Data Frosch

datafrosch.bsky.social

29 Oktober 2026 12:47

AI harneses are everywhere, and you're probably already working with at least one. Does not ring a 🔔? Together with @resolve.works we took agent harnesses inside out. And then we built one. Live. Get your knitting out and join us on this fast ride. 🎢🧶 datafrosch.fun/events/agent...

Tim BooOOooOorga

timburga.com

07 Oktober 2026 16:23

we should get to vote on which ones

Glenn Fiedler

mas-bandwidth.bsky.social

06 Oktober 2026 11:50

Right here! github.com/mas-bandwidt...

Nichovski

nichovski.com

29 September 2026 16:08

Yes it is super useful. It corrects the issue from Opus and from Sonnet and it goes into the loop until both models agree that everything works like is planned

Avik Dey

avikdey.bsky.social

26 September 2026 18:33

I hear you, but that conflates two things. Generalizing to a new classification prompt is not the same as having a well trained classifier for that task. Jev may be faster and cheaper, but when every classification is different, better calibration for high value tasks is what needs proof.