techcam

6 ms·

1. 1.▲
   by techcam 6mo ago

   Happy to explain how the scoring works since that’s the obvious first question. The core idea is: Safety Score = 100 − riskScore The risk score is based on structural prompt properties that tend to correlate with failures in production syst
2. 2.▲
   Show HN: CLI to score AI prompts after a prod failure(costguardai.io)

   1 points by techcam 6mo ago|1 comments
3. 3.▲
   by techcam 6mo ago

   The tricky part is that prompts can look “correct” but still behave unpredictably depending on phrasing.
4. 4.▲
   by techcam 6mo ago

   We ran into something similar with API costs — small changes in behavior can have surprisingly large downstream effects.
5. 5.▲
   by techcam 6mo ago

   This resonates — most of the hard problems show up after you ship, not before.
6. 6.▲
   by techcam 6mo ago

   Feels like we have great tooling for code, but prompts are still mostly trial-and-error. Curious how people are validating them today.
7. 7.▲
   by techcam 6mo ago

   I’ve been noticing the same — a lot of failures aren’t obvious “jailbreaks,” they’re just subtle prompt structure issues that only show up in production.