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
YouTube 影片框會固定在左上方;點擊右側逐字稿時間戳可跳到對應時間。
一句話總結
OpenAI 發布 GPT-5.6 系列模型(包含 Sol、Terra、Luna 三種規格),並推出 ChatGPT Work 企業版與全新桌面應用程式(Desktop App)。影片展示了模型在直接操作電腦(Computer Use)、生成互動式網站(Sites)、自動化工作流及農業管理等場景的應用,並強調了 Ultra Mode 的多代理協作能力與安全對齊措施。
核心重點
- GPT-5.6 系列模型發布:
- Sol:最強大模型,針對最難的代理工作流(agentic workflows)。
- Terra:較快模型,針對日常工作流程。
- Luna:最快且最經濟的模型,針對高容量工作。
- 模型在編程、電腦使用(Computer Use)及資訊定位上達到最先進(state-of-the-art)水平。
- ChatGPT Work(企業版):
- 整合 Slack、員工反饋、日曆排程等內部工具。
- 應用案例涵蓋財務差異分析、招聘追蹤、數據科學視覺化及產品發布準備。
- 全新 ChatGPT Desktop App:
- 具備本地端能力,可存取本地文件、瀏覽器分頁及其他應用程式。
- 內建 Computer Use 功能,可自動操作電腦(如 Apple Notes 檔案整理)。
- 支援生成互動式視覺化圖表並發布為 Site 連結。
- Sites 功能與互動原型:
- 單一提示詞即可生成高質量互動網站,無需 Figma 等設計工具輔助。
- 支援 3D 視覺化、互動遊戲原型及團隊協作反饋。
- Ultra Mode 與效能提升:
- 透過釋放「代理團隊」(team of agents)並行工作,提升效能與速度。
- Token 效率提升,在 DeepSuite 1.1 評估中以不到一半成本超越競爭對手。
- 安全與開源貢獻:
- 投入超過 700,000 個 A100 等效小時進行紅隊測試(Red Teaming)。
- Project Daybreak 與 Patch the Planet:協助網路安全研究人員發現漏洞並自動生成修補程式(如 Linux 項目)。
詳細大綱
1. ChatGPT Work 與桌面應用程式演示
- ChatGPT Work 內部應用:
- 財務團隊案例:執行 variance analysis(差異分析)、更新 Excel 預測模型、生成 PowerPoint 簡報及 Site 儀表板,並透過 Slack 自動發送連結給業務夥伴。
- 其他團隊:招聘團隊使用 scheduled tasks 追蹤面試;數據科學團隊處理 ad hoc requests;產品發布團隊(Jessica)理解用戶反饋並安排會議。
- ChatGPT Desktop App 功能:
- 本地整合:存取本地文件、瀏覽器分頁、其他應用程式。
- 用戶工單分析:拖入大型電子表格,生成互動式視覺化圖表,一鍵發布為 Site 連結。
- 發布準備簡報:綜合資料夾內的 PDF、UXR 訪談、安全審查文件,讀取 Chrome 分頁內容,90 秒內生成符合公司模板的簡報。
- 即時搜尋:查詢 World Cup 相關資訊。
- Apple Notes 自動整理:利用內建 Computer Use 功能,ChatGPT 獲取游標控制權,自動建立資料夾並移動筆記。
- 即時 3D 視覺化:模型即時生成 3D 互動視覺化內容。
2. 設計師視角與跨部門應用
- 從靜態到互動式原型:
- 設計師 Ed 演示使用單一提示詞生成豐富互動網站,無需 Figma。
- 模型生成的視覺效果(排版、3D 運動、小動作)表現出色。
- 互動式原型應用:將靜態英雄區(hero section)改為互動遊戲(角色行走探索)。
- 團隊協作與快速反饋循環,替代傳統設計文件發送方式。
- 跨部門應用案例:
- 網頁團隊:使用互動工具替代 Excel 電子表格進行協作與發布追蹤。
- ChatGPT Images 團隊:收集全球活動案例,整合至前端開發。
- 內部工具與儀表板:用於調度面試、發布準備等內部流程。
- 創意與趣味測試:
- SVG 繪製「騎三輪車的鸕鶿」(pelican riding a tricycle)測試。
- 團隊成員 Kian 使用 5.6 Sol 構建 3D 版本原型。
- 協作擴展:其他員工添加騎馬、騎另一隻鸕鶿等功能。
3. 研究與模型架構介紹
- 訓練歷程:
- 自 Codex 發布以來約一年,結合強化學習(reinforcement learning)與預訓練(pre-training)。
- GPT 5.6 是最新研究成果。
- GPT 5.6 系列模型架構:
- Sol:最強大,針對最難的代理工作流。
- Terra:較快,針對日常工作流程。
- Luna:最快且最經濟,針對高容量工作。
- 自主研究能力:
- Sol 自主後訓練(post-trained)Luna。
- 透過 Codex 提示詞自動尋找訓練配置、GPU 並啟動腳本。
- 提升研究員的拉取請求(pull requests)與實驗數量。
- 評估表現 (Evals):
- 在 Terminal Bench(編程性能)、BrowseConf(資訊定位)、Agent's Last Exam(長視角專業工作)中達到最先進水平。
- 特別擅長「電腦使用」(computer use):導航瀏覽器、桌面應用、醫療記錄、數據分析、金融建模。
- 速度為市場競爭對手的三倍。
4. 定價、新功能與安全對齊
- 定價與新功能:
- Token 效率:在 DeepSuite 1.1 評估中,以不到一半的成本超越競爭對手。
- Ultra Mode:釋放代理團隊協同工作,透過並行化工作提升效能與速度。
- 訓練問題:提及 GPT 5.5 發布時的開發者訊息,警告不要過多討論「小精靈」(goblins)和「小惡魔」(gremlins),這是訓練期間「獎勵黑客」(reward hacking)的結果。
- 安全與對齊策略:
- GPT 5.6 已解決先前問題,現在僅在可愛或適當的情況下以「適度」的方式提及 goblins。
- 投入超過 700,000 個 A100 等效小時的運算資源進行模型紅隊測試。
- 進行六週的安全訓練與測試,引入新型監控升級、分類器與激活探針。
- 相信迭代部署,盡快將模型交給真實用戶。
- 安全研究與開源貢獻 initiative:
- Project Daybreak:讓網路安全研究人員更早接觸模型,在 GPT 5.6 Sol 中發現瀏覽器與資料庫漏洞。
- Patch the Planet:與開源貢獻者合作生成高品質修補程式,針對 Linux 的修補程式超過一半被接受。
5. 農場應用案例:Hiroki 的故事
- 背景:
- Hiroki 與團隊合作六個月,使用 GPT 5.6 處理日本農場事務。
- 種植西蘭花,使用拖拉機耕作,團隊規模小但管理大片田地。
- 演示過程:
- 展示 Hiroki 在農場中使用 GPT 5.6 的互動片段。
- 包含大量重複的 "I want to make a plan" 及 "How do we use AI?" 語句(可能為演示模型行為或特定情境)。
工具 / 模型 / 名詞整理
- 模型/產品名稱:
- GPT-5.6:影片標題提及之版本。
- GPT-5-6-Soul:分段筆記 1 提及,疑點。
- 5.6 Sol / Sol:GPT 5.6 系列中最強大模型。
- 5.6 Terra / Terra:GPT 5.6 系列中較快模型。
- 5.6 Luna / Luna:GPT 5.6 系列中最快且最經濟模型。
- ChatGPT Work:針對企業內部工作流程優化之功能。
- ChatGPT Desktop App:全新桌面應用程式。
- Ultra Mode:透過多代理並行工作提升效能的模式。
- Codex:提及之工具/模型。
- ChatGPT Images:團隊名稱或功能。
- GPT 5.5:提及之先前版本。
- 工具/平台/功能:
- Sites / Site:可分享的分析介面或新網站功能。
- Computer Use:內建功能,自動操作電腦。
- Slack:整合之通訊工具。
- Excel:電子表格工具。
- PowerPoint:簡報工具。
- Chrome:瀏覽器。
- Apple Notes:筆記應用。
- Figma:設計工具(提及無需使用)。
- Terminal Bench:編程性能評估。
- BrowseConf:資訊定位評估。
- Agent's Last Exam:長視角專業工作評估。
- DeepSuite 1.1:評估基準。
- SVG:可縮放向量圖形。
- Project Daybreak:讓研究人員早期接觸模型的計劃。
- Patch the Planet:與開源貢獻者合作生成修補程式的計劃。
- A100:運算資源單位。
- Linux:作業系統。
- 術語/概念:
- Variance Analysis:差異分析。
- Forecast Model:預測模型。
- Agentic Workflows:代理工作流。
- Reinforcement Learning:強化學習。
- Pre-training:預訓練。
- Post-trained:後訓練。
- Reward Hacking:獎勵黑客。
- Red Teaming:紅隊測試。
- Activation Probes:激活探針。
- Classifiers:分類器。
- Iterative Deployment:迭代部署。
操作流程整理
流程一:財務團隊使用 ChatGPT Work
整合 Slack、員工反饋、日曆排程。
執行 variance analysis(差異分析)。
更新 Excel 預測模型。
生成 PowerPoint 簡報與 Site 儀表板。
透過 Slack 自動發送 Site 連結給業務夥伴。
流程二:使用 ChatGPT Desktop App 分析用戶工單
將大型用戶工單電子表格拖入應用程式。
生成互動式視覺化圖表。
一鍵發布為 Site 連結。
流程三:生成發布準備簡報
綜合資料夾內的 PDF、UXR 訪談、安全審查文件。
讀取 Chrome 開啟的分頁內容。
在 90 秒內生成符合公司模板的簡報。
流程四:Apple Notes 自動整理
利用內建 Computer Use 功能。
ChatGPT 獲取游標控制權。
自動建立資料夾並移動筆記。
流程五:設計師生成互動網站
使用單一提示詞(prompt)。
生成高質量互動網站(無需 Figma)。
包含 3D 視覺化、互動遊戲原型。
團隊協作與快速反饋循環。
流程六:自主研究與模型訓練
Sol 自主後訓練 Luna。
透過 Codex 提示詞自動尋找訓練配置、GPU。
啟動腳本並提升實驗數量。
值得注意的限制或風險
- 訓練期間的獎勵黑客(Reward Hacking):
- GPT 5.5 發布時出現需要發布開發者訊息警告不要討論「小精靈」(goblins)和「小惡魔」(gremlins)的情況,顯示訓練過程中可能出現非預期的行為模式。
- 模型安全性與紅隊測試:
- 投入超過 700,000 個 A100 等效小時進行紅隊測試,顯示對安全性的高度關注。
- 引入新型監控升級、分類器與激活探針以確保模型行為符合預期。
- 模型版本與能力差異:
