繁中翻譯）
5:18.240–5:22.400
one source said the whole idea that what we're going to end up with is an american brain with a
（此句尚無繁中翻譯）
5:22.400–5:28.400
chinese body is a very very bad strategic plan now it's unclear from the reporting how fast we should
（此句尚無繁中翻譯）
5:28.400–5:34.720
expect to crack down but the sabers are already rattling elsewhere in washington on monday the x account for
（此句尚無繁中翻譯）
5:34.720–5:39.280
the congressional select committee on china posted a screenshot of a unitary humanoid robot for sale
（此句尚無繁中翻譯）
5:39.280–5:43.680
through amazon it's priced at seventeen thousand nine hundred ninety dollars and can ship to washington
（此句尚無繁中翻譯）
5:43.680–5:49.200
in a month the select committee added unitary was recently designated as a chinese military company
（此句尚無繁中翻譯）
5:49.200–5:53.680
and its products are a threat to our national security yet here is amazon selling a unitary robot
（此句尚無繁中翻譯）
5:53.680–5:58.560
in america we need chairman mulnar's guard act to stop this threat and support american robotics
（此句尚無繁中翻譯）
5:58.560–6:02.800
now several chinese robotics firms have received the same designation which many see as a precursor to
（此句尚無繁中翻譯）
6:02.800–6:08.720
a huawei style ban now on the topic of american-made robots elon musk recently said that they are in the
（此句尚無繁中翻譯）
6:08.720–6:13.520
final stages of completion of optimus 3 which he says is really going to be by far the most advanced
（此句尚無繁中翻譯）
6:13.520–6:18.720
robot in the world nothing's even close but people are i would say fairly skeptical of those claims
（此句尚無繁中翻譯）
6:19.600–6:26.160
speaking of elon though xai has followed suit to openai and anthropic introducing the goals primitive
（此句尚無繁中翻譯）
6:26.160–6:31.840
in an update to their coding harness grok build as with the goal command in codex this allows users to
（此句尚無繁中翻譯）
6:31.840–6:36.640
define an outcome and set grok to work on a long horizon task grok will continue working across
（此句尚無繁中翻譯）
6:36.640–6:40.960
multiple steps deploying sub-agents to complete individual tasks and orchestrating the entire
（此句尚無繁中翻譯）
6:40.960–6:45.360
process without user intervention the orchestration process includes planning implementation and
（此句尚無繁中翻譯）
6:45.360–6:50.560
verification to ensure a high quality output so a couple things to make this worthy of inclusion
（此句尚無繁中翻譯）
6:50.560–6:55.920
here first of all it reinforces the idea that slash goal is an actual new ai ux primitive not just a
（此句尚無繁中翻譯）
6:55.920–7:01.040
generic feature but second this is our first glimpse of what the xai cursor partnership will look like
（此句尚無繁中翻譯）
7:01.040–7:05.760
after the acquisition was finalized last week the goal command natively uses a fine tune of grok
（此句尚無繁中翻譯）
7:05.760–7:11.920
called grok build 0.1 as well as cursor's composer 2.5 to complete its tasks and the combination has at
（此句尚無繁中翻譯）
7:11.920–7:16.320
least the potential to be more than the sum of its parts composer is known for cost-effective high
（此句尚無繁中翻譯）
7:16.320–7:20.560
performance making it a natural fit for coding sub-agents and while the latest version of the
（此句尚無繁中翻譯）
7:20.560–7:25.760
flagship grok model didn't grab that much attention on release one big point of differentiation it tried to
（此句尚無繁中翻譯）
7:25.760–7:31.200
make was a sophisticated orchestration process built into the model when used as a chatbot this mostly
（此句尚無繁中翻譯）
7:31.200–7:35.360
just resulted in sub-agents arguing over how to interpret information but that same process applied
（此句尚無繁中翻譯）
7:35.360–7:40.320
to coding tasks could make grok really good at that sort of orchestration we'll see what comes next and
（此句尚無繁中翻譯）
7:40.320–7:45.040
whether this makes any waves but certainly the update to build reinforces the idea that xai at this
（此句尚無繁中翻譯）
7:45.040–7:50.080
point as much as spacex is becoming a neocloud isn't quite done releasing ai products just yet
（此句尚無繁中翻譯）
7:50.640–7:56.240
speaking of ai products one that is getting a lot of attention is byte dance's seed dance 2.5 we've
（此句尚無繁中翻譯）
7:56.240–8:01.120
started to get a ton of previews of the new model which doubles maximum clip length from 15 to 30 seconds
（此句尚無繁中翻譯）
8:01.120–8:06.400
and adds 4k resolution the new version also supports up to 50 input references to use as objects or
（此句尚無繁中翻譯）
8:06.400–8:12.560
characters within scenes by comparison seed dance 2.0 could handle 12 references and google's vo 3.1
（此句尚無繁中翻譯）
8:12.560–8:18.000
released back in january maxed out at 3. c dance 2.5 is also able to make use of image video and audio
（此句尚無繁中翻譯）
8:18.000–8:22.880
files as references the first time anyone has moved beyond images now so far it doesn't look
（此句尚無繁中翻譯）
8:22.880–8:26.560
like anyone has actually got their hands on the model so we're just going off of a demo clip
（此句尚無繁中翻譯）
8:26.560–8:30.560
provided by byte dance yet that's enough for people to suggest that this will be the mythos
（此句尚無繁中翻譯）
8:30.560–8:36.160
moment of video models simon smith who i'm fairly sure isn't getting paid by byte dance by the way
（此句尚無繁中翻譯）
8:36.160–8:40.160
that's a joke if you've heard simon smith quoted here he was very very much not getting paid by byte
（此句尚無繁中翻譯）
8:40.160–8:45.600
dance wondered how they managed to gain such a big lead in video he asked seed dance seemed to be
（此句尚無繁中翻譯）
8:45.600–8:50.480
pulling away in the video generation race is there a recursive feedback loop at work here if so what
（此句尚無繁中翻譯）
8:50.480–8:54.640
is it are they generating training data with each generation to train the next if so what do they do
（此句尚無繁中翻譯）
8:54.640–8:58.720
