← All IntelClip / Other
Using hallucination probes as a proxy for model 'lostness'
From The State of Model Routing — NVIDIA, Cognition, OpenRouter · ≈38:21
“So, probes that work on either the internal state like internal state of the models directly.”
“So, that kind of gives you a proxy for how lost it is.”
“So, you can use like different kinds of probes to understand like the perplexity within a model.”
What’s in it
- Explains hallucination probes that peek at a model's internal state
- Covers linear probes and magnitude analysis as hallucination detectors
- Frames perplexity as a proxy for how 'lost' a model is
Clip transcript
know. >> So, uh just to add on that um um you have a lot of like these days there are a lot of hallucination probes. So, probes that work on either the internal state like internal state of the models directly. So, you can have some form of either magnitude analysis done or linear probes or just uh the end types of probes that you can see and you can essentially rate like how how much you think is is tending towards hallucination. Uh so, that kind of gives you a proxy for how lost it is. Uh like how lost a model is in its thinking. Uh so, you can use like different kinds of probes to understand like the perplexity within a model. >> Well, that's interesting. So, yeah,
Recommended reads
Comments
Sign in to comment.
Loading comments…