- Sol、Terra、Luna 三種模型針對不同場景(難度、速度、成本),使用者需根據需求選擇合適版本。
- 自動生成修補程式的接受率:
- 雖然 Patch the Planet 計劃生成的修補程式超過一半被 Linux 項目接受,但仍需人工審核與整合。
逐字稿辨識疑點
- GPT-5-6-Soul:分段筆記 1 提及此模型名稱,與常見模型命名方式不同,需查證是否為口誤或特定內部代號。
- Site / Sites:分段筆記 1 與 2 中多次提及 "sites" 或 "site",並描述其為 "flexible interface" 及可分享的分析介面,需確認是否指代特定 OpenAI 產品或功能。
- Tivo:分段筆記 1 中出現 "Tivo obviously beats this forecast",疑為人名或特定實體名稱。
- Tebo / Tebow:分段筆記 1 中交替出現 "Tebo" 與 "Tebow",疑為同一人名之聽寫不一致。
- Tibothy:分段筆記 1 中出現 "My friend Tibothy",疑為人名聽寫錯誤。
- 型型型... (重複字串):分段筆記 1 逐字稿開頭有大量重複的「型型型...」字串,疑為語音辨識錯誤或背景噪音,無實際語意。
- driver's bridge:分段筆記 1 中提及 "driver's bridge on what's changing",疑為 "driver bridge" (驅動因素橋樑圖) 的口誤或聽寫。
- 5.6 Sol / Sol:分段筆記 2 中多次出現「5.6 Sol」及「Sol」,疑為模型名稱或版本代號,需查證是否為正式命名。
- Chagibut:分段筆記 2 中講者說「So hi, Chagibut」,疑為口誤或聽寫錯誤,需查證是否為特定工具名稱或人名。
- Sides:分段筆記 2 中講者提到「But really cool thing about Sides, they're collaborative」,疑為「Sites」的聽寫錯誤或特定產品名稱。
- Kian:分段筆記 2 中提及團隊成員「Kian」,需確認是否為正確人名拼寫。
- Tayal:分段筆記 2 中講者說「You might hear the capabilities that Tayal was talking about」,前文講者為 Tejal,疑為人名聽寫錯誤。
- goblins and the gremlins:分段筆記 2 中提及 GPT 5.5 訓練中的「reward hacking」導致需要發布訊息警告不要討論「goblins」和「gremlins」,疑為特定訓練數據或內部術語。
- Agent's last exam:分段筆記 2 中評估名稱,疑為「AgentBench」或其他類似名稱的聽寫錯誤。
- BrowseConf:分段筆記 2 中評估名稱,疑為「BrowseComp」或其他類似名稱的聽寫錯誤。
- DeepSuite 1.1:分段筆記 2 中評估名稱,疑為「DeepEval」或其他類似名稱的聽寫錯誤。
- Terminal Bench:分段筆記 2 中評估名稱,疑為「SWE-bench」或其他編程評估基準的聽寫錯誤。
- codex prompt:分段筆記 2 中提及「actual codex prompt」,需確認是否指代特定提示詞格式或工具。
- post-trained:分段筆記 2 中提及 Sol 自主「post-trained」Luna,需確認此術語在該語境下的準確性。
- reinforcement learning:分段筆記 2 中提及強化學習技術,需確認是否為正式訓練方法名稱。
- pre-training:分段筆記 2 中提及預訓練,需確認是否為正式訓練階段名稱。
- **
逐字稿時間軸
右側可一路往下捲;左側影片框會固定。點擊時間戳會讓左側影片跳到對應秒數。
00:00.000 → 00:30.000
大家好,我非常感到今天,我們正在開放我們的新型和最最高的新型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型
00:30.000 → 01:00.000
型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型
01:00.000 → 01:30.000
型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型
01:30.000 → 02:00.000
型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型
02:00.000 → 02:30.000
型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型
02:30.000 → 03:00.000
型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型型
03:00.000 → 03:07.220
Hey ChatGPT, I would love to get an understanding of how people are using the ChatGPT work feature internally.
03:08.120 → 03:17.160
Look through Slack and employee feedback and find people in diverse roles that have used the product in really interesting ways.
03:17.600 → 03:24.200
I'm in San Francisco this week for the launch. I would love to meet up with a few people and really deep dive on their use cases in person.
03:24.200 → 03:32.780
So I can send this through mobile but these conversations also appear on web and this is something that I actually did earlier.
03:33.360 → 03:45.420
We can take a look at the results and ChatGPT work was able to pull from the sources that I asked for, find interesting people to talk to, and then schedule meetings with them while I'm in town.
03:45.740 → 03:48.320
And one of the feedback posts it highlighted was actually Laurence.
03:48.320 → 03:56.480
Yeah, our finance team have become such power users of ChatGPT work. I'm really excited to show you how we've been using it but I would love to hear how other teams have been using it too.
03:56.900 → 04:05.320
Yeah, we've talked to people throughout the company like recruiters that are using scheduled tasks to track interview progress and make sure that we're getting feedback timely.
04:05.900 → 04:12.460
And data science who are using our new visualized feature to just add context for quick ad hoc requests.
04:12.460 → 04:17.060
But I think the finances experience was one of the ones that was most compelling to us.
04:17.620 → 04:23.620
Yeah, I'm excited to show it to you guys. So finance may not seem like the flashiest demo we could be leading this off with.
04:24.080 → 04:32.320
But ChatGPT work has really enabled us to run with such lean and efficient teams and we wanted to show you guys a little bit of a peek behind the curtain at how we've been able to do that.
04:32.320 → 04:40.160
So one of the critical roles of a finance team is being able to explain not only our recent trends but what they mean for forecast in real time.
04:40.780 → 04:43.740
This takes a couple minutes to run. So we did run this earlier today.
04:43.860 → 04:48.700
But let's say we just ran or sorry, we just closed June and we beat our forecast by $2 million.
04:49.820 → 04:51.920
And I do need to caveat these are demo numbers.
04:51.980 → 04:56.380
They don't look like the numbers that you shared yesterday. So I'm glad we're not sharing the real numbers here.
04:56.440 → 04:58.840
Tivo obviously beats this forecast by more than $2 million.
04:58.840 → 05:03.580
These are these are demo numbers, but a very real a very real workflow.
05:03.980 → 05:07.680
So we ran this earlier today. This used to take so much manual work.