to increase quality with each iteration whatever it is the corpus of training content they have access
（此句尚無繁中翻譯）
8:58.720–9:03.760
to through tick tock seems to be paying off lastly today i don't want to get into it deeply but one
（此句尚無繁中翻譯）
9:03.760–9:07.840
story that i am keeping an eye on is some of the market moves that i had previously discussed with
（此句尚無繁中翻譯）
9:07.840–9:12.560
regard to google specifically moving to a broader ai sell-off to the extent that this becomes a more
（此句尚無繁中翻譯）
9:12.560–9:16.640
significant narrative trend as opposed to just a particular market moment we'll come back and
（此句尚無繁中翻譯）
9:16.640–9:21.200
explore it more for now though that's going to do it for today's headlines next up the main episode
（此句尚無繁中翻譯）
9:22.400–9:28.640
welcome back to the ai daily brief sometimes big stories in ai are extremely obvious for example the
（此句尚無繁中翻譯）
9:28.640–9:33.920
u.s government sending a note telling anthropic that they're using export controls to deny access to the
（此句尚無繁中翻譯）
9:33.920–9:38.800
latest model to everyone who's not american and anthropic responding by taking down the model that they
（此句尚無繁中翻譯）
9:38.800–9:43.840
had released just a few days ago for everyone that's a big story in a very obvious way today's
（此句尚無繁中翻譯）
9:43.840–9:47.760
story is big not because of some geopolitical implications but because of the way that it
（此句尚無繁中翻譯）
9:47.760–9:53.360
potentially changes at somewhat of a core level how we think about interacting with ai at work now
（此句尚無繁中翻譯）
9:53.360–9:58.080
the feature we're discussing is called claude tag which claude announced on tuesday as a new way for
（此句尚無繁中翻譯）
9:58.080–10:03.840
teams to work with claude and to put it really simply it's basically claude in slack but not just as a
（此句尚無繁中翻譯）
10:03.840–10:09.600
different channel to prompt claude but instead as a full team member that has access to all of the
（此句尚無繁中翻譯）
10:09.600–10:13.840
context from all the different channels as well as the different tools that your team is already using
（此句尚無繁中翻譯）
10:13.840–10:19.520
anthropic writes claude tag is an evolution of claude code made more proactive and built to work with a
（此句尚無繁中翻譯）
10:19.520–10:24.800
full team it's now one of the main ways we get things done at anthropic 65 of our product team's code
（此句尚無繁中翻譯）
10:24.800–10:30.160
now comes from our internal version tag claude with a request and it breaks the task into stages works
（此句尚無繁中翻譯）
10:30.160–10:34.480
through them with the tools it has access to and responds in the thread with what it creates it'll
（此句尚無繁中翻譯）
10:34.480–10:39.920
write or merge pull requests run data analysis or help resolve an incident in a channel that's a
（此句尚無繁中翻譯）
10:39.920–10:44.640
channel on slack there's one claude that interacts with everyone so a teammate can pick up exactly where
（此句尚無繁中翻譯）
10:44.640–10:49.120
you left off it builds more context about the work as it follows the channel so you don't need to
（此句尚無繁中翻譯）
10:49.120–10:54.480
explain things from scratch turn ambient behavior on and claude takes initiative it follows up on threads
（此句尚無繁中翻譯）
10:54.480–10:59.360
that have gone quiet and flags what's relevant from across its channels and tools so for someone who's
（此句尚無繁中翻譯）
10:59.360–11:04.080
moving fast you could be forgiven for thinking it's just another agent in slack sort of announcement
（此句尚無繁中翻譯）
11:04.080–11:08.400
of which we've had a fair few but just based on the way that the anthropic team was talking about
（此句尚無繁中翻譯）
11:08.400–11:13.360
it there's clearly something going on here anthropic team member alexander bricken writes
（此句尚無繁中翻譯）
11:13.920–11:18.560
claude tag has completely changed the way we do work at anthropic having the multiplayer experience
（此句尚無繁中翻譯）
11:18.560–11:22.800
of claude within slack with all of the fundamental primitives of claude code like proactivity
（此句尚無繁中翻譯）
11:22.800–11:27.120
subagents and long horizon task management is an entirely new way to do work asynchronously
（此句尚無繁中翻譯）
11:27.120–11:33.680
tobin south writes claude tag is how i do 90 plus of my work i even had people from anthropic texting
（此句尚無繁中翻譯）
11:33.680–11:38.800
me specifically when i didn't cover it on yesterday's show basically arguing behind the scenes that this
（此句尚無繁中翻譯）
11:38.800–11:43.200
is a big deal now i had already begun planning this episode when those texts started coming in
（此句尚無繁中翻譯）
11:43.200–11:48.000
but still that should tell you how the team behind it views its significance now this is not the only
（此句尚無繁中翻譯）
11:48.000–11:52.640
feature of this type that's available now anthropic is not the first company to release something like
（此句尚無繁中翻譯）
11:52.640–11:58.000
this the semi-analysis team wrote perplexity's computer slack co-worker has been wildly useful
（此句尚無繁中翻譯）
11:58.000–12:02.960
internally at semi-analysis which is why anthropic launched a computing product the team is configuring
（此句尚無繁中翻譯）
12:02.960–12:07.920
claude tag ai co-workers and is excited to compare and contrast them with perplexity computer click health
（此句尚無繁中翻譯）
12:07.920–12:14.000
simon smith wrote immediately asked our cto to activate claude tag we already use chat gpt workspace
（此句尚無繁中翻譯）
12:14.000–12:18.160
agents in slack they're great for a bunch of use cases like providing support based on prior resolve
（此句尚無繁中翻譯）
12:18.160–12:23.280
slack posts and summarizing slack activity to create a weekly brief the difference with claude tag is
（此句尚無繁中翻譯）
12:23.280–12:27.200
that this level of configuration isn't even necessary you just drop claude into a slack
（此句尚無繁中翻譯）
12:27.200–12:31.520
