WTF Is Going On / Weights, Tools & Frameworks

How enabling two settings tripled our scores on the ARC-AGI-3 benchmark

OpenAI announcement yesterday
read it at the source — OpenAI →
HN · 33 points · 5 comments @sungkim.bsky.social · 21 likes @techmeme.com · 5 likes
@timkellogg.me “GPT-5.6 is SOTA on ARC-AGI-3 when they ran the benchmark, they misconfigured the model, dropping reasoning blocks after compaction. it’s properly configured in codex & chatgpt openai.com/index/how-tw...” 34 likes

Also on the board

UK lawmaker suing Musk’s xAI seeks order to stop Grok generating sexualised images

9 outlets 6h
Elon Musk's xAI sues Minnesota over law to ban 'nudify' apps

9 outlets yesterday
The utter utopian unwisdom of AI bros

2 outlets 3h
Google DeepMind Disbands Its Nobel-Prize Winning AlphaFold Team

2 outlets 4h
Meta’s AI Spendings Are Burning Almost All of Its Cash Flow

2 outlets 9h