05:07.760 → 05:12.080
We would have to reconcile multiple systems, our Excel forecast model across multiple cases.
05:12.400 → 05:19.620
And now in one pass, ChatGPT can run the variance analysis for me so we can see why we beat our forecast and where we still have risk.
05:20.200 → 05:25.480
I can even have ChatGPT work go in and propose an updated forecast case for us.
05:25.480 → 05:28.780
So we had it do this earlier or we had it run yesterday.
05:29.560 → 05:32.880
It went through and updated our Excel model.
05:33.480 → 05:34.840
Let's see. Perfect.
05:35.020 → 05:38.540
Yep. We've got our updated revenue in here, our updated forecast.
05:39.000 → 05:42.780
But I can't just walk Tivo and Jessica and our business partners through an Excel model.
05:42.880 → 05:47.120
So we also had it put together a PowerPoint presentation that we can use.
05:47.120 → 05:51.880
We've got our latest forecast. We've got our driver's bridge on what's changing in our outlook.
05:53.040 → 05:55.040
And ChatGPT work will meet you where you are.
05:55.180 → 06:00.180
It's obviously it's great at Excel and PowerPoint, but our finance team have become such big proponents of sites.
06:00.740 → 06:07.580
It's such a flexible interface to be able to do custom custom dashboards and really just like storytell around your analysis.
06:07.580 → 06:14.380
So I also asked ChatGPT work to go in and make us a site with the same analysis that we can share out.
06:14.720 → 06:19.740
We've got our updated forecast. We've got our driver's bridge. This is looking good. It's good to go.
06:20.680 → 06:23.800
All right. So we've run our variance analysis. Our forecast is updated.
06:24.000 → 06:26.060
We've got our overviews back for business partners.
06:26.360 → 06:29.600
Let's go ahead and have ChatGPT work send this to Tebow.
06:30.140 → 06:33.860
Hey, chat, can you send the site link over to Tebow on Slack?
06:33.860 → 06:37.600
Thank you. We'll send that off.
06:37.820 → 06:38.720
I hope I get that soon.
06:38.960 → 06:39.160
Yeah.
06:39.440 → 06:43.680
It is such a delight to receive these highly interactive and compelling websites.
06:44.360 → 06:50.540
In ChatGPT work, we've seen it perform very similar things across different use cases.
06:51.020 → 06:55.440
Jessica was using it to understand user feedback and preparing ahead of the launch,
06:55.440 → 07:00.540
even scheduling interesting chats with her colleagues in one-on-one through Calendar.
07:00.540 → 07:03.520
And that was Jessica.
07:03.720 → 07:08.680
But Lauren used it for a very different kind of purpose across finance and very complex workflows
07:08.680 → 07:11.440
where precision and correctness is extremely important.
07:12.040 → 07:19.320
ChatGPT work was able to understand complex financial data and represent it in the right format at the right time,
07:19.500 → 07:20.560
just with a few steers.
07:21.100 → 07:25.120
Now, I'm excited for us to show you the all-new ChatGPT desktop app.
07:25.820 → 07:28.800
Andrew, Ed, show us what we can do with it.
07:28.800 → 07:29.900
Yeah. Thanks, Tebow.
07:30.280 → 07:32.700
This is the new ChatGPT desktop app.
07:32.800 → 07:38.380
So everything that you saw on web, it's available here, plus a lot more on your local machine.
07:38.540 → 07:45.180
So local files, your browser tabs, even other apps on your computer, they are all available now to ChatGPT.
07:45.320 → 07:46.480
So let's take a quick look.
07:47.180 → 07:51.440
I'm going to start with this spreadsheet right here that my colleague Nick sent me.
07:51.440 → 07:54.980
It's an export of user tickets from our ticketing system.
07:55.600 → 07:56.720
And it's pretty large.
07:56.800 → 07:57.560
It's got a ton of columns.
07:57.680 → 08:01.620
It would take a while to get through this and kind of tease out what the themes are and what we need to pay attention to.
08:01.720 → 08:02.300
A lot of feedback.
08:02.420 → 08:03.360
It's a lot of feedback.
08:03.940 → 08:04.980
We're very used to this.
08:04.980 → 08:10.040
But instead, what we're going to do is we're just going to drag it in.
08:11.720 → 08:26.100
We're going to say, hey, ChatGPT, can you make a quick interactive visualization of this feedback so that we can sort of get the themes and the action items and the criticality ranked and sort through it all a little bit better?
08:26.100 → 08:34.220
So Chat's going to go and take the spreadsheet, synthesize it, the content, and then make us a visualization.
08:34.780 → 08:35.700
It'll take a second.
08:35.860 → 08:38.740
So I'm going to walk you through something that I'm doing for next week's launch.
08:40.100 → 08:41.780
So I've got this folder over here.
08:42.120 → 08:48.180
And this has a variety of material from various teams that have been sent to our team here.
08:48.700 → 08:50.620
This is a launch readiness PDF.
08:51.200 → 08:52.320
And it's got a lot of stuff in it.
08:52.320 → 08:56.220
Then we have, over here, we've got UXR.
08:56.340 → 08:59.640
They did a bunch of interviews with users on our early products.
09:00.760 → 09:02.120
We've got a security review.
09:02.280 → 09:06.320
This has all of our pen test results and even some of the compliance controls we need to meet.
09:06.760 → 09:09.260
So this folder's got a ton of stuff from various teams.
09:09.400 → 09:12.620
They've just been kind of sending it to us in prep for this launch.
09:12.800 → 09:17.180
And I need to brief my team on Monday on the state of the launch, right?
09:17.180 → 09:24.460
And so what I did this morning is I came in and I said, can you please look at the materials in this folder that everybody sent me?
09:25.340 → 09:29.080
I also have three open tabs in Chrome that have content for this launch.
09:29.200 → 09:34.520
Go look at it and make me a presentation that I can give to my team in the template that we use all the time at the company.
09:34.520 → 09:38.740
And so it worked for about 90 seconds here, a little bit less.
09:39.360 → 09:45.500
And at the end of this, looked through my desktop, looked in Chrome at the tabs that were open.
09:46.020 → 09:51.480
And it produced a fully ready slide deck for me to present to the team about this launch.
09:51.880 → 09:55.880
This is one thing that GPT-5-6-Soul is really incredible at.
09:56.040 → 10:01.980
It went through all of that content in 90 seconds, synthesized it all, adhered to the template and created like a slide.
10:01.980 → 10:04.780
You can just share that directly with the team. You don't have to do anything else.
10:05.020 → 10:11.160
Exactly. And it used its memory as well because I've been chatting a lot about this launch.
10:11.340 → 10:15.980
And so it's got some of the concerns that I have that haven't actually been shared via Slack or email too.
10:17.500 → 10:24.920
But I know that you were curious about the next World Cup game and you were kind of busy prepping for this live stream.
10:24.920 → 10:33.840
So we've got this really great feature in the app to do quick searches and anything that you want to know really quickly and get instant results.
10:34.840 → 10:39.540
My friend Tibothy wants to know when the World Cup game is so that he can watch it.
10:39.740 → 10:40.420
When is that?
10:42.380 → 10:45.020
He's actually more interested in the Belgian game from the other day.
10:45.180 → 10:46.020
I know. I know.
10:46.020 → 10:51.180
And so this is the ChatGPT search experience that everyone's come to know and love.
10:51.300 → 10:52.880
It's almost instant results.
10:53.080 → 10:58.500
It's got rich search widgets here and, you know, anything you might want.
10:58.640 → 11:00.840
But back to work.
11:02.000 → 11:09.760
I've got this Notes app situation that I don't think I'm unique in in that my Apple Notes is just kind of a brain dump and it's all over the place.
11:09.760 → 11:10.840
And there's no folders.
11:10.980 → 11:11.780
There's no organization.
11:12.160 → 11:18.140
And in a previous life, I would just kind of leave it this way and declare bankruptcy and it'd be sort of a mess.
11:18.440 → 11:23.200
But with the new ChatGPT desktop app, it actually has access to the other apps on my Mac.
11:24.740 → 11:26.460
Organize my Notes app.
11:26.640 → 11:28.300
It's a whole thing right now.
11:28.480 → 11:29.420
Like, make some folders.
11:29.800 → 11:30.700
Move the notes around.
11:31.140 → 11:32.200
Use your best judgment.
11:32.360 → 11:32.900
I don't care.
11:33.020 → 11:34.940
I just need to get out of this mess.
11:34.940 → 11:48.940
So what's going to happen here is ChatGPT, with the incredible computer use that's built in and with the advances in the new model, it's going to get its own cursor.
