TypeSafe's Allie Laabs discusses bounded decisions, combining Jev with language models, and asking several questions about the same input in one request.
Original by CodeRabbitGetting startedOverview31 min 27 secPublished
Look for places where software needs a category, rating or yes-or-no judgment rather than a written answer.
Keep ordinary rules for decisions code can already make exactly, and reserve learned judgment for questions about meaning.
Group independent questions about unchanged input, even when some answers are used only on certain paths. Questions that need a previous result or new state still have to wait.
Worth knowing
Vendor interview, not comparative testing. Examples and performance claims are the guest's account, not reproduced measurements. The discussion of batching does not make genuinely dependent questions independent.