Aisha KhanWrite test cases once, run them against every prompt change and model upgrade, and see which answers got worse before your users do.
Write test cases once, run them against every prompt change and model upgrade, and see which answers got worse before your users do.
Discussion 5
Tried it on a real case from last week and it got 90% of the way there. The last 10% is where I'd still want a human, which you already handle.
Which model are you running under the hood? The latency is great.
How do you deal with hallucinations? Is there a way to see which source each answer came from?
Pricing per seat or per usage? Small teams care a lot about predictable bills.
Yes! Just added it, refresh and you should see it.