Good post. I’ve been thinking about doing offline testing of LLM tasks a bit these days and have come to the conclusion that old school testing is the best until more mature features can be developed. Specifically, I mean running a power analysis to determine sample size, random sampling based on that and then running tests like a z test to see if there is a difference and between what bounds. Tests are expensive and I wish there was a better way for realizable offline evals.
Good post. I’ve been thinking about doing offline testing of LLM tasks a bit these days and have come to the conclusion that old school testing is the best until more mature features can be developed. Specifically, I mean running a power analysis to determine sample size, random sampling based on that and then running tests like a z test to see if there is a difference and between what bounds. Tests are expensive and I wish there was a better way for realizable offline evals.
Have you seen LLM testing tools like promptfoo?
Yes, I have seen it and BrainTrust too. Unfortunately, need FOSS without vendor at scale.
i - luckily - passed my statistics exam this summer, it's however cool to visualize what's happening
[flagged]
- Brand new burner account
- upset about “AI slop” (the image is clearly not AI)
- mentions tech buzzwords that annoy you
- claiming the article is not as rigorous as an academic paper
Perhaps I’m just old school. But I miss the HN where the best way to get upvotes was to be insightful and not to send low-effort snarky replies
One of those points is not like the other, Marimo’s feature to deploy a notebook as WASM is a very nice feature imo.
Clearly marked as unsplash
[flagged]