→ChatGPT lands in Word; OpenAI says Excel and PowerPoint usage has surged recently
ChatGPT is now built into Word: it can turn rough notes into a draft, rephrase paragraphs, proofread, suggest edits, and catch formatting issues. OpenAI's Sherwin Wu says Excel and PowerPoint usage has spiked recently, and adding Word completes the Office suite integration. The post doesn't disclose launch date, pricing, or feature limits.
#OpenAI#Microsoft#Sherwin Wu
editor take
ChatGPT now lives inside Word, completing the Office suite—but no launch date or pricing yet.
FEATUREDAI HOT (Curated Pool)· aihot-apiZH20:49 · 09·17
→Anthropic shares three internal metrics to track how fast AI is building AI
Anthropic published a measurement framework and an internal snapshot to give the public visibility into the pace of frontier AI development. The headline number: Claude now leads 26% of Anthropic's AI R&D tasks, up from under 1% in February 2026. Two other metrics track oversight of AI agents and compute allocation. Anthropic plans to embed independent third-party evaluators to verify the data, but cross-lab comparison still lacks a common methodology.
#Anthropic#Claude#Epoch AI
why featured
Featured · importance 88 · hook + knowledge + resonance
editor take
Anthropic measures itself: Claude now leads 26% of its AI R&D, up from <1% in Feb, possibly hitting 80% by year-end.
sharp
This is worth opening because Anthropic put a number on 'AI building AI' for the first time. They worked with Epoch AI on an automation scale from AL0 (all human) to AL5 (fully autonomous). Claude sits at AL4 — 'AI leads, human supervises.' The 26% figure looks dramatic, but don't panic yet. It measures task leadership, not full replacement, and it's self-reported data with no third-party verification in place yet. Anthropic says independent evaluators are coming, but there's no cross-lab methodology for comparison. I'd read this as Anthropic getting ahead of regulatory pressure — the direction matters, but I'd discount the absolute numbers until someone outside checks them.
FEATUREDAI HOT (Curated Pool)· aihot-apiZH17:52 · 09·17
→Claude redesigns Projects from a folder into a hosted, multi-threaded conversational project
Anthropic overhauled Claude's Projects: it's no longer a folder of chats, but a hosted project space that supports multiple parallel conversation threads with cross-thread context. You can group related conversations into one Project, and Claude remembers context across threads. Team plans get shared projects with permission controls. The post doesn't spell out free-tier project limits or max threads per project.
#Anthropic#Claude#Product update
why featured
Featured · importance 78 · hook + knowledge + resonance
editor take
Anthropic turned Claude Projects from a chat folder into a multi-threaded workspace with cross-thread memory; team plans get shared projects.
sharp
This is worth a look because it changes how you work, not the model itself. Old Projects was basically a chat folder. Now you can run multiple parallel threads inside one project, and Claude remembers context across them. Team plans add shared projects with permissions — it turns scattered one-off chats into a persistent workspace with memory. The post doesn't spell out free-tier limits on project count or max threads per project, and there's no detail on how much cross-thread context it actually holds. I'd discount it a bit until real users test the limits.
● P1AI HOT (Curated Pool)· aihot-apiZH15:38 · 09·17
→Noam Brown on 10,000-agent swarms solving math problems and recursive self-improvement
Noam Brown, a core contributor to OpenAI's o1 reasoning models, now works on multi-agent systems. His team just solved a Millennium Prize Problem using 10,000 agents, 130 billion tokens, and 88 hours of compute. Brown frames multi-agent as parallel test-time compute: a single agent hits a latency wall, so you throw more agents at the problem to go faster, at the cost of some efficiency. In the 5.6 release's Ultra Mode, 4 agents cut solve time in half; 16 agents push it further, especially on parallel-friendly tasks like math. The conversation also covers what math progress signals for recursive self-improvement, degrading chain-of-thought quality, and how to verify alignment before kicking off RSI.