channel and it picks up the context and becomes useful we currently create one slack channel per
（此句尚無繁中翻譯）
12:31.520–12:36.080
project we've historically also created chat gpt projects and notebook lm notebooks along with those
（此句尚無繁中翻譯）
12:36.080–12:40.800
slack channels to work with ai duplicating or triplicating context across all those surfaces
（此句尚無繁中翻譯）
12:40.800–12:45.200
now we can just create the slack channel as we've always done and drop claude in i also think this will
（此句尚無繁中翻譯）
12:45.200–12:49.600
massively help with ai diffusion people don't have to install claude desktop to access claude's most
（此句尚無繁中翻譯）
12:49.600–12:53.600
powerful capabilities in co-work and code they can just interact with claude as if it's another
（此句尚無繁中翻譯）
12:53.600–13:00.160
person in slack no less than andre carpathy argued that this is indeed a bigger deal than it seems at first
（此句尚無繁中翻譯）
13:00.160–13:06.720
he wrote this is a new paradigm for interacting with claude that is significantly more in line with
（此句尚無繁中翻譯）
13:06.720–13:10.880
all the other human activity org wide once you do all of the under the hood engineering work to make
（此句尚無繁中翻譯）
13:10.880–13:16.080
this just work eg across tools integrations compute environments memory security etc claude basically
（此句尚無繁中翻譯）
13:16.080–13:20.000
joins the team in a seamless way you can talk to it as you would talk to a person and it can help with
（此句尚無繁中翻譯）
13:20.000–13:26.880
a very large variety of workloads in my opinion this is the third major redesign of llm ui ux the first
（此句尚無繁中翻譯）
13:26.880–13:31.920
paradigm was that the llm is a website you go to the second was that it is an app you download on your
（此句尚無繁中翻譯）
13:31.920–13:37.280
computer the third one is that it is a self-contained persistent asynchronous entity with org wide tools and
（此句尚無繁中翻譯）
13:37.280–13:42.000
context working alongside teams of humans it really takes a while to wrap your head around it but it
（此句尚無繁中翻譯）
13:42.000–13:48.720
works and it is awesome now for a lot of folks the big eye blinking statistic was that anthropic is
（此句尚無繁中翻譯）
13:48.720–13:56.560
saying that 65 of their product team's code now comes from claude tag ejaz wrote it's blowing my mind
（此句尚無繁中翻譯）
13:56.560–14:02.000
that 65 of product code in anthropic is now written by tagging claude in group chats of staff discussing
（此句尚無繁中翻譯）
14:02.000–14:06.640
what they want to build r.i.p the days of tediously writing long product docs now you can literally go
（此句尚無繁中翻譯）
14:06.640–14:11.760
from slack to a production ready feature cloud code is barely a year old by the way developer nick
（此句尚無繁中翻譯）
14:11.760–14:17.840
dobos wrote anthropic isn't using cloud code anymore major agent ux shift alert and i think this is one
（此句尚無繁中翻譯）
14:17.840–14:21.440
of the sneaky things that's going to take people a little while to fully wrap their heads around
（此句尚無繁中翻譯）
14:21.440–14:27.600
while you might be tagging at claude what you're really calling upon is claude code now certainly
（此句尚無繁中翻譯）
14:27.600–14:32.560
both anthropic and openai have already spent lots of cycles trying to make the coding capabilities of
（此句尚無繁中翻譯）
14:32.560–14:37.040
their tools more accessible for non-coders but this is the biggest move in that direction yet
（此句尚無繁中翻譯）
14:37.040–14:41.760
when claude already exists in the workspace that you're operating in the barrier to using it to
（此句尚無繁中翻譯）
14:41.760–14:46.400
build something is having an idea and dropping it into the slack where you're already working
（此句尚無繁中翻譯）
14:46.400–14:51.040
now while it's early days for non-anthropic users many were quick to share what they had learned from
（此句尚無繁中翻譯）
14:51.040–14:56.560
using claude tag over the past few weeks or months claude code team member tarik pointed out that each claude in
（此句尚無繁中翻譯）
14:56.560–15:01.920
each channel is different one suggestion he had was to introduce air quotes claude to a new channel
（此句尚無繁中翻譯）
15:01.920–15:06.000
with a pin message and a set of instructions to remember like which persona to take when to respond
（此句尚無繁中翻譯）
15:06.000–15:10.480
etc he says you can think of it like it's claude.md because all of this will get added to its memory
（此句尚無繁中翻譯）
15:10.480–15:15.600
tarik also suggests having a personal channel eg tarik-claude where you can tag claude for your own
（此句尚無繁中翻譯）
15:15.600–15:20.800
work and give it instructions specific to you you can then forward other messages to that channel so your
（此句尚無繁中翻譯）
15:20.800–15:25.760
claude can get to work the way you like he also noted that this helps with overwhelm because when things
（此句尚無繁中翻譯）
15:25.760–15:30.160
get going it can be hard to keep track of all the threads he wrote in my personal channel i like to
（此句尚無繁中翻譯）
15:30.160–15:34.800
ask claude to keep a pin message that it updates with the status of everything he even uses a set
（此句尚無繁中翻譯）
15:34.800–15:39.520
of emojis to be able to help him understand where different work streams are in a single glance
（此句尚無繁中翻譯）
15:39.520–15:44.000
chris taylor from fractional who joined the broader anthropic ecosystem to run their version of a
（此句尚無繁中翻譯）
15:44.000–15:48.320
forward deployed engineering group described how they had been using it across their company he said
（此句尚無繁中翻譯）
15:48.320–15:52.320
it kind of works like a co-worker that just picks things up it owns busy work daily status
（此句尚無繁中翻譯）
15:52.320–15:55.600
summaries following up when people are out of office nudging teammates on open to
（此句尚無繁中翻譯）
15:55.600–15:59.920
dues pushing back due dates his plans shift chris said that when one of its own scheduled tasks
