Post

Also "we have the greatest model of all time but we're not going to do a live stream about it"

there's a blog post by openai detailing why the default harness is bad, you can search for it.

:shrugs: if ed zitron can how can i say no

Fresh off the press: arcprize.org/blog/astra "GPT-6 Astra scores 62.7% for $26K on ARC-AGI-3 Semi-Private with our Standard Harness, and 99.9% for $19K with a Provider Adapter Harness."

Yea, they’re doing their blog post thing of testing on ARC-AGI-3 with a different endpoint and retaining reasoning and using their internal compaction. I think that’s fair but not when comparing to all the other models still being nerfed with those issues

WE DON’T

Wtf gemini 3.8 flash is cooking on deep swe lol

😵Lord Ψ says that, recycled information can never be conscious.😵