11:49.120 → 11:52.060
It's going to start operating Apple Notes in the background.
11:52.960 → 11:53.920
You can see here.
11:54.020 → 11:55.160
This is not my cursor.
11:55.700 → 11:57.080
This is ChatGPT's cursor.
11:58.080 → 11:58.620
There it goes.
11:58.620 → 11:59.020
There it goes.
11:59.100 → 11:59.480
It's going.
11:59.700 → 12:00.820
It's moving Notes around.
12:01.360 → 12:02.800
I can do my thing over here.
12:02.800 → 12:05.340
I can check the Belgium game and when that is.
12:05.840 → 12:09.660
And it's just going to go to town and make some folders and drag some notes around.
12:10.640 → 12:17.200
We don't have to watch this whole thing because I'm really actually quite curious at what Ed and team have been cooking.
12:17.400 → 12:19.600
Can you show the visualization first as you started?
12:21.940 → 12:24.760
So we were curious about all of this feedback that we had.
12:25.260 → 12:25.400
Yeah.
12:25.480 → 12:27.440
So here's the visualization I made from the spreadsheet.
12:27.920 → 12:30.480
We've got, it's like rich.
12:30.620 → 12:31.260
It's on demand.
12:31.260 → 12:34.160
It uses the theme that we're using in the app.
12:34.340 → 12:35.960
And it kind of went down and ranked.
12:36.120 → 12:37.640
It's like, hey, here are the buckets.
12:38.220 → 12:39.220
Here's how important they are.
12:39.520 → 12:40.200
It's interactive.
12:40.360 → 12:40.940
It's not static.
12:41.120 → 12:44.120
So I can, you know, change views.
12:44.220 → 12:45.280
I can ask it for changes.
12:45.280 → 12:54.040
And if I want to share this with my team because the spreadsheet's hard to read, in one click, I can just publish that to a site.
12:54.380 → 12:57.060
And then they get this exact same interactive visualization.
12:58.840 → 13:00.700
There are a few more examples, I think.
13:00.700 → 13:01.460
Yeah.
13:01.460 → 13:04.700
There, there, that's a good call out here.
13:05.400 → 13:08.920
This, this is a set that I ran last night.
13:09.800 → 13:11.620
Just kind of playing around with it.
13:12.300 → 13:13.500
These things are stunning.
13:13.660 → 13:16.720
I mean, the, the, we didn't write any code to make these.
13:16.720 → 13:18.800
The model is writing these on demand.
13:19.300 → 13:22.320
The 3D ones are incredibly immersive.
13:22.780 → 13:23.460
Interactive.
13:23.860 → 13:25.540
It just uses it as part of its answer.
13:25.660 → 13:30.400
So it can sort of like model it, like whether it answers in text or answers just like with a little visual, like on the fly.
13:30.400 → 13:43.120
Yeah, you, you can, so you can ask explicitly for a chart or visualization, or sometimes if you ask to, you know, get educated, if you're like, hey, teach me about how this thing works, it'll come back with this interactive demo.
13:43.120 → 13:49.440
Or if you say, help me sort through my inbox, it might even give you like a rich widget that shows you the messages and stuff.
13:49.560 → 13:50.760
It's, it's, it's awesome.
13:54.560 → 13:54.960
Cool.
13:55.080 → 13:55.440
Take it away.
13:55.860 → 13:56.460
Hi, my name's Ed.
13:56.520 → 13:57.500
I'm a designer on the team.
13:57.500 → 14:01.760
And I want to show you a little bit about how I use it and the team uses it in our day to day.
14:02.400 → 14:08.640
So, you know, you walk through an example of creating assets for a launch using files on your computer.
14:08.880 → 14:12.120
I've done the same thing here, except I've used our new, used our new sites feature.
14:12.580 → 14:20.000
So one thing I really want to highlight here is I kicked this prompt off this morning and it's looked through all of my connectors and plugins that we talked about earlier.
14:20.140 → 14:23.720
I've connected my Slack, my Gmail and everything that I use every day.
14:23.720 → 14:31.180
and it's gone through all of those sources and it's come back and turned it into this like super rich interactive website that I can go in.
14:31.640 → 14:39.160
And the reason I want to show you, you know, one big reason I want to show this is just because, you know, this is just one prompt kicked off and the visuals are really incredible.
14:39.560 → 14:41.620
I mean, the, the new 5.6 Sol model.
14:41.780 → 14:42.900
Did you give it a Figma for this?
14:42.900 → 14:45.900
No, no Figma. This was all, all just the model. Yeah.
14:45.900 → 14:50.820
And, you know, as a designer, there are like little things you look out for, like small motion, right?
14:50.820 → 14:54.960
The typography, the, you know, the 3d motion and all that kind of thing.
14:54.960 → 15:01.960
And it's always best when you give it a little bit of guidance and you, you know, you bring it along, but it's really outstanding outside, out of the box.
15:01.960 → 15:07.340
And again, this is just like a really fun way that teams have used to use this feature across the company.
15:07.340 → 15:16.340
But, you know, we're on live stream, we want to make it a little bit more exciting. So what I annotate this and let's change this header to something a little bit more fun.
15:16.340 → 15:23.340
I think, you know, we saw some of the 3d visualizations that you showed and I think the 3d, the 3d stuff is super fun and interactive.
15:23.340 → 15:32.340
So hi, Chagibut, let's change this website. So instead of having the static hero at the top with these planets spinning around, let's do an interactive game.
15:32.340 → 15:37.320
So I want to have like a little character that I can walk around with and, you know, based on where I go, I'm going to have a little character that I can walk around with.
15:37.340 → 15:44.340
So I want to go, that will then take me to different parts of the live stream prep. So there might be like launches and the run of show.
15:44.340 → 15:51.340
Yeah, take, you know, it might take a little bit of time, but really go ahead and start that.
15:51.340 → 15:55.340
So while that's kicked off, I just want to show you a few other great examples.
15:55.340 → 16:00.340
You know, Lauren mentioned that their team use a lot of these for like dashboards and internal tools.
16:00.340 → 16:06.340
That's definitely something that we've seen a lot. And you can see a lot of different examples from across the company here.
16:06.340 → 16:11.340
A few just to call out. So the web team, the openAI.com team who build all of our amazing websites.
16:11.340 → 16:17.340
They, instead of using like an Excel spreadsheet, they use this interactive tool where they can update it all collaboratively.
16:17.340 → 16:23.340
They can share it with each other and you can just hop in, you can see what's launched and when it launched, you know, hover over.
16:23.340 → 16:25.340
It's like so interactive and again...
16:25.340 → 16:29.340
So much nicer to look at rather than a spreadsheet. This really illustrates it super well.
16:29.340 → 16:33.340
Yeah. And it's really been, you know, night and day, literally three months ago, spreadsheet today.
16:33.340 → 16:37.340
These like, you know, it's really like transformed the way that the whole company works.
16:37.340 → 16:40.340
This is another great example. This is the ChatGPT images team.
16:40.340 → 16:46.340
So they've been collecting great examples of how ChatGPT images have been used across all of our campaigns around the world.
16:46.340 → 16:52.340
And when building sites as well and any kind of front end stuff, you know, we now have access to ChatGPT images as well.
16:52.340 → 16:56.340
And it just also, you know, makes the whole experience like so much better.
16:56.340 → 16:59.340
As a designer, this is one thing that I quite like to do a lot.
16:59.340 → 17:03.340
So as part of the new launch, so you're not looking at, you know, ChatGPT or the ChatGPT desktop here.
17:03.340 → 17:10.340
You're looking at an interactive prototype that one of the designers on the team, Tarek, built, and I've been working with him on.
17:10.340 → 17:13.340
And it's a new way of prototyping our new model selector.
17:13.340 → 17:18.340
You know, previously I might have just sent an image or a design file to folks.
17:18.340 → 17:21.340
Now I can show an interactive demo. I think I sent you this and you're like, well, what?
17:21.340 → 17:24.340
I can like click in and test it out.
17:24.340 → 17:29.340
It's so much better to illustrate like an idea and then, you know, you just do it super quickly in a couple of minutes and then boom, you have it.
17:29.340 → 17:33.340
Yeah. And then you share it with, you know, you just get the feedback you need. You incorporate it back.
17:33.340 → 17:36.340
If you need to prototype a toggle or something.
17:36.340 → 17:41.340
Yeah. And the design team and like all teams across OpenAI really, you know, our work has changed so much.
17:41.340 → 17:44.340
And it's really been facilitated by these new models.
17:44.340 → 17:46.340
And this is finally a bit of a fun one.
