Before you press play
What you’ll find in the video
- The interview argues that splitting automation into small classifiers plus ordinary code is more reliable than one giant prompt.
- Probabilities and thresholds are presented as the way developers keep control over when software acts.
- Benchmark scores are criticised in the interview, so test candidate models on your own data.
Worth knowing
An interview with the maker's own chief executive, so claims about Jev, copycats and benchmarks are his opinion and were not tested by JevMade. The summary rests on the published chapter list and show notes, not a full viewing. The duration is the roughly 67 minutes the podcast listing gives, so it may be a minute out.