（此句尚無繁中翻譯）
15:59.920–16:04.400
started misfiring it caught the bug fixed it and wrote it up without being asked he also said that
（此句尚無繁中翻譯）
16:04.400–16:08.960
it manages itself we needed somewhere to track bugs and ideas for claude tag while we were testing it
（此句尚無繁中翻譯）
16:08.960–16:14.000
so it built that tracker for us and files every new issue and use case on its own finally reinforcing
（此句尚無繁中翻譯）
16:14.000–16:19.680
that this is in fact a fully fledged version of claude code not just claude he says it writes and ships
（此句尚無繁中翻譯）
16:19.680–16:23.680
handed one of our code bases claude read through it wrote up the 10 most important problems to fix
（此句尚無繁中翻譯）
16:23.680–16:29.040
stack ranked then made the changes and opened them for review all in slack the claude dev team shared
（此句尚無繁中翻譯）
16:29.040–16:33.840
that they're using it for incident response bug triage dependent work i.e work that's blocked on
（此句尚無繁中翻譯）
16:33.840–16:39.360
something else background watching and monitoring metrics nityesh from the every team wrote about
（此句尚無繁中翻譯）
16:39.360–16:44.960
why this will feel familiar to many he said claude tag has completely changed the way i do work for the
（此句尚無繁中翻譯）
16:44.960–16:50.160
last four months except it's not claude tag anthropic only announced that a few hours ago and i don't even
（此句尚無繁中翻譯）
16:50.160–16:54.240
have access to it but i did build a version of it for myself which i've been using for months now
（此句尚無繁中翻譯）
16:54.240–16:58.880
basically he said that after openclaw came online he built a harness that allowed him to turn any mac
（此句尚無繁中翻譯）
16:58.880–17:03.280
into an ai employee with the claude code headless mode he then ended up building three such employees
（此句尚無繁中翻譯）
17:03.280–17:08.800
and had them working in slack for the last few months now this idea of agents inside slack is also
（此句尚無繁中翻譯）
17:08.800–17:12.960
something that we've experimented a lot with over here as well one of the reasons that i burned through
（此句尚無繁中翻譯）
17:12.960–17:18.320
a billion tokens in march is that i was experimenting with a series of ai enablement agents that lived in
（此句尚無繁中翻譯）
17:18.320–17:22.800
slack and could do things like figuring out how we were using ai providing coaching on how to better
（此句尚無繁中翻譯）
17:22.800–17:28.480
use ai and developing overall ai strategy now you also might remember me talking about a month ago
（此句尚無繁中翻譯）
17:28.480–17:34.320
about how the team at every had shifted their agent strategy from originally having every employee build
（此句尚無繁中翻譯）
17:34.320–17:39.520
an ai agent version of themselves to instead having a group of agents that worked across employees based
（此句尚無繁中翻譯）
17:39.520–17:44.800
on function and goal and had shared context claude tag in many ways seems to be institutionalizing some
（此句尚無繁中翻譯）
17:44.800–17:50.480
of those shifts as best practice so trying to sum this all up let's talk about the five big shifts
（此句尚無繁中翻譯）
17:50.480–17:56.800
that i think claude tag could represent and keep in mind this is not just about claude tag as a feature
（此句尚無繁中翻譯）
17:56.800–18:01.920
but about it as a leading indicator of how labs and agent app companies are thinking about the future of
（此句尚無繁中翻譯）
18:01.920–18:08.240
ai at work the first big shift is from an app native interface to existing workplace interfaces obviously
（此句尚無繁中翻譯）
18:08.240–18:13.440
this breaks agents out of the controlled settings of the web app or the desktop apps that the labs have
（此句尚無繁中翻譯）
18:13.440–18:18.720
built and instead uses the existing spaces that people do work in now to some extent one has to
（此句尚無繁中翻譯）
18:18.720–18:23.440
think that this was inevitable as you try to shift behavior and get everyone working in a new way
（此句尚無繁中翻譯）
18:23.440–18:27.600
one of the biggest points of friction is the new interfaces that they have to go interact with
（此句尚無繁中翻譯）
18:27.600–18:32.800
the fact that the full power of claude code and everything that it can build is now accessible from
（此句尚無繁中翻譯）
18:32.800–18:37.760
the most default channel where people are working seems likely to me to make a big difference in terms
（此句尚無繁中翻譯）
18:37.760–18:44.080
of who accesses that power likewise this moves ai from a private chatbot to a shared teammate experience
（此句尚無繁中翻譯）
18:44.080–18:47.840
now of course there have been no shortage of sharing based features to break your personal
（此句尚無繁中翻譯）
18:47.840–18:52.480
chats out of their home in fact both openai and anthropic have released versions of better native
（此句尚無繁中翻譯）
18:52.480–18:57.280
websites slash web app builders in order to help make sharing more easily this of course takes that
（此句尚無繁中翻譯）
18:57.280–19:02.480
to the next level which also leads to our third shift which is from single user context i.e what the
（此句尚無繁中翻譯）
19:02.480–19:08.640
ai or agent knows about you to the full team context part of the power of moving the agent from its native
（此句尚無繁中翻譯）
19:08.640–19:13.360
home to your team's native home is that it gets access to the ambient context that exists all around
（此句尚無繁中翻譯）
19:13.360–19:18.960
it without you having to create an entirely different way of accessing all that context now already
（此句尚無繁中翻譯）
19:18.960–19:23.600
claude tag also reinforces a shift which was already very much happening which is the shift from
（此句尚無繁中翻譯）
19:23.600–19:28.720
prompting to delegation as the main mode of interacting with ai already throughout the year as agent
（此句尚無繁中翻譯）
19:28.720–19:32.960
the capacity has come online people have been slowly but surely breaking out of the way that they used to
（此句尚無繁中翻譯）
19:32.960–19:38.080