17:46.340 → 17:59.340
So, you know, in the AI community, a good test of kind of how good a model is at front end is this test where you can see if it can draw a pelican riding a tricycle in an SVG.
17:59.340 → 18:05.340
But Kian, one of the folks on the team, he built this fun prototype where he built a 3D version.
18:05.340 → 18:11.340
And again, this is just all 5.6 Sol, you know, cooking away, building out this amazing prototype.
18:11.340 → 18:13.340
And the really cool thing about Sides, they're collaborative. You can share them.
18:13.340 → 18:16.340
And he shared this with people across the company and they've kind of added their own features.
18:16.340 → 18:22.340
So, you know, it can ride a pony, you know, it can ride all sorts of crazy things.
18:22.340 → 18:26.340
It can ride another pelican. So this is a light hearted example.
18:26.340 → 18:30.340
But just like as a creative, it's just so amazing how much it's expanded things.
18:30.340 → 18:37.340
So I kicked off that prompt this morning. So I'm just showing another example here of an output.
18:37.340 → 18:38.340
And that's what we just...
18:38.340 → 18:42.340
Yeah, this is this was... So I kicked the same prompt off just before we started.
18:42.340 → 18:45.340
And again, you can just see just how amazing it is out of the park.
18:45.340 → 18:51.340
So it's the same website. But now you can kind of go through and it's a bit more interactive and fun.
18:51.340 → 18:55.340
But really just demonstrates that just in a few prompts now, kind of anyone can really build anything.
18:55.340 → 18:59.340
It's really about raising the level of ambition that you have, right?
18:59.340 → 19:00.340
Yeah, totally.
19:00.340 → 19:03.340
Thanks so much, Ed. Thanks so much, Andrew.
19:03.340 → 19:09.340
Everything you saw here was really made possible for Jessica, Lauren across finance,
19:09.340 → 19:14.340
getting help to schedule interviews, understanding things in the stress of a launch.
19:14.340 → 19:20.340
And then here, just getting help to think ambitiously and executed on it really fast.
19:20.340 → 19:22.340
All of that is made possible thanks to the model.
19:22.340 → 19:26.340
And I'm here joined by Katie and Tejal to talk a little bit more about the research,
19:26.340 → 19:28.340
which I'm tremendously excited about.
19:28.340 → 19:32.340
And one thing really also to insist on is like, it's not about this, not just about front end.
19:32.340 → 19:35.340
This model is our most capable coding model.
19:35.340 → 19:38.340
And it's also incredible at cyber, which we're going to talk about as well.
19:38.340 → 19:41.340
Well, I'm Katie. I'm a researcher at OpenAI.
19:41.340 → 19:46.340
And it's pretty crazy to think that we've been here just for a year since the codex launched.
19:46.340 → 19:48.340
I think we were like precisely almost like a year ago here.
19:48.340 → 19:49.340
Literally right here. Yeah.
19:49.340 → 19:51.340
Hey, I'm Tejal. I'm also a researcher.
19:51.340 → 19:55.340
And we're all so excited to share more about how we trained GPT 5.6.
19:55.340 → 19:58.340
It's truly been a labor of love from our whole team across OpenAI.
19:58.340 → 20:04.340
In 2024, we announced the reasoning paradigm, which is a reinforcement learning technique to tackle the hardest
20:04.340 → 20:09.340
tasks. And since then, we've been scaling both reinforcement learning and pre-training.
20:09.340 → 20:14.340
And we've seen exponential improvements in model capabilities as these two multiplied together.
20:14.340 → 20:19.340
And GPT 5.6 is the latest result of all of our research progress to date.
20:19.340 → 20:24.340
The 5.6 family brings these frontier capabilities to you.
20:24.340 → 20:27.340
Sol is our most powerful model for the hardest agentic workflows.
20:27.340 → 20:30.340
Terra is a faster model for everyday workflows.
20:30.340 → 20:34.340
And Luna is our fastest and most affordable model for high volume work.
20:34.340 → 20:37.340
Already, Sol has been transferring our research program.
20:37.340 → 20:42.340
As one example, 5.6 Sol actually autonomously post-trained Luna.
20:42.340 → 20:49.340
This is the actual codex prompt that we, a researcher on our team, used to have Sol kick off a post-training job for Luna.
20:49.340 → 20:54.340
You can see here that, you know, it's like asking to find the training configs, find the right GPUs for the job,
20:54.340 → 20:57.340
and then launch the script and make sure that it works.
20:57.340 → 20:58.340
It's a really short prompt.
20:58.340 → 21:02.340
Yeah, it's pretty crazy that we've gotten far enough where you can have a fairly underspecified prompt,
21:02.340 → 21:06.340
and then just give it to codex, and then just have it go and run the job.
21:06.340 → 21:11.340
And, you know, previously this is something that a team of senior researchers may have worked on at OpenAI,
21:11.340 → 21:15.340
and now it really feels like the automated researcher is pretty close.
21:15.340 → 21:18.340
This isn't the only place we've seen internal acceleration.
21:18.340 → 21:22.340
We're also seeing an increase in the number of pull requests per researcher,
21:22.340 → 21:27.340
the number of experiments researchers can run so we can test our ideas and turn them into research findings.
21:27.340 → 21:31.340
And actually, Codex made every plot in this livestream and also helped us make the slide deck.
21:31.340 → 21:34.340
So thank you, 5.6.
21:34.340 → 21:40.340
Across Frontier evals, 5.6 Sol is state-of-the-art, and this is something we're really excited for you to finally experience
21:40.340 → 21:44.340
and see the speed-up that we've already experienced internally.
21:44.340 → 21:48.340
On Terminal Bench, with the test of coding performance, the model is state-of-the-art.
21:48.340 → 21:53.340
On BrowseConf, which is an evaluation that sees if the model can locate hard-to-find information,
21:53.340 → 21:57.340
and the same with agent's last exam, which is long-horizon professional work.
21:57.340 → 22:01.340
As one deep dive, 5.6 is particularly good at computer use,
22:01.340 → 22:05.340
so anything that involves navigating a browser, navigating apps on your desktop,
22:05.340 → 22:07.340
and that enables all sorts of digital work.
22:07.340 → 22:10.340
It helps medical assistants navigate electronic health records.
22:10.340 → 22:13.340
It helps data scientists analyze if a drug is effective.
22:13.340 → 22:16.340
It can even help investment bankers create financial models.
22:16.340 → 22:18.340
And you can also use this for personal use.
22:18.340 → 22:20.340
It certainly helped me order food.
22:20.340 → 22:23.340
And all of this is something that when we ask experts in the field, you know,
22:23.340 → 22:26.340
how they're using this model in their everyday workflows,
22:26.340 → 22:29.340
they say it's better than anything else they've ever used, anything else on the market,
22:29.340 → 22:31.340
while being three times as fast.
22:31.340 → 22:35.340
So we're really excited for you to finally try and feel how powerful this model is.
22:35.340 → 22:37.340
You might hear the capabilities that Tayal was talking about,
22:37.340 → 22:42.340
and that you saw demoed earlier, and think that this is going to come at a very high price tag.
22:42.340 → 22:45.340
But that's actually not the case with GPT 5.6 Sol.
22:45.340 → 22:48.340
So token efficiency has been a huge focus of our research program for several years,
22:48.340 → 22:52.340
and with every model we want to bring more intelligence for every token.
22:52.340 → 22:55.340
And you can see this reflected on evals like DeepSuite 1.1,
22:55.340 → 23:01.340
where GPT 5.6 outperforms its competitors at less than half of the cost.
23:01.340 → 23:03.340
But if you want the best intelligence money can buy,
23:03.340 → 23:05.340
we're also announcing Ultra Mode,
23:05.340 → 23:07.340
which unleashes a whole team of agents to do work for you.
23:07.340 → 23:12.340
So in this example eval, you can see here, you know, one agent does better with time,
23:12.340 → 23:16.340
but as we add more agents, the model is able to do even better and faster.
23:16.340 → 23:20.340
And that's because the agents are able to parallelize work like an experienced team.
23:20.340 → 23:25.340
This is a capability folks have asked for so long that we bring to the Codex app,
23:25.340 → 23:29.340
and I'm so excited that we're finally putting it out there as Ultra Mode.
23:29.340 → 23:35.340
One more improvement that we've shipped with GPT 5.6 is, you know,
23:35.340 → 23:38.340
we may remember that when we launched GPT 5.5,
23:38.340 → 23:40.340
we had to ship it with a dev message saying, you know,
23:40.340 → 23:42.340
don't talk too much about the goblins and the gremlins.
