#detector
4 discoveries · Hand-picked discoveries from the open web, newest first.
-
I Made Qwen a Medieval Scholar. The Lie Detector Believed It Too.
Why should you care? If you… What this means for you Build lie detectors or deception probes Your probe can read a model’s belief faithfully and still not be what the answer comes from. Held-out accuracy will not show this; ablating, steering, and patching the probe direction will, in minutes per model. Monitor deployed models that run with system prompts or personas A persona rewrites the representation your probe reads, so your probe would agree with the persona’s false answers instead of flagging them. Steer models toward honesty (representation engineering) Pushing along the truth directio
LessWrong · EN
-
Show HN: Deterministic prompt injection detector using formal logic
This is a prompt-injection detector I built using NLP and SMT2. Comments URL: https://news.ycombinator.com/item?id=50501132 Points: 1 # Comments: 0
Show HN · EN
-
Show HN: Indic-Turn-Detector – open-weights end-of-turn model for voice agents
Text Classification model with 0.5B parameters. Download, explore files and run it on Hugging Face.
Show HN · EN
-
Show HN: I used Jev to build an off-topic detector for meetings
A fun demo exploring how faster, cheaper model judgments can help tell when a meeting is going off track.
Show HN · EN