OpenAI wants humans to check its science
OpenAI handed its science results to outside experts and asked them to do the hard part. The proofs are only useful if someone can check and explain them. If you sell research tools, that is the lesson to take today. Build the verification step into your product before a customer asks for it.
- OpenAI asks outside experts to verify its science
- China's AI stocks fall about 40 percent
- Strawberry's browser agent leaves beta
- Quick hits: Microsoft, SurgePix, n8n, G42, Verizon
- AI Index and new members
Top stories
All news →What happened: OpenAI says its models produced three research results and is asking outside experts to turn them into understanding.
- The outputs include a Lean-formalized disproof of the Exponential Time Hypothesis for 3-SAT, a proof of the local smoothing conjecture, and a physics result containing a proof of pure Yang-Mills existence.
- Outside experts work under no-NDA terms, participation is voluntary and unpaid, and OpenAI claims no academic credit.
- Three release strategies were proposed, and OpenAI admits it does not yet know which will work best.
Why it matters: A lab is treating model output as a dataset and asking humans to supply interpretation and recognition, a model other AI-for-science startups may copy. Formal verification is what made these claims checkable, so teams selling research tools should plan for it.
What happened: China's AI and chip stocks have fallen sharply, and traders blamed high valuations, tax burdens and the risk of US limits on Chinese data center equipment.
- The STAR50 index closed at 1,315.50 on October 26, down 2.23% that session, according to Reuters.
- AI-related stocks make up 42% of China's total market capitalization, based on Goldman Sachs figures.
- Optical component makers Zhongji Innolight and Eoptolink are down about 60% from their highs, and the CSI300 sits about 20% below its June peak.
Why it matters: Capital for Chinese AI companies is getting tighter, and one investment chief expects listing plans to stall for many firms. Teams that depend on Chinese hardware or listings should watch export-curb risk closely.
What happened: Strawberry released version 1.0 of its AI browser, which can run agents on schedules and triggers without a prompt.
- The release adds scheduled runs and triggers tied to new emails or leads, and large research jobs are split across parallel sub-agents.
- Meetings on Zoom, Microsoft Teams and Google Meet are recorded locally, and agents pause for human approval before sending emails or changing records.
- Strawberry was founded in 2023 and announced a $6 million raise led by General Catalyst and EQT Ventures, with a Series A planned.
Why it matters: The agent works inside the browser, where logins and team context already sit, so distribution is the real edge. Investors should watch whether the data sovereignty pitch turns into paying users.
Quick hits
- Microsoft is offering buyouts to about 60,000 staff, which may free up talent for startups. TechCrunch →
- SurgePix lets AI agents edit images one layer at a time, with skills on a public GitHub repo. Product Hunt →
- A new no-code tool builds branded chat widgets for n8n, selling one-time licenses, not subscriptions. Product Hunt →
- G42 may build an AI data center in Greece, but location, capacity and pricing are still unset. Product Hunt →
- Verizon signed a $2 billion-plus AI infrastructure deal with AWS and raised its subscriber forecast. Reuters →


