·
Finding signal on Twitter is more difficult than it used to be. We curate the best tweets on topics like AI, startups, and product development every weekday so you can focus on what matters.
PLUS: OpenAI nears proof of Hodge Conjecture
Anthropic says Claude now leads 26% of its AI research and development, up from less than 1% in February.
OpenAI researchers expect to finish a proof of the Hodge Conjecture soon, one of the remaining Millennium Prize problems.
Claude Projects lets you describe work and spin off parallel threads that keep running after you close your laptop.
Vercel's AI Gateway data shows open-weight models took 56% of token volume for the first time in August.
Astra took a third of OpenAI spend within 48 hours and twice Fable 5.1's share at launch.
Anthropics says only 6% of AI R&D compute currently goes to safety work.
Z.ai says GLM-5.3's Infra Agent built production inference stack in under two weeks, calling it recursive self-improvement.
Top OpenAI researchers are spending $7-8k daily on Codex, a run rate of $2-3M per year.
Trade data shows more than $3B of chips smuggled into China via Malaysia between April 2024 and June 2025.
Nvidia's Jensen Huang bans political discourse at work but takes presidential calls on speakerphone for the whole room.
Three Hacktron AI researchers used Claude Opus 5 to break into OpenAI's GitHub monorepo through an employee's Codex account.
GPT-6 Astra cracked a previously unsolved 1941 German Enigma message in about 10 hours autonomously.
Musk, Zuckerberg, and Huang all opposed a FINRA-style AI industry body that Demis Hassabis pitched to Trump.
NVIDIA researchers released Agora, a shared memory system for research agents that ran 13 workers for 12 days without tasks.
Ole Lehmann says you can start a hardware company for $250 using Astra, JLC3DP, and an ESP32 chip.
Stripe launched Directory, an improving index of agent-buyable businesses that increases agentic conversion rate and reduces tokens.
Facebook admitted in federal court that only 17% of time on Facebook goes to friends' content, down from 22% in 2023.
OpenAI's Noam Brown says recursive self-improvement is the priority for new AI models by a pretty wide margin.
AnthropicAI says Claude optimized inference for 30 open-source models, making them 4x faster on average with custom GPU software.
World Labs released Atlas, which generates real-time flythrough from 32 images with pixel-perfect camera control.
Cognition gave Devin access to a Ramp card and told it to make money; it made $75 through cold outreach.
Notion launched Skills API, letting teams edit skills in Notion and use them in ChatGPT, Claude, and all agents.
Epoch AI launched Benchmark Reviews, auditing 15 AI benchmarks with 4 verified, 9 flawed, and 2 inconclusive.