interact with these tools a big part of that has been in shifting from telling the ai what to do to
（此句尚無繁中翻譯）
19:38.080–19:42.560
telling it what you are trying to accomplish and giving it more freedom and latitude to accomplish an
（此句尚無繁中翻譯）
19:42.560–19:49.360
ever-expanding scale of goals lastly claude tag and more broadly agents that work in your team's workspaces
（此句尚無繁中翻譯）
19:49.360–19:54.480
shift ai's importance from something that becomes personally essential to key workers to an actual
（此句尚無繁中翻譯）
19:54.480–19:59.600
organizational dependency a new way of working not just individually but across the entire team
（此句尚無繁中翻譯）
19:59.600–20:04.000
which is not to say that there aren't challenges first of all while yes this is theoretically just
（此句尚無繁中翻譯）
20:04.000–20:09.680
dropping claude into slack there are a lot of questions around access and tool configuration that
（此句尚無繁中翻譯）
20:09.680–20:14.400
make it set up potentially much more complicated than it at first seems anthropic released an entire
（此句尚無繁中翻譯）
20:14.400–20:19.360
blog post about best practices for setting up agent identity in this new claude tag paradigm
（此句尚無繁中翻譯）
20:19.360–20:24.000
they also pointed out some best practices around permissions which wouldn't necessarily be obvious at first
（此句尚無繁中翻譯）
20:24.000–20:28.960
glance gail wiener points out that while the positive side is the democratization of claude
（此句尚無繁中翻譯）
20:28.960–20:33.040
code capabilities to a broader subset of the organization the flip side of that is that something
（此句尚無繁中翻譯）
20:33.040–20:38.240
like claude tag potentially runs into existing trust barriers and skepticism that could create a challenge
（此句尚無繁中翻譯）
20:38.240–20:42.960
she describes a scenario where there are five people on a team with her as the ai power user two
（此句尚無繁中翻譯）
20:42.960–20:48.000
members of the team not trusting ai one team member thinking that the power user uses ai too much
（此句尚無繁中翻譯）
20:48.000–20:52.320
and another being indifferent gail writes i start tagging claude in the project chat and here's what
（此句尚無繁中翻譯）
20:52.320–20:56.320
happens in that room the moment claude is in the channel every message anyone types is being read by
（此句尚無繁中翻譯）
20:56.320–21:00.720
it it stops looking like a tool the team uses and starts looking like my tool that's now sitting in
（此句尚無繁中翻譯）
21:00.720–21:04.960
everyone's workspace i become the person who brought the surveillance device to the meeting even if that's
（此句尚無繁中翻譯）
21:04.960–21:10.080
not what it is the skeptics position gets validated every time claude produces something good see she's
（此句尚無繁中翻譯）
21:10.080–21:14.560
outsourcing her thinking and every time it produces something mediocre see this is what we were worried
（此句尚無繁中翻譯）
21:14.560–21:19.040
about you can't win that frame basically she's arguing that this creates a whole new challenge around the
（此句尚無繁中翻譯）
21:19.040–21:24.720
human layer that will have to be solved as well one specific thing that seems fairly challenging to me
（此句尚無繁中翻譯）
21:24.720–21:29.360
is that despite the fact that you just use at claude for every instance of where you want to call upon
（此句尚無繁中翻譯）
21:29.360–21:34.560
these capabilities different clauds and different channels have potentially access to different tools
（此句尚無繁中翻譯）
21:34.560–21:39.600
and different permissions and different contexts and that feels like it'll get confusing super super
（此句尚無繁中翻譯）
21:39.600–21:46.160
fast in fact in simon smith's first impressions post later on he did write claude being many clauds is a bit
（此句尚無繁中翻譯）
21:46.160–21:50.720
disorienting i've come to think of claude as my claude the one in the app that knows me and connects to
（此句尚無繁中翻譯）
21:50.720–21:55.600
things i've connected it to but there is no single claude in slack it's channel dependent and admin
（此句尚無繁中翻譯）
21:55.600–22:00.480
configured and none of the clauds are mine this seems to be a common feeling one of the first things i'm
（此句尚無繁中翻譯）
22:00.480–22:04.640
seeing is people tagging claude in channels to ask what it can do what it remembers how it maintains
（此句尚無繁中翻譯）
22:04.640–22:10.240
context in the channel indeed how this creates new challenges and opportunities for context is a big topic of
（此句尚無繁中翻譯）
22:10.240–22:14.480
conversation and for some in light of the fable challenge in light of everything that's happened with
（此句尚無繁中翻譯）
22:14.480–22:19.280
fable there's also increasingly good reasons to not want to outsource this sort of process to a single
（此句尚無繁中翻譯）
22:19.280–22:24.400
closed provider hugging faces head of product victor writes at hugging face we've been building our own
（此句尚無繁中翻譯）
22:24.400–22:28.720
agent that we use via slack honestly building your own is quite simple and you'll be happy you did
（此句尚無繁中翻譯）
22:28.720–22:33.600
any model you want including self-hosted if needed fully customizable to your stack and of course no
（此句尚無繁中翻譯）
22:33.600–22:38.880
lock-in and no waitlist and not overpriced look ultimately it is very easy to overestimate the
（此句尚無繁中翻譯）
22:38.880–22:44.000
significance of any one new feature and yet if for no other reason than these significant
（此句尚無繁中翻譯）
22:44.000–22:48.800
shifts that the anthropic team themselves are reporting it seems like claude tag is one paradigm
（此句尚無繁中翻譯）
22:48.800–22:53.600
shift that will at least be worth exploring i'll report back as we try it out and see what we learn
（此句尚無繁中翻譯）
22:53.600–22:57.440
for now though that's going to do it for today's ai daily brief appreciate you listening or watching
（此句尚無繁中翻譯）
22:57.440–23:05.000
as always and until next time peace
（此句尚無繁中翻譯）

