by the way, yesterday Hamel gave a huge shoutout to Phoenix by Arize. what do you use?

I feel the point of pytest is to spot regressions in properties you can assert deterministically (e.g. length, or that the model does mention some things) while evals are for vibe checks and further refine the dataset for future fine tuning / few shots.

I tried adding few shots and it does help a lot. Without them, it uses `groupby` everywhere. sigh.

Incredibly powerful. Thanks for sharing.

did you see @marimo.io ?

Uh glad to find you here prof! Amazing starter pack, needed that.

wut is the permaban thing real? that's absurd.

Here we are! 🐍

Freshly made! @pydata-milan.bsky.social

@pietroppeter.bsky.social we should add python/pydata milano!

uh is there a link for this talk? :)

So can I replace watchdog with watchfiles when working with, say, streamlit?

Uuuh I see @gi0baro.bsky.social contributing there.

will definitely run some experiments with it!

also perhaps @tomaarsen.bsky.social