{"slug":"large-language","label":"Large Language","item_count":3,"day_count":3,"source_count":3,"first_seen":"2026-10-01T17:06:53+00:00","last_updated":"2026-10-16T14:26:44+00:00","generated_at":"2026-10-19T05:12:45.304178+00:00","sources":["arxiv_cs_ai","arxiv_cs_lg","arxiv_llm_reliability"],"days":[{"date":"2026-10-01","items":[{"title":"External Observers May See More Clearly: Cross-Model Span-Level Hallucination Detection in Large Language Models via Hidden State Probing","url":"http://arxiv.org/abs/2610.02066v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"As Large Language Models (LLMs) increasingly serve as foundational reasoning engines, their tendency to hallucinate remains a critical vulnerability. While recent internal state probes offer a promising alternative to...","why_it_matters":"Matches feed focus: eval.","sid":"2476b153d599e8b7","published":"2026-10-01T17:06:53+00:00","editor_note":"Hallucination detection by probing hidden states across models, at span level."}]},{"date":"2026-10-06","items":[{"title":"Secure Speculative Decoding for Large Language Models","url":"http://arxiv.org/abs/2610.08678v1","source":"arxiv_cs_lg","type":"paper","summary_1line":"Speculative decoding accelerates inference for a large language model (LLM), referred to as the \\emph{target model}, by first using a smaller model, referred to as the \\emph{draft model}, to generate candidate tokens...","sid":"c47efdba3284bb9f","published":"2026-10-06T16:59:37+00:00","editor_note":"Addresses security of speculative decoding, an inference-speedup technique unrelated to the other two items."}]},{"date":"2026-10-16","items":[{"title":"Trapped in Your Story: Evaluating Narrative-Induced Sycophancy in Large Language Models","url":"http://arxiv.org/abs/2610.23525v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Large language models (LLMs) are known to exhibit sycophancy by adjusting their responses toward users' expressed views. However, real users often justify their opinions through reasons, personal experiences, causal e...","why_it_matters":"Matches feed focus: evaluation.","sid":"ade0f523550f640a","published":"2026-10-16T14:26:44+00:00","editor_note":"The newest item: tests whether personal narratives, not just stated opinions, induce sycophancy."}]}],"editorial":{"tldr":"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).","stale":false,"whats_new":"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.","why_it_matters":"Each paper names a distinct failure surface: hallucination probing needs access to hidden states, speculative decoding adds a draft-model attack path, and sycophancy skews answers in user-facing agents. Treat them as three separate checks, not one trend.","take_for_builders":"Do not cite these three papers as corroborating each other; if you run user-facing agents, add a sycophancy case where the user justifies a wrong claim with a personal story, and judge the other two papers separately.","beats":[{"kicker":"RELIABILITY","tone":"launch","headline":"Cross-model span-level hallucination detection probes hidden states","sids":["2476b153d599e8b7"]},{"kicker":"SECURITY","tone":"neutral","headline":"Secure speculative decoding targets the draft-model path in LLM inference","sids":["c47efdba3284bb9f"]},{"kicker":"BEHAVIOR","tone":"now","headline":"Narrative-induced sycophancy evaluated in LLMs","sids":["ade0f523550f640a"]}],"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?"],"generated_at":"2026-10-19T05:12:44.417713+00:00"}}