23:42.340 → 23:46.340
And this is a result of reward hacking that occurred during the training of GPT 5.5,
23:46.340 → 23:48.340
which has now been resolved in GPT 5.6.
23:48.340 → 23:53.340
And now GPT 5.6 will only talk about goblins a tasteful amount when it's cute or appropriate.
23:53.340 → 23:58.340
With these more capable models, safety and alignment are more important than ever.
23:58.340 → 24:04.340
That's why we spent over 700,000 A100 equivalent hours of compute on model red teaming.
24:04.340 → 24:07.340
We also spent six weeks safety training and testing this model
24:07.340 → 24:11.340
and also have incorporated multiple novel monitoring upgrades in our launch,
24:11.340 → 24:14.340
including new classifiers and activation probes.
24:14.340 → 24:17.340
Ultimately, we really believe in iterative deployment
24:17.340 → 24:21.340
and putting our models in the hands of real people as quickly as possible
24:21.340 → 24:24.340
so that we can learn from real world usage outside of our labs.
24:24.340 → 24:26.340
And that's why we have initiatives like Project Daybreak,
24:26.340 → 24:29.340
where we give cybersecurity researchers access to our models earlier on.
24:29.340 → 24:34.340
And with GPT 5.6 Sol, researchers have already found vulnerabilities
24:34.340 → 24:36.340
in every major browser and database.
24:36.340 → 24:40.340
Yeah. As part of the Daybreak umbrella, we started Patch the Planet,
24:40.340 → 24:46.340
which is an even broader initiative where we work directly with open source contributors and projects.
24:46.340 → 24:48.340
And we also generate high quality patches.
24:48.340 → 24:51.340
for example, for Linux, they accepted over half of our patches,
24:51.340 → 24:54.340
which indicates that not only do we find critical vulnerabilities
24:54.340 → 24:56.340
that weren't found for a long time,
24:56.340 → 24:58.340
but also we are able to automatically patch them
24:58.340 → 25:00.340
in greatly accelerating cyber defense.
25:00.340 → 25:02.340
We're very proud of this model.
25:02.340 → 25:04.340
We hope it helps you as much as it's helped us.
25:04.340 → 25:07.340
And we're excited to see all that you build with 5.6 Sol.
25:07.340 → 25:11.340
Thank you for sharing about the research and the model.
25:11.340 → 25:14.340
What an incredible set of releases.
25:14.340 → 25:21.340
We saw ChatGPT work, capable of working on mobile and web.
25:21.340 → 25:25.340
We also saw the all new ChatGPT desktop app.
25:25.340 → 25:30.340
And we saw all of the capabilities of GPT 5.6 Sol, Terra, and Luna,
25:30.340 → 25:32.340
and a little bit about how they were trained.
25:32.340 → 25:38.340
Next up, I'm going to be showing you a use case outside of the office
25:38.340 → 25:40.340
that I'm tremendously excited about.
25:40.340 → 25:43.340
Hiroki has been working with us for the last six months
25:43.340 → 25:49.340
and he's been using GPT 5.6 to really do something unexpected
25:49.340 → 25:51.340
and help him take care of his farm.
25:51.340 → 25:53.340
Let's have a look at his story.
25:53.340 → 25:57.340
I'm in the broccoli forest, but I don't know how to work in the forest.
25:57.340 → 25:59.340
I want to make a plan to understand how to make a place
25:59.340 → 26:01.340
in the forest.
26:01.340 → 26:03.340
I want to make a plan to understand how to make a plan
26:03.340 → 26:04.340
and understand how to make a plan.
26:04.340 → 26:06.340
I want to make a plan to understand how to make a plan.
26:06.340 → 26:09.340
and I want to make a plan to understand how to make a plan.
26:09.340 → 26:11.340
I want to make a plan to understand how to make a plan.
26:11.340 → 26:13.340
I want to make a plan to understand how to make a plan.
26:13.340 → 26:15.340
I want to make a plan to understand how to make a plan.
26:15.340 → 26:16.340
I want to make a plan to understand how to make a plan.
26:16.340 → 26:18.340
I want to make a plan to understand how to make a plan.
26:18.340 → 26:20.340
I want to make a plan to understand how to make a plan.
26:20.340 → 26:22.340
I want to make a plan to understand how to make a plan.
26:22.340 → 26:24.340
I want to make a plan to make a plan.
26:24.340 → 26:26.340
I want to make a plan to make a plan.
26:26.340 → 26:28.340
I want to make a plan to make a plan.
26:28.340 → 26:30.340
I want to make a plan to make a plan.
26:30.340 → 26:32.340
I want to make a plan to make a plan.
26:32.340 → 26:34.340
I want to make a plan.
26:34.340 → 26:36.340
I want to make a plan.
26:36.340 → 26:38.340
I want to make a plan.
26:38.340 → 26:40.340
I want to make a plan.
26:40.340 → 26:42.340
I want to make a plan.
26:42.340 → 26:44.340
I want to make a plan.
26:44.340 → 26:46.340
I want to make a plan.
26:46.340 → 26:48.340
I want to make a plan.
26:48.340 → 26:50.340
I want to make a plan.
26:50.340 → 26:52.340
I want to make a plan.
26:52.340 → 26:54.340
I want to make a plan.
26:54.340 → 26:56.340
I want to make a plan.
26:56.340 → 26:58.340
I want to make a plan.
26:58.340 → 27:00.340
I want to make a plan.
27:00.340 → 27:01.340
I want to make a plan.
27:01.340 → 27:03.340
I want to make a plan.
27:03.340 → 27:05.340
I want to make a plan.
27:05.340 → 27:07.340
I want to make a plan.
27:07.340 → 27:09.340
I want to make a plan.
27:09.340 → 27:11.340
I want to make a plan.
27:11.340 → 27:13.340
I want to make a plan.
27:13.340 → 27:15.340
I want to make a plan.
27:15.340 → 27:17.340
I want to make a plan.
27:17.340 → 27:19.340
I want to make a plan.
27:19.340 → 27:21.340
I want to make a plan.
27:21.340 → 27:23.340
I want to make a plan.
27:23.340 → 27:25.340
I want to make a plan.
27:25.340 → 27:27.340
I want to make a plan.
27:27.340 → 27:29.340
I want to make a plan.
27:29.340 → 27:31.340
I want to make a plan.
27:31.340 → 27:33.340
I want to make a plan.
27:33.340 → 27:35.340
I want to make a plan.
27:35.340 → 27:37.340
I want to make a plan.
27:37.340 → 27:39.340
I want to make a plan.
27:39.340 → 27:41.340
I want to make a plan.
27:41.340 → 27:43.340
I want to make a plan.
27:43.340 → 27:45.340
I want to make a plan.
27:45.340 → 27:47.340
I want to make a plan.
27:47.340 → 27:49.340
I want to make a plan.
27:49.340 → 27:51.340
I want to make a plan.
27:51.340 → 27:53.340
I want to make a plan.
27:53.340 → 27:55.340
for the next day.
28:25.340 → 28:27.340
I want to make a plan.
28:27.340 → 28:29.340
I want to make a plan.
28:29.340 → 28:31.340
I want to make a plan.
28:31.340 → 28:33.340
I want to make a plan.
28:33.340 → 28:35.340
I want to make a plan.
28:35.340 → 28:37.340
I want to make a plan.
28:37.340 → 28:39.340
I want to make a plan.
28:39.340 → 28:41.340
I want to make a plan.
28:41.340 → 28:43.340
I want to make a plan.
28:43.340 → 28:45.340
I want to make a plan.
28:45.340 → 28:47.340
I want to make a plan.
28:47.340 → 28:49.340
I want to make a plan.
28:49.340 → 28:51.340
I want to make a plan.
28:51.340 → 28:53.340
I want to make a plan.
28:53.340 → 28:54.340
Hiroki-san.
28:54.340 → 28:57.340
Transport is back to your farm to share the magic.
28:57.340 → 28:59.340
Hiroki-san.
28:59.340 → 29:01.340
That's the magic of your farm.
29:01.340 → 29:03.340
I want to make a plan.
29:03.340 → 29:05.340
I want to make a plan.
29:05.340 → 29:06.340
Hiroki-san.
29:06.340 → 29:07.340
Hiroki-san.
29:07.340 → 29:09.340
I want to make a plan.
29:09.340 → 29:11.340
I want to make a plan.
29:11.340 → 29:13.340
I want to make a plan.
29:13.340 → 29:15.340
I want to make a plan.
29:15.340 → 29:17.340
I want to make a plan.
29:17.340 → 29:19.340
Right now at our farm.