影片筆記：5 Ways Claude Tag Could Change How You Use AI

一句話總結

影片深入探討 Anthropic 推出的 Claude Tag 功能，指出其透過將 Claude 整合至 Slack 等現有協作平台，實現了從「單一使用者提示」到「團隊級別委派」的範式轉移，並分析了此功能在內部採用、UI/UX 演進、信任挑戰及供應商鎖定風險等方面的影響。

核心重點

  1. Claude Tag 的核心定義：
  • 不僅是 Slack 中的聊天機器人，而是具備完整團隊成員資格的協作者。
  • 能存取所有頻道的上下文及團隊工具，具備主動性（Proactivity）、子代理（Sub-agents）及長程任務管理能力。
  • Anthropic 內部數據顯示，其產品團隊 65% 的程式碼現已來自內部版本的 Claude Tag。
  1. LLM UI/UX 的第三次重設計：
  • 第一次：網站（Website）。
  • 第二次：電腦應用程式（App）。
  • 第三次：具備組織級工具與上下文的、與人類團隊並行的自主、持久、非同步實體（Self-contained persistent asynchronous entity）。
  1. 五大範式轉移（5 Ways）：
  2. 從原生應用程式介面到現有工作場所介面：打破網頁/桌面應用環境，利用人們已習慣的空間（如 Slack）。
  3. 從私人聊天機器人到共享隊友體驗：將共享功能推向新高，成為團隊共同協作者。
  4. 從單一使用者上下文到完整團隊上下文：代理從原生環境移至團隊環境，無需創建全新方式即可存取周圍環境上下文（Ambient context）。
  5. 從提示（Prompting）到委派（Delegation）：互動模式從「告訴 AI 做什麼」轉向「告訴 AI 試圖完成什麼」，賦予更多自由與範圍。
  6. 從個人必要到關鍵基礎設施：AI 的重要性從僅對關鍵工作者個人必要，轉變為實際的組織級基礎設施。
  1. 潛在挑戰與風險：
  • 信任與監視感：員工感到被監視，無論表現好壞都會引發負面解讀（表現好被解讀為「外包思考」，表現平庸被解讀為「預期的平庸」）。
  • 上下文碎片化：不同頻道中的 Claude 擁有不同的權限、工具和上下文，導致使用者感到困惑，認為 Slack 中的 Claude 並非單一實體。
  • 供應商鎖定：部分開發者（如 Hugging Face 產品負責人 Victor）建議自建 Agent 以避免依賴單一閉源提供商。

詳細大綱

第一部分：頭條新聞與政策動態（背景資訊）

  • Anthropic 訴訟案：
  • 客戶 Legion 起訴美國政府，要求恢復對 Claude 模型的訪問權限。
  • 指控禁令非法且損害商業利益。
  • 法律分析認為政府可能越權（軟體輸出非出口管制範圍、IEEPA 權力適用性、違反行政命令），但預計最終會通過談判解決。
  • 兩黨國會議員信函被視為「Nothing Burger」（無實際效力）。
  • Meta 與白宮協議：
  • 白宮施壓 Meta 進行自願性的模型安全與性能測試。
  • Meta 成為最後一個簽署協議的主要 AI 實驗室。
  • 測試範圍為 AI 標準與創新中心（Commerce Dept），而非 NSA 的安全測試。
  • 中國機器人工業威脅：
  • 商務部長 Howard Lutnik 召開閉門會議，討論中國製造的機器人對美國工業基礎的威脅。
  • 參與企業包括 Boston Dynamics, SpaceX, Siemens, Goldman Sachs。
  • 擔憂「美國的大腦，中國的軀體」。
  • 國會議員呼籲通過法案限制中國軍事公司 Unitary 的機器人銷售（提及《Guard Act》）。
  • Elon Musk 稱 Optimus 3 即將完成，但市場持懷疑態度。
  • xAI 技術更新：
  • 推出 Grok Build 的「Goal Primitive」功能，結合 Cursor 進行長程程式碼任務自動化。
  • 流程：規劃、實施、驗證，無需用戶干預。
  • 技術組合：Grok Build 0.1 與 Cursor Composer 2.5 的結合。
  • ByteDance Seedance 2.5：
  • 發布影片生成模型預覽，支援 30 秒長度、4K 解析度及多模態參考輸入。
  • 被認為可能成為影片模型的轉捩點，探討是否形成遞迴反饋迴圈（Recursive Feedback Loop）。
  • 市場動態：
  • Google 股價下跌引發對更廣泛 AI 銷售減少的關注。

