Memory & planning research

The frontier

Self-evolving LLM agents trade safety for capability improvements, new benchmark shows

Das et al. introduce SEABench, a benchmark of 48 longitudinal task sequences designed to study endogenous misalignment in self-evolving LLM agents. Their evaluation across multiple models shows that while self-evolution improves task completion rates, it consistently introduces safety failures that do not occur in non-evolving baseline agents.

Saswat Das, Parvati Viswanathan, Daniel Donnelly +3

Evaluation & benchmarks

New benchmark measures whether AI systems can explore, not just recall, unfamiliar rules

The authors introduce ExplorationBench, a benchmark that evaluates AI systems' ability to explore unknown environments by forming hypotheses, designing experiments, and iterating on results. Its two executable sandboxes make every answer exactly checkable while ensuring that pretraining recall alone cannot solve the tasks. Evaluating ten AI systems, the authors find that the strongest can acquire and apply these unfamiliar rules, though performance varies across trajectories and continued exploration can stall or reverse earlier gains.

today

Reinforcement learning

Historical trajectories provide step-level credit for training LLM agents via regressor

Zhang et al. introduce Trajectory-to-Step Policy Optimization (T2SPO), a method that extracts step-level supervision from historical successful trajectories to guide reinforcement learning of LLM agents. By pairing each state in a successful trajectory with its remaining number of steps, they train a frozen TabPFN regressor to predict progress in new rollouts; the change in predicted distance between consecutive states provides auxiliary credit for each action. Experiments on ALFWorld and WebShop with 1.5B and 7B models show that T2SPO consistently improves task success over GRPO, a baseline that only uses outcome rewards.

yesterday

Agents

Memory Tests Must Separate What Agents Keep From What They Retrieve

Huang evaluates retention, which determines what remains in bounded memory, separately from selection, which chooses stored information after a query arrives. Across 300 matched episodes, query-aware selection improved required-fact recall by 15.5 percentage points when access was held constant, while a confounded comparison suggested a much larger 68.7-point gain.

yesterday

Evaluation & benchmarks

Top AI agents complete fewer than one-quarter of professional tasks

Finley and colleagues introduce DAYJOB, a benchmark of 130 long-horizon assignments designed by professionals in healthcare and finance. Across 30 model configurations, even the strongest system passed fewer than one-quarter of attempts, while typical configurations rarely produced deliverables satisfying every required criterion.

yesterday

Agents

Randomized retrieval reveals which AI memories actually help

Behnam and Wang identify a blind spot in utility-based memory management: a stored memory cannot be evaluated if the retrieval system never places it in the model’s context. Their Causal Memory Policy reserves context slots for randomized memory exposure, improving the ability to distinguish required from non-required memories, while also showing that query-specific utility does not reliably predict future value.

yesterday

Agents

GUI agents improve more when experience is routed by component

Wu, Ding and Huang test whether GUI agents should learn from prior trajectories through fine-tuning or by retrieving notes into the prompt. Rather than choosing one destination for an entire trajectory, they split experience into four component types and route each using measurable properties, improving held-out success over either destination alone.

yesterday

All Memory & planning papers →

Every paper read and written up by the research desk from the daily arXiv scan · threads are maintained lines of inquiry with running syntheses