DeepMind's proof agent lands in Science with 9 Erdős solves
Google DeepMind's AlphaProof Nexus is now peer-reviewed in Science: 9 of 353 open Erdős problems solved and 44 of 492 OEIS conjectures proved, with caveats from Spanish scholars.
Page 4
View archive →Nadella calls for an emergency brake on AI models
Microsoft CEO Satya Nadella laid out five principles for AI trust architecture, from separating model and harness to assuming the model is already compromised.
Cloudflare's Clef-omni Decision Model Now Handles Video and Audio
Cloudflare's new Clef-omni scores audio and full video input in a single pass: about 1.5 seconds for a 21-second clip, at $0.15 per million input tokens.
Ten minutes of AI help makes people quicker to quit once it's gone
A UC Berkeley study finds that after about 10 minutes of ChatGPT-assisted problem solving, participants scored lower and skipped more questions once the tool was taken away.
Claude-built Adobe clones now target Microsoft Office
Brandon Thomas's Rust-based ArtCraft series has grown from seven Adobe-style apps to eleven, adding open-source stand-ins for Word, Excel, PowerPoint and AutoCAD.
Codex raises Sol's default context to 500K tokens
Codex now defaults GPT-6.1 Sol to a 500K-token context, up from 272K, to cut down on context compaction. The old default matched the API's long-context surcharge threshold.
Grok Bot Gets Its Own Email Inbox
xAI's Grok Bot began assigning users dedicated mail.grokbot.com inboxes on October 9. Team accounts need admin approval, and many details remain unpublished.
ARC-AGI-2 Kaggle leader hits 88.06%
TUFA Labs tops the 2026 ARC-AGI-2 leaderboard at 88.06%, versus last season's 24.03% private-set win. The two figures are measured differently.
AI decompilation moves old games into the browser overnight
Hundreds of AI-decompiled browser ports of classic games have appeared in two to three weeks, and rights holders have yet to say how they will respond.
Redwood uses distillation to audit models; confession hits 84%
Redwood Research's new paper distills untrusted models into same-base students to make them confess hidden traits, lifting one trait's confession rate from 22% to 84%.