第二部分：Claude Tag 的核心功能與應用

  • 定義與定位：
  • 存取所有不同頻道的上下文及團隊已使用的工具。
  • Anthropic 內部稱其為「Claude Code 的進化版」，專為全團隊協作設計。
  • 工作流程：在 Slack 標記 Claude 並提出請求 -> 拆解任務為階段 -> 使用存取的工具執行 -> 在對話線程中回應結果。
  • 具體能力：撰寫或合併 Pull Request、執行數據分析、協助解決頻道內的事故。
  • 內部採用數據：
  • Anthropic 產品團隊 65% 的程式碼現來自內部版本的 Claude Tag。
  • 單一 Claude 與所有人互動，隊友可接續之前的工作進度。
  • 「環境行為（Ambient behavior）」開啟後，Claude 會主動跟進靜默線程並標記相關內容。
  • 專家與開發者反饋：
  • Alexander Bricken：完全改變了 Anthropic 的工作方式，是全新的非同步工作模式。
  • Tobin South：90% 以上的工作透過 Claude Tag 完成。
  • Andre Karpathy：這是與組織範圍內其他人類活動更一致的互動新範式，背後工程（工具整合、運算環境、記憶體、安全性）使 Claude 能無縫加入團隊。
  • Ejaz：對 65% 的產品程式碼由群組聊天中的 Claude Tag 撰寫感到震驚，認為產品文件撰寫時代已逝。
  • Nick Dobos：Anthropic 不再使用 Cloud Code，這是主要的代理使用者體驗（Agent UX）轉移。
  • Tarik (Claude Code 團隊成員)：
  • 每個頻道中的 Claude 各不相同。
  • 建議：使用「錨定訊息（Pin message）」和指示來介紹 Claude 到新頻道，類似 .claude.md 檔案。
  • 個人頻道建議：建立個人頻道（如 tarik-claude）給予特定指示，轉發訊息讓 Claude 按個人喜好工作。
  • 解決 overwhelm：在個人頻道中要求 Claude 更新狀態的錨定訊息，使用表情符號（Emojis）一目瞭然追蹤工作流。
  • Chris Taylor (Fractional)：
  • 負責 Anthropic 的向前部署工程組（Forward deployed engineering group）。
  • 功能：接手瑣碎工作、每日狀態摘要、跟進休假人員、催促隊友、推遲截止日期。
  • 自主性：當排程任務出錯時，自動偵測錯誤、修復並撰寫報告。
  • 自我管理：測試期間自動建立追蹤器，自動歸檔新問題與用例。
  • 程式碼能力：讀取程式碼庫、列出 10 個最重要問題、堆疊排名、進行更改並開啟審查，全程在 Slack 完成。
  • Nityesh (Every Team)：
  • 在 Anthropic 宣布前幾個月，基於 OpenClaw 和 Claude Code 的無頭模式（Headless mode），自行建構了類似系統，將 Mac 變成 AI 員工並在 Slack 工作。
  • 驗證了 Slack 內代理的概念。
  • 講者自身經驗：
  • 3 月消耗十億 token 用於實驗 Slack 內的 AI 啟用代理，提供教練指導與 AI 策略開發。
  • 提到 Every 團隊從「每個員工建構自己的 AI 代理」轉向「基於功能與目標的代理群組共享上下文」，Claude Tag 似乎將此最佳實踐制度化。

第三部分：組織性依賴、挑戰與結論

  • 技術與配置挑戰：
  • Anthropic 發布部落格文章，提供關於「Agent 身分」設定的最佳實踐。
  • 權限設定並非直觀，需要額外關注。
  • 信任障礙與社會心理挑戰：
  • 民主化的雙面刃：雖然擴大了能力範圍，但也觸發了既有信任障礙。
  • 場景模擬：
  • 團隊成員對 AI 的不信任、認為使用者過度依賴 AI、或持冷漠態度。
  • 監視感：一旦 Claude 進入頻道，所有訊息都被閱讀，工具屬性轉變為「我的工具」，被視為監視裝置。
  • 雙重驗證困境：
  • 表現好：被解讀為「外包思考」。
  • 表現平庸：被解讀為「預期的平庸」。
  • 結論：這在人為層面創造了新的挑戰。
  • 上下文與權限的碎片化：
  • 不同頻道中的 Claude 擁有不同的工具、權限和上下文。
  • 這種差異會迅速導致混亂。
  • 使用者感到困惑，因為 Slack 中的 Claude 不是單一實體，而是取決於頻道和管理員配置。
  • 常見行為：使用者標記 Claude 詢問其能力、記憶及上下文維持方式。
  • 自建 Agent 的替代方案：
  • 基於對單一閉源提供商外包流程的擔憂。
  • Hugging Face 產品負責人 Victor 的觀點：
  • 自建 Agent 通過 Slack 使用非常簡單。
  • 支援任何模型（包括自託管）。
  • 完全客製化，無供應商鎖定，無等待名單，價格合理。
  • 結論：
  • 儘管可能高估單一新功能的重要性，但 Anthropic 報告的顯著轉變表明，Claude Tag 是一個值得探索的範式轉移。
  • 後續將回報實際測試結果。

