{"slug":"policy-optimization","label":"Policy Optimization","item_count":3,"day_count":3,"source_count":3,"first_seen":"2026-10-14T15:27:50+00:00","last_updated":"2026-10-23T06:08:17+00:00","generated_at":"2026-10-27T00:13:38.037892+00:00","sources":["arxiv_agent_systems_research","arxiv_cs_ai","arxiv_llm_reliability"],"days":[{"date":"2026-10-14","items":[{"title":"Lifted State Policy Optimization: Internalizing Self-Correction through Transition Geometry","url":"http://arxiv.org/abs/2610.20350v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Preference optimization ranks final answers and trajectory RL maximizes aggregate returns, yet neither credits the local change that turns a flawed reasoning state into a better one. We introduce Lifted State Policy O...","why_it_matters":"Matches feed focus: eval.","sid":"a42e86ea9f9f98f3","published":"2026-10-14T15:27:50+00:00","editor_note":"Introduces Lifted State Policy Optimization, crediting the local change from a flawed reasoning state to a better one."}]},{"date":"2026-10-22","items":[{"title":"ReflectPO: Reflection-Guided Policy Optimization for Long-Horizon LLM Agents","url":"http://arxiv.org/abs/2610.30933v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"With external tools, large language models (LLMs) can tackle complex tasks through sustained environmental interaction. LLM-based agents are therefore increasingly used to tackle complex long-horizon tasks in real-wor...","why_it_matters":"Matches feed focus: agent.","sid":"9cbcaf295333969d","published":"2026-10-22T08:54:24+00:00","editor_note":"Extends policy optimization with reflection guidance for long-horizon, tool-using LLM agents."}]},{"date":"2026-10-23","items":[{"title":"SOPO: Socratic Guided Policy Optimization for Span-Level Hallucination Detection","url":"http://arxiv.org/abs/2610.32212v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Large Language Models (LLMs) have gained widespread adoption in various natural language processing tasks, but suffer from hallucination issues where they generate unfaithful or inconsistent content. While Reinforceme...","why_it_matters":"Matches feed focus: evaluation.","sid":"78e0961a0062764b","published":"2026-10-23T06:08:17+00:00","editor_note":"Uses Socratic guided policy optimization to detect hallucinated spans rather than whole answers."}]}],"editorial":{"tldr":"On Oct 14 a paper proposed Lifted State Policy Optimization to credit the local change that turns a flawed reasoning state into a better one. On Oct 22 ReflectPO applied reflection-guided policy optimization to long-horizon LLM agents.","stale":false,"whats_new":"SOPO (Oct 23) applies Socratic guided policy optimization to span-level hallucination detection in LLM output.","why_it_matters":"Three arXiv papers in ten days tailor policy optimization to self-correction, agent horizons and hallucination detection; all are preprints, so treat them as research signals, not production tooling.","take_for_builders":"Treat these as preprints: wait for released code and independent replication before changing any training or eval pipeline.","beats":[{"kicker":"SELF-CORRECTION","tone":"launch","headline":"Lifted State Policy Optimization targets self-correction","sids":["a42e86ea9f9f98f3"]},{"kicker":"AGENTS","tone":"rising","headline":"ReflectPO brings reflection-guided optimization to long-horizon agents","sids":["9cbcaf295333969d"]},{"kicker":"NOW","tone":"now","headline":"SOPO aims policy optimization at span-level hallucination detection","sids":["78e0961a0062764b"]}],"open_questions":["Do any of these three methods publish code or reproducible results on public benchmarks?"],"generated_at":"2026-10-27T00:13:24+00:00"}}