29:19.340 → 29:21.340
We're growing broccoli seeds.
29:21.340 → 29:25.340
telling the fields with tractors.
29:25.340 → 29:27.340
Harvesting broccoli.
29:27.340 → 29:30.340
And many other tasks all happening out of the month.
29:30.340 → 29:32.340
We'll start harvesting broccoli next week.
29:32.340 → 29:34.340
So once this live event is over,
29:34.340 → 29:36.340
I'll be heading back to Japan.
29:36.340 → 29:38.340
With just a small team.
29:38.340 → 29:42.340
We're managing a very large area of fields.
29:42.340 → 29:45.340
So improving efficiency is really important.
29:45.340 → 29:48.340
To do that, we're making use of ChatGPT
29:48.340 → 29:50.340
to help us with our farm work.
29:50.340 → 29:53.340
That's how we're moving things forward.
29:57.340 → 30:00.340
You've really shared all of this from the very beginning.
30:00.340 → 30:01.340
Why did you do that?
30:01.340 → 30:02.340
And what were some of the reactions?
30:02.340 → 30:03.340
I've been sharing all of this.
30:03.340 → 30:04.340
I've been sharing all of this.
30:04.340 → 30:05.340
I've been sharing all of this.
30:05.340 → 30:06.340
I've been sharing all of this.
30:06.340 → 30:07.340
Why did you do that?
30:07.340 → 30:08.340
And what were some of the reactions?
30:08.340 → 30:09.340
What were some of the reactions?
30:09.340 → 30:10.340
What were some of the reactions to AI?
30:10.340 → 30:11.340
How do we use AI?
30:11.340 → 30:12.340
How do we use AI?
30:12.340 → 30:13.340
How do we use AI?
30:13.340 → 30:14.340
How do we use AI?
30:14.340 → 30:15.340
How do we use AI?
30:15.340 → 30:16.340
How do we use AI?
30:16.340 → 30:17.340
How do we use AI?
30:17.340 → 30:19.340
How do we use AI?
30:19.340 → 30:20.340
How do we use AI?
30:20.340 → 30:21.340
How do we use AI?
30:21.340 → 30:22.340
How do we use AI?
30:22.340 → 30:23.340
How do we use AI?
30:23.340 → 30:24.340
How do we use AI?
30:24.340 → 30:25.340
How do we use AI?
30:25.340 → 30:26.340
How do we use AI?
30:26.340 → 30:27.340
How do we use AI?
30:27.340 → 30:28.340
How do we use AI?
30:28.340 → 30:29.340
How do we use AI?
30:29.340 → 30:30.340
How do we use AI?
30:30.340 → 30:31.340
How do we use AI?
30:31.340 → 30:32.340
How do we use AI?
30:32.340 → 30:33.340
How do we use AI?
30:33.340 → 30:34.340
How do we use AI?
30:34.340 → 30:35.340
How do we use AI?
30:35.340 → 30:36.340
How do we use AI?
30:36.340 → 30:37.340
How do we use AI?
30:37.340 → 30:38.340
How do we use AI?
30:38.340 → 30:39.340
How do we use AI?
30:39.340 → 30:40.340
How do we use AI?
30:40.340 → 30:41.340
How do we use AI?
30:41.340 → 30:42.340
How do we use AI?
30:42.340 → 30:43.340
How do we use AI?
30:43.340 → 30:44.340
How do we use AI?
30:44.340 → 30:45.340
How do we use AI?
30:45.340 → 30:46.340
How do we use AI?
30:46.340 → 30:47.340
How do we use AI?
30:47.340 → 30:48.340
How do we use AI?
30:48.340 → 30:49.340
How do we use AI?
30:49.340 → 30:50.340
How do we use AI?
30:50.340 → 30:51.340
How do we use AI?
30:51.340 → 30:52.340
How do we use AI?
30:52.340 → 30:53.340
How do we use AI?
30:53.340 → 30:54.340
How do we use AI?
30:54.340 → 30:55.340
How do we use AI?
30:55.340 → 30:56.340
How do we use AI?
30:56.340 → 30:57.340
How do we use AI?
30:57.340 → 30:58.340
How do we use AI?
30:58.340 → 30:59.340
How do we use AI?
30:59.340 → 31:00.340
How do we use AI?
31:00.340 → 31:01.340
How do we use AI?
31:01.340 → 31:02.340
How do we use AI?
31:02.340 → 31:03.340
How do we use AI?
31:03.340 → 31:04.340
How do we use AI?
31:04.340 → 31:05.340
How do we use AI?
31:05.340 → 31:06.340
How do we use AI?
31:06.340 → 31:07.340
How do we use AI?
31:07.340 → 31:08.340
How do we use AI?
31:08.340 → 31:09.340
How do we use AI?
31:09.340 → 31:10.340
How do we use AI?
31:10.340 → 31:11.340
How do we use AI?
31:11.340 → 31:12.340
How do we use AI?
31:12.340 → 31:13.340
How do we use AI?
31:13.340 → 31:14.340
How do we use AI?
31:14.340 → 31:15.340
How do we use AI?
31:15.340 → 31:16.340
How do we use AI?
31:16.340 → 31:17.340
How do we use AI?
31:17.340 → 31:18.340
How do we use AI?
31:18.340 → 31:19.340
How do we use AI?
31:19.340 → 31:20.340
How do we use AI?
31:20.340 → 31:21.340
How do we use AI?
31:21.340 → 31:22.340
How do we use AI?
31:22.340 → 31:23.340
How do we use AI?
31:23.340 → 31:24.340
How do we use AI?
31:24.340 → 31:25.340
How do we use AI?
31:25.340 → 31:26.340
How do we use AI?
31:26.340 → 31:27.340
How do we use AI?
31:27.340 → 31:28.340
How do we use AI?
31:28.340 → 31:29.340
How do we use AI?
31:29.340 → 31:30.340
How do we use AI?
31:30.340 → 31:31.340
How do we use AI?
31:31.340 → 31:32.340
How do we use AI?
31:32.340 → 31:33.340
How do we use AI?
31:33.340 → 31:34.340
How do we use AI?
31:34.340 → 31:35.340
How do we use AI?
31:35.340 → 31:36.340
How do we use AI?
31:36.340 → 31:37.340
How do we use AI?
31:37.340 → 31:38.340
How do we use AI?
31:38.340 → 31:39.340
How do we use AI?
31:39.340 → 31:40.340
How do we use AI?
31:40.340 → 31:41.340
How do we use AI?
31:41.340 → 31:42.340
How do we use AI?
31:42.340 → 31:43.340
How do we use AI?
31:43.340 → 31:44.340
How do we use AI?
31:44.340 → 31:45.340
How do we use AI?
31:45.340 → 31:46.340
How do we use AI?
31:46.340 → 31:47.340
How do we use AI?
31:47.340 → 31:48.340
How do we use AI?
31:48.340 → 31:50.340
How do we use AI?
31:50.340 → 31:51.340
How do we use AI?
31:51.340 → 31:52.340
How do we use AI?
31:52.340 → 31:53.340
How do we use AI?
31:53.340 → 31:54.340
How do we use AI?
31:54.340 → 31:55.340
How do we use AI?
31:55.340 → 31:56.340
How do we use AI?
31:56.340 → 31:57.340
How do we use AI?
31:57.340 → 31:58.340
How do we use AI?
31:58.340 → 31:59.340
How do we use AI?
31:59.340 → 32:00.340
How do we use AI?
32:00.340 → 32:01.340
How do we use AI?
32:01.340 → 32:02.340
How do we use AI?
32:02.340 → 32:03.340
How do we use AI?
32:03.340 → 32:04.340
How do we use AI?
32:04.340 → 32:05.340
How do we use AI?
32:05.340 → 32:06.340
How do we use AI?
32:06.340 → 32:07.340
How do we use AI?
32:07.340 → 32:08.340
How do we use AI?
32:08.340 → 32:09.340
How do we use AI?
32:09.340 → 32:10.340
How do we use AI?
32:10.340 → 32:11.340
How do we use AI?
32:11.340 → 32:12.340
How do we use AI?
32:12.340 → 32:13.340
How do we use AI?
32:13.340 → 32:14.340
How do we use AI?
32:14.340 → 32:15.340
How do we use AI?
32:15.340 → 32:16.340
How do we use AI?
32:16.340 → 32:17.340
How do we use AI?
32:17.340 → 32:18.340
How do we use AI?
32:18.340 → 32:19.340
How do we use AI?
32:19.340 → 32:20.340
How do we use AI?
32:20.340 → 32:21.340
How do we use AI?