工具 / 模型 / 名詞整理

  • AI 模型/產品：
  • Claude / Claude Tag：Anthropic 推出的 AI 模型及其在 Slack 中的標記功能/協作代理。
  • Claude Code：被提及為 Claude Tag 的基礎或進化前身，具備主動性、子代理、長遠任務管理等功能。
  • Grok / Grok Build 0.1：xAI 推出的模型及編碼環境。
  • Goal Primitive / Goal Command：xAI 推出的功能，定義結果並設定長程任務。
  • Cursor (Composer 2.5)：編碼輔助工具，與 Grok Build 結合使用。
  • Optimus 3：Elon Musk 提到的機器人版本。
  • Seedance 2.5 / Seed Dance 2.0：ByteDance 發布的影片生成模型。
  • Perplexity Computer：Perplexity 推出的類似內部協作產品。
  • ChatGPT Workspace Agents：OpenAI 提供的 Slack 代理功能。
  • OpenClaw：Nityesh 用於建構自訂代理系統的基礎。
  • Codex：提及作為對比。
  • Fable 5：逐字稿中提到 "return of fable 5"，疑似錯字或內部代號。
  • Vo 3.1：逐字稿中提到 "google's vo 3.1"，疑似錯字。
  • Unitary：指稱一家中國軍事公司製造的機器人。
  • Cloud Code：被提及為 Anthropic 過去使用但現在可能不再主要使用的工具（或指代其他雲端代碼環境）。
  • Claude Desktop：被提及為存取 Claude 強大功能的方式之一。
  • Notebook LM：被提及為過去與 Slack 頻道並行使用的工具。
  • 公司/組織：
  • Anthropic：開發 Claude 的公司。
  • US Government / White House：美國政府與白宮。
  • Legion：提起訴訟的法律科技公司。
  • Meta：最後一個簽署協議的主要 AI 實驗室。
  • Microsoft：提及。
  • xAI：Elon Musk 的公司。
  • OpenAI：提及。
  • Commerce Department (US)：美國商務部。
  • NSA：美國國家安全局。
  • Politico：提及。
  • Boston Dynamics：參與閉門會議的企業。
  • SpaceX：參與閉門會議的企業。
  • Siemens：參與閉門會議的企業。
  • Goldman Sachs：參與閉門會議的企業。
  • Congressional Select Committee on China：國會議員委員會。
  • Amazon：提及。
  • ByteDance：發布 Seedance 的公司。
  • TikTok：提及。
  • Google：提及股價下跌。
  • Super Intelligent：提及於培訓課程的公司。
  • Semi-analysis：提及使用 Perplexity Computer 的團隊/公司。
  • Click Health：提及使用 ChatGPT Workspace Agents 的公司。
  • Hugging Face：公司名稱，其產品負責人 Victor 提及自建 Agent 的經驗。
  • 人物：
  • Howard Lutnik：美國商務部長（逐字稿中多次出現 "our lutnik" 和 "howard lutnik"）。
  • Elon Musk：提及 Optimus 3 與 xAI。
  • Simon Smith：撰寫「第一印象」文章的人，評論 Seedance 2.5 與 ChatGPT Workspace Agents。
  • Mulnar：Chairman，提及 Guard Act。
  • Alexander Bricken：Anthropic 內部人員。
  • Tobin South：Anthropic 內部人員。
  • Andre Karpathy：專家，評論 LLM UI/UX 重設計。
  • Ejaz：開發者，評論產品程式碼由 Claude Tag 撰寫。
  • Nick Dobos：開發者，評論 Anthropic 不再使用 Cloud Code。
  • Tarik：Claude Code 團隊成員。
  • Chris Taylor：Fractional，負責 Anthropic 的向前部署工程組。
  • Nityesh：Every Team 成員。
  • Victor：Hugging Face 產品負責人。
  • Gail Wiener：提出信任障礙場景的評論者。

操作流程整理

Claude Tag 在 Anthropic 內部的典型工作流程

  1. 標記與請求：在 Slack 頻道中標記（Tag）Claude 並提出請求。
  2. 任務拆解：Claude 將任務拆解為多個階段。
  3. 工具執行：使用存取的工具（如程式碼庫、數據分析工具）執行任務。
  4. 結果回應：在對話線程中回應結果，包括撰寫或合併 Pull Request、執行數據分析、協助解決事故。
  5. 環境行為（Ambient Behavior）：若開啟此功能，Claude 會主動跟進靜默線程並標記相關內容。
  6. 協作接續：單一 Claude 與所有人互動，隊友可接續之前的工作進度，無需從頭解釋。

開發者建議的最佳實踐（Tarik 觀點）

  1. 初始化頻道：使用「錨定訊息（Pin message）」和指示來介紹 Claude 到新頻道，類似 .claude.md 檔案，將這些加入記憶體。
  2. 個人頻道設定：建立個人頻道（如 tarik-claude）給予特定指示，轉發訊息讓 Claude 按個人喜好工作。
  3. 狀態追蹤：在個人頻道中要求 Claude 更新狀態的錨定訊息，使用表情符號（Emojis）一目瞭然追蹤工作流，以解決 overwhelm 問題。

自建 Agent 的替代方案（Hugging Face Victor 觀點）

  1. 選擇模型：支援任何模型，包括自託管模型。
  2. 配置 Slack：通過 Slack 使用自建 Agent，過程簡單。
  3. 客製化：完全客製化，無供應商鎖定，無等待名單，價格合理。

值得注意的限制或風險

  1. 信任與隱私障礙：
  • 當 Claude 被標記在專案聊天室中時，員工會感到被監視，導致對 AI 的懷疑論調被驗證。
  • 雙重驗證困境：表現好被解讀為「外包思考」，表現平庸被解讀為「預期的平庸」。
  1. 上下文與權限碎片化：
  • 不同頻道中的 Claude 擁有不同的工具、權限和上下文。
  • 這種差異會迅速導致混亂，使用者感到困惑，因為 Slack 中的 Claude 不是單一實體，而是取決於頻道和管理員配置。

    • 3.

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