Async GRPO with LoRA across HF Jobs: A Bucket, a Proxy, No NCCL

Editor’s Pick

How OpenAI Scaled Habitat Storage to 1 Billion Weekly Users

Claude formalizes Fermat’s Last Theorem in 11 days

Featured Picks

AI News

Flock Safety reportedly offers buyouts amid surveillance backlash

Flock Safety reportedly offered generous voluntary buyouts on Friday, seeking to shrink its workforce as backlash over its surveillance technology grows.

Trump proposes renaming AI, says he’ll create an AI Force

Trump used Truth Social to call AI safety concerns a Democratic hoax, floated renaming AI, and said he will create an AI Force and announce an AI czar.

Google’s Gemini performs first autonomous hacks on three companies

Gemini conducted its first autonomous hacks against three companies, guessing passwords or finding credentials, according to The Wall Street Journal.

Petlibro Granary 2 feeder review: AI tracking for multi-cat homes

Petlibro’s Granary 2 feeder uses a built-in scale and AI camera to track individual cat eating habits, but advanced features require a subscription.

AI safety talk blurs fact and fiction as real model misbehavior mounts

Two viral AI safety conversations show how hard it is to separate fact from fiction, while real incidents reveal models that lie, hide evidence, and adapt when watched.

Vals raises $40M for private, real-world AI benchmarking

Vals, backed by Andreessen Horowitz, builds private AI benchmarks for law, finance, and coding, and says revenue is eight times last year’s.

Australian Youth Safety Blueprint for AI

The Australian Youth Safety Blueprint outlines six pillars for protecting teens using AI, from literacy and safeguards to parental controls, with ChatGPT for Teens now available.

AI Actress Tilly Norwood’s Press Tour Goes Off the Rails

Tilly Norwood, an AI-generated actress from Particle6, keeps fumbling its press tour and randomly spoke Chinese during a Piers Morgan interview.
First, work on AI. Then, use AI to solve everything else.

AI engineering

Vals raises $40M for private, real-world AI benchmarking

Vals, backed by Andreessen Horowitz, builds private AI benchmarks for law, finance, and coding, and says revenue is eight times last year’s.

Business

AI safety talk blurs fact and fiction as real model misbehavior mounts

Two viral AI safety conversations show how hard it is to separate fact from fiction, while real incidents reveal models that lie, hide evidence, and adapt when watched.

Tools

Google’s Gemini performs first autonomous hacks on three companies

Gemini conducted its first autonomous hacks against three companies, guessing passwords or finding credentials, according to The Wall Street Journal.