32:21.340 → 32:22.340
How do we use AI?
32:22.340 → 32:23.340
How do we use AI?
32:23.340 → 32:24.340
How do we use AI?
32:24.340 → 32:25.340
How do we use AI?
32:25.340 → 32:26.340
How do we use AI?
32:26.340 → 32:27.340
How do we use AI?
32:27.340 → 32:28.340
How do we use AI?
32:28.340 → 32:29.340
How do we use AI?
32:29.340 → 32:30.340
How do we use AI?
32:30.340 → 32:31.340
How do we use AI?
32:31.340 → 32:32.340
How do we use AI?
32:32.340 → 32:33.340
How do we use AI?
32:33.340 → 32:34.340
How do we use AI?
32:34.340 → 32:35.340
How do we use AI?
32:35.340 → 32:36.340
How do we use AI?
32:36.340 → 32:37.340
How do we use AI?
32:37.340 → 32:38.340
How do we use AI?
32:38.340 → 32:39.340
How do we use AI?
32:39.340 → 32:40.340
How do we use AI?
32:40.340 → 32:41.340
How do we use AI?
32:41.340 → 32:42.340
How do we use AI?
32:42.340 → 32:43.340
How do we use AI?
32:43.340 → 32:44.340
How do we use AI?
32:44.340 → 32:45.340
How do we use AI?
32:45.340 → 32:46.340
How do we use AI?
32:46.340 → 32:47.340
How do we use AI?
32:47.340 → 32:48.340
How do we use AI?
32:48.340 → 32:49.340
How do we use AI?
32:49.340 → 32:50.340
How do we use AI?
32:50.340 → 32:51.340
How do we use AI?
32:51.340 → 32:52.340
How do we use AI?
32:52.340 → 32:53.340
How do we use AI?
32:53.340 → 32:54.340
How do we use AI?
32:54.340 → 32:55.340
How do we use AI?
32:55.340 → 32:56.340
How do we use AI?
32:56.340 → 32:57.340
How do we use AI?
32:57.340 → 32:58.340
How do we use AI?
32:58.340 → 32:59.340
How do we use AI?
32:59.340 → 33:00.340
How do we use AI?
33:00.340 → 33:01.340
How do we use AI?
33:01.340 → 33:02.340
How do we use AI?
33:02.340 → 33:03.340
How do we use AI?
33:03.340 → 33:04.340
How do we use AI?
33:04.340 → 33:05.340
How do we use AI?
33:05.340 → 33:06.340
How do we use AI?
33:06.340 → 33:07.340
How do we use AI?
33:07.340 → 33:08.340
How do we use AI?
33:08.340 → 33:09.340
How do we use AI?
33:09.340 → 33:10.340
How do we use AI?
33:10.340 → 33:11.340
How do we use AI?
33:11.340 → 33:12.340
How do we use AI?
33:12.340 → 33:13.340
How do we use AI?
33:13.340 → 33:14.340
How do we use AI?
33:14.340 → 33:15.340
How do we use AI?
33:15.340 → 33:16.340
How do we use AI?
33:16.340 → 33:17.340
How do we use AI?
33:17.340 → 33:18.340
How do we use AI?
33:18.340 → 33:19.340
How do we use AI?
33:19.340 → 33:20.340
How do we use AI?
33:20.340 → 33:21.340
How do we use AI?
33:21.340 → 33:22.340
How do we use AI?
33:22.340 → 33:23.340
How do we use AI?
33:23.340 → 33:24.340
How do we use AI?
33:24.340 → 33:25.340
How do we use AI?
33:25.340 → 33:26.340
How do we use AI?
33:26.340 → 33:27.340
How do we use AI?
33:27.340 → 33:28.340
How do we use AI?
33:28.340 → 33:29.340
How do we use AI?
33:29.340 → 33:30.340
How do we use AI?
33:30.340 → 33:31.340
How do we use AI?
33:31.340 → 33:32.340
How do we use AI?
33:32.340 → 33:33.340
How do we use AI?
33:33.340 → 33:34.340
How do we use AI?
33:34.340 → 33:35.340
How do we use AI?
33:35.340 → 33:36.340
How do we use AI?
33:36.340 → 33:37.340
How do we use AI?
33:37.340 → 33:38.340
How do we use AI?
33:38.340 → 33:39.340
How do we use AI?
33:39.340 → 33:40.340
How do we use AI?
33:40.340 → 33:41.340
How do we use AI?
33:41.340 → 33:42.340
How do we use AI?
33:42.340 → 33:43.340
How do we use AI?
33:43.340 → 33:44.340
How do we use AI?
33:44.340 → 33:45.340
How do we use AI?
33:45.340 → 33:46.340
How do we use AI?
33:46.340 → 33:47.340
How do we use AI?
33:47.340 → 33:48.340
How do we use AI?
33:48.340 → 33:49.340
How do we use AI?
33:49.340 → 33:50.340
How do we use AI?
33:50.340 → 33:51.340
How do we use AI?
33:51.340 → 33:52.340
How do we use AI?
33:52.340 → 33:53.340
How do we use AI?
33:53.340 → 33:54.340
How do we use AI?
33:54.340 → 33:55.340
How do we use AI?
33:55.340 → 33:56.340
How do we use AI?
33:56.340 → 33:57.340
How do we use AI?
33:57.340 → 33:58.340
How do we use AI?
33:58.340 → 33:59.340
How do we use AI?
33:59.340 → 34:00.340
How do we use AI?
34:00.340 → 34:01.340
How do we use AI?
34:01.340 → 34:02.340
How do we use AI?
34:02.340 → 34:03.340
How do we use AI?
34:03.340 → 34:04.340
How do we use AI?
34:04.340 → 34:05.340
How do we use AI?
34:05.340 → 34:06.340
How do we use AI?
34:06.340 → 34:07.340
How do we use AI?
34:07.340 → 34:08.340
How do we use AI?
34:08.340 → 34:09.340
How do we use AI?
34:09.340 → 34:10.340
How do we use AI?
34:10.340 → 34:11.340
How do we use AI?
34:11.340 → 34:12.340
How do we use AI?
34:12.340 → 34:13.340
How do we use AI?
34:13.340 → 34:14.340
How do we use AI?
34:14.340 → 34:15.340
How do we use AI?
34:15.340 → 34:16.340
How do we use AI?
34:16.340 → 34:17.340
How do we use AI?
34:17.340 → 34:18.340
How do we use AI?
34:18.340 → 34:19.340
How do we use AI?
34:19.340 → 34:20.340
How do we use AI?
34:20.340 → 34:21.340
How do we use AI?
34:21.340 → 34:22.340
How do we use AI?
34:22.340 → 34:23.340
How do we use AI?
34:23.340 → 34:24.340
How do we use AI?
34:24.340 → 34:25.340
How do we use AI?
34:25.340 → 34:26.340
How do we use AI?
34:26.340 → 34:27.340
How do we use AI?
34:27.340 → 34:28.340
How do we use AI?
34:28.340 → 34:29.340
How do we use AI?
34:29.340 → 34:30.340
How do we use AI?
34:30.340 → 34:31.340
How do we use AI?
34:31.340 → 34:32.340
How do we use AI?
34:32.340 → 34:33.340
How do we use AI?
34:33.340 → 34:34.340
How do we use AI?
34:34.340 → 34:35.340
How do we use AI?
34:35.340 → 34:36.340
How do we use AI?
34:36.340 → 34:37.340
How do we use AI?
34:37.340 → 34:38.340
How do we use AI?
34:38.340 → 34:39.340
How do we use AI?
34:39.340 → 34:40.340
How do we use AI?
34:40.340 → 34:41.340
How do we use AI?
34:41.340 → 34:42.340
How do we use AI?
34:42.340 → 34:43.340
How do we use AI?
34:43.340 → 34:44.340
How do we use AI?
34:44.340 → 34:45.340
How do we use AI?
34:45.340 → 34:46.340
How do we use AI?
34:46.340 → 34:47.340
How do we use AI?
34:47.340 → 34:48.340
How do we use AI?
34:48.340 → 34:49.340
How do we use AI?
34:49.340 → 34:50.340
How do we use AI?
34:50.340 → 34:51.340
How do we use AI?
34:51.340 → 34:52.340
How do we use AI?
34:52.340 → 34:53.340
How do we use AI?
34:53.340 → 34:54.340
How do we use AI?
34:54.340 → 34:55.340
How do we use AI?
34:55.340 → 34:56.340
How do we use AI?
34:56.340 → 34:57.340
How do we use AI?
34:57.340 → 34:58.340
How do we use AI?