#Reasoning#Agent#Noam Brown#OpenAI
why featured
Featured · importance 98 · hook + knowledge + resonance
editor take
Noam Brown reveals a 10,000-agent system for math, but the model isn't public and details are all from a podcast — treat this as a directional signal, not a product launch.
sharp
Two sources covered this, but both trace back to a single Dwarkesh podcast episode — no blog post, no paper, no public demo. Noam Brown says OpenAI used 10,000 agents running for 88 hours and burning 130 billion tokens to solve a Millennium Prize math problem. All numbers come from his spoken remarks, so there's no way to cross-check.
The logic he lays out: reasoning models get better the longer they think, but serial latency becomes unbearable. Parallelizing across many agents trades some efficiency for speed, and math problems happen to be highly parallelizable. The idea isn't new, but the scale is — this is the first time anyone from a major lab has talked about running 10,000 agents on a single hard problem.
I'd discount this a bit for now. No pricing was mentioned, and it's unclear whether 88 hours is wall-clock time or GPU time. He didn't specify which Millennium Problem was solved or what verification looked like. What's solid: OpenAI is betting heavily on multi-agent as the next scaling axis. What's missing: any signal on when this becomes a product rather than a research flex.
→Unsloth ships Docker image and desktop app to train & run 500+ models locally
Unsloth released a Docker image and Unsloth Desktop to train and run 500+ models locally with zero setup. It includes a new GUI and notebook workflows, supporting both NVIDIA and AMD GPUs. The post doesn't disclose specific performance numbers or the full model list, but the install guide is live.
#Unsloth
editor take
Unsloth shipped a Docker image and desktop app for zero-setup local training of 500+ models, but no performance numbers yet — I'd hold off on the hype.
FEATUREDAI HOT (Curated Pool)· aihot-apiZH00:26 · 09·17
→GitHub used Copilot agents to migrate the Copilot runtime from TypeScript to 830K lines of Rust
GitHub engineers used Copilot's agent mode to migrate the core Copilot runtime from TypeScript to Rust, producing roughly 830K lines of code. The migration ran in three phases: file-by-file translation by agents, test-driven bug fixing, and performance/security review. After migration, service startup dropped from 30s to 3s, memory usage fell to one-third, and time-to-first-token went from 11s to 1.2s. The team stresses that humans stayed in the loop—agents did the heavy lifting, engineers owned architecture, code review, and test coverage. The post doesn't spell out exact cost savings but states the migration was done 'with a smaller team in less time.'
#Code#GitHub#GitHub Copilot
why featured
Featured · importance 82 · hook + knowledge + resonance
editor take
GitHub used Copilot's agent mode to migrate its core runtime from TypeScript to 830K lines of Rust, cutting startup from 30s to 3s.
sharp
The numbers here are hard to ignore: startup from 30s to 3s, memory down to one-third, time-to-first-token from 11s to 1.2s. GitHub laid out the three-phase process clearly—file-by-file translation by agents, test-driven bug fixing, then human-led architecture and security review.
I'd discount this a bit: it's GitHub eating its own dogfood with a codebase and tool they know inside out. An outside team might not hit the same efficiency. The post says it was done "with a smaller team in less time" but doesn't give headcount or dollar figures.
The useful bit isn't "AI replaces devs"—it's the division of labor. Agents handled the grunt work of translation and bug fixing; engineers owned architecture, test coverage, and final sign-off. 830K lines of Rust is a real project, and the fact that it shipped says agent mode is already a serious tool for large-scale refactors.
→Use MCP Plugin to Offload Codex Planning to GPT-6 Pro, Save Pro Weekly Quota
The article body is blocked by WeChat, only the title remains. It describes using an MCP plugin to let GPT-6 Pro take over Codex planning tasks to save Pro weekly quota. No details on setup, savings, or effectiveness are disclosed.
#GPT-6 Pro#Codex#MCP
editor take
Body blocked by WeChat. Title only: use MCP plugin to offload Codex planning to GPT-6 Pro to save weekly quota. No details on setup, savings, or effectiveness.