External Observers May See More Clearly: Cross-Model Span-Level Hallucination Detection in Large Language Models via Hidden State Probing
Hallucination detection by probing hidden states across models, at span level.
AI Storyline
3 items · 3 sources · 3 days
Operational story trace
Large Language
Follow in this browser to see new updates on your Live feed.
Latest change
Oct 16 adds an arXiv study showing that LLMs shift toward a user's view when the user justifies it with a personal narrative, widening sycophancy testing beyond bare opinions.
Earlier contextThe story so far
This page groups three arXiv papers by the shared phrase "large language models" in their titles; they do not form one developing story. They cover hidden-state hallucination detection (Oct 1), a security attack on speculative decoding (Oct 6), and narrative-induced sycophancy (Oct 16).
editor-curated · source-linked
Arc
Oct 1Oct 16 · now
RELIABILITY · Oct 1Cross-model span-level hallucination detection probes hidden states1 source · show source ▾
SECURITY · Oct 6Secure speculative decoding targets the draft-model path in LLM inference1 source · show source ▾
Secure Speculative Decoding for Large Language Models
Addresses security of speculative decoding, an inference-speedup technique unrelated to the other two items.
BEHAVIOR · Oct 16Narrative-induced sycophancy evaluated in LLMs1 source · show source ▾
Trapped in Your Story: Evaluating Narrative-Induced Sycophancy in Large Language Models
The newest item: tests whether personal narratives, not just stated opinions, induce sycophancy.
Day 1 Thursday, Oct 1, 2026
External Observers May See More Clearly: Cross-Model Span-Level Hallucination Detection in Large Language Models via Hidden State Probing
Hallucination detection by probing hidden states across models, at span level.
Day 2 Tuesday, Oct 6, 2026
Secure Speculative Decoding for Large Language Models
Addresses security of speculative decoding, an inference-speedup technique unrelated to the other two items.
Day 3 Friday, Oct 16, 2026
Trapped in Your Story: Evaluating Narrative-Induced Sycophancy in Large Language Models
The newest item: tests whether personal narratives, not just stated opinions, induce sycophancy.
What to watch — open questions
- Do hidden-state hallucination probes transfer across model families without per-model training?
- Does the speculative-decoding defense add measurable latency over unprotected drafting?
How this thread was built
editor wrote the arc · 3 beats
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.