3 ms·

Is this the reverse of LLM? Ie, "what's the capital of France?" LLM picks from "Paris" 99.9% / "London" 0.001%, LLM then with some randomness output you "Paris

by gogoout 10d ago

Is this the reverse of LLM? Ie, "what's the capital of France?" LLM picks from "Paris" 99.9% / "London" 0.001%, LLM then with some randomness output you "Paris". For Jev, you ask it to give probability of a set of answers "what's the capital of France?" choose from answers (Paris/London), it then gives you (99.9%/0.001%)

aatd86 10d ago
99.90% ;D It basically gives you the softmax value instead of using it to generate the output token. (said loosely)
brooksy 10d ago
Essentially, although LLM token probabilities tend to be miscalibrated (mostly bc of posttraining). Jev is meant to be particularly calibrated