Source item

Sam Altman cooked super hard with Astra and looks like AGI is coming. Benchmark performance for Astra is a dramatic leap if you look at some of the before/afters: eg. science benchmark from 22.4% -> 64.6%, explotbench 78.5% -> 100%, SRE-bench 68.7% -> 99.2%. will spend some more time to look into it