Multi-turn RL improves LLM agent long-horizon performance
Agentic RL gains concentrate in error recovery and long-horizon tasks but require sufficient interaction budgets.
Stories per day · 30 days
2 RSS
Relateddistillation2image-editing2reinforcement-learning2self-forcing2acedit1asymmetric-conditioning1benchmarks1coding-agents1jailbreak-defense1llm-agents1llm-safety1rl1safety-alignment1swe-bench1user-simulation1
15 stories, newest first · one card per story
Stories tagged #multi-turn
ShallowDefend puts an explicit safety decision before multi-turn replies
Assess, Decide, Respond is trained with reinforcement learning and led multi-turn safety on 14 benchmarks across two model families.
SWE-DIVE evaluates coding agents in multi-turn sessions
A user simulator and conversation validator run inside Harbor to score whether the agent solves the task.
ACEdit: Stable Multi-Turn Image Editing via Asymmetric Conditioning
ACEdit stabilizes multi-turn image editing by distilling a teacher while the student edits its own outputs.
ACEdit uses asymmetric conditioning to stabilize multi-turn image edits
The student edits its own prior outputs while a teacher stays on the clean image with a compound instruction.
EAU co-evolves environment, agent, and user during RL
Authors report a 14.7–15.0% score gain over the strongest baseline and up to 36.7% more ColBench completions.
NVIDIA PivotOPD Teaches Multi-Turn AI Agents to Recover From Pivotal Mistakes
NVIDIA's PivotOPD trains multi-turn LLM agents to avoid and recover from early pivotal mistakes.
You Changed Your Mind, The Model Didn't: Demystifying Intent in Multi-Turn Dialogue
Intent-Eval shows rejected or replaced user changes can still derail multi-turn LLM task execution.
TRACE: Trajectory Return Attribution and Contrastive Erasure for Multi-Turn Safety
TRACE cut multi-turn attack success to the lowest rate in all 35 model-attack pairs with at most a 1.23-point utility drop.
Can Computation from Earlier Problems Help LLMs Solve New Ones?
STAIR reuses earlier response keys and values, lifting later-turn accuracy by up to 11.67 points on Qwen.
Does the Unsafe Gradient Survive a Conversation? On the Fragility of Gradient-Based Jailbreak Detection in Multi-Turn Dialogue
Gradient jailbreak detection that succeeds on synthetic chats fails on realistic multi-turn conversations.
On-Policy Self-Distillation for Multi-Turn Image Editing
MT-OPSD uses on-policy self-distillation so image editors stay stable across iterative multi-turn edits.
When Users Change Their Minds: Measuring and Repairing Intent Drift in LLM Agents
Researchers measure intent drift in LLM agents and show explicit state tracking only partly closes the gap.
Critical-State RL: Diagnosing Trainable States for Multi-Turn Tool Use
Critical-State RL identifies which multi-turn tool-use calls to train, improving missing-function task performance about 14 points on BFCL v4.
Measuring LLM Sycophancy under Sustained Multi-Turn Pressure
SPINE benchmark shows LLM sycophantic collapse rises with conversation length as an adaptive user pushes a mistaken position for up to 25 turns.