Claims Browser

Large language model agents in governed organizations require persona (instructions, tone, self-presentation) to evolve freely while keeping execution (stateful, audited work) traceable, which cannot be satisfied cheaply in a single trust domain.

"Large language model (LLM) agents in governed organizations must let the persona (instructions, tone, self-presentation) evolve freely, while keeping execution (stateful, audited work) traceable. A single trust domain does not satisfy both cheaply."

Persona-Execution Separation (PES) architecture, where persona and execution reside in different trust domains connected by a governed contract bridge, solves the dual requirements of free persona drift and execution traceability.

"We present Persona-Execution Separation (PES): persona and execution reside in different trust domains, connected by a governed contract bridge. The persona is singly-homed and may drift; execution is faceless and audited."

Under LLM representational indistinguishability, any single-domain mechanism meeting free drift, execution traceability, and decoupling must re-introduce typed change objects, an external gate, and a stable audit anchor, essentially rebuilding PES at higher coupling cost.

"Under LLM representational indistinguishability, any single-domain mechanism that meets all three must re-introduce typed change objects, an external gate, and a stable audit anchor: PES rebuilt at higher coupling cost."

NPO achieves comparable or better performance than GEPA with fewer rollouts

"NPO achieves comparable or better performance than GEPA with fewer rollouts"

Prompt optimization can deliver performance gains comparable to fine-tuning model weights while reducing computational costs

"prompt optimization emerging as a promising approach capable of delivering performance gains comparable to those achieved by fine-tuning model weights, while reducing computational costs in both optimization and serving"

Recent prompt optimizer developments favor unnecessarily complex approaches

"recent developments increasingly favor unnecessarily complex prompt optimizers"

Stronger teacher reasoning can partially substitute for optimizer-side search complexity in prompt optimization

"its advantage increases with stronger teacher models, suggesting that stronger teacher reasoning can partially substitute for optimizer-side search complexity"

NPO-optimized prompts transfer well to other student models, especially within the same model family

"NPO-optimized prompts elicit similar performance improvements when applied verbatim to other student models, especially across models within the same family"

Simple, linear prompt optimization can rival substantially more sophisticated and complex search procedures

"simple, linear prompt optimization can rival substantially more sophisticated and complex search procedures"

Efficiently improving autonomous agents across diverse tasks is central to accelerating recursive self-improvement in agentic AI

"Efficiently improving autonomous agents across diverse tasks is central to accelerating recursive self-improvement (RSI) in agentic AI"

GPT-5.6-sol matches or outperforms the best existing methods on almost all evaluated instances of inventory control, queueing network control, and assortment optimization problems

"The strongest model we test, gpt-5.6-sol, matches or outperforms the best existing method on almost all evaluated instances."

LLMs can design effective algorithms at level 2, where the algorithm is fixed before seeing evaluation instances, and still match specialized methods

"This holds even at level 2, where the returned algorithm is fixed before seeing the evaluation instances."

LLM performance on algorithm design improved sharply across models released less than eight months apart, suggesting this capability is advancing quickly

"Performance also improves sharply across models released less than eight months apart, suggesting that this capability is moving quickly."

A single untuned LLM query can produce algorithms competitive with specialized methods for well-specified operations research problems

"for the well-specified operations problems we study, a single untuned LLM query can already produce algorithms competitive with specialized methods."

Frontier LLMs should be considered a serious empirical baseline for algorithm design in well-specified operations research problems

"These results suggest that frontier LLMs can be a serious empirical baseline for algorithm design in well-specified OR problems."

Existing propose-and-verify methods for agent harness adaptation waste rollouts on unrelated behaviors and allow aggregate scores to obscure specific regressions

"Existing propose-and-verify methods typically score every candidate on a fixed task set, wasting rollouts on unrelated behaviors and allowing aggregate scores to obscure specific regressions."

HarnessLens improves average held-out performance by 7.6-13.6% while consuming substantially less evaluation budget than competing baselines across three agent harnesses and four benchmarks

"Across three agent harnesses and four benchmarks, HarnessLens improves average held-out performance by 7.6-13.6% while consuming substantially less evaluation budget than competing baselines."

Behavior-aware verification with explicit attribution enables more reliable and sample-efficient harness evolution under constrained interaction budgets

"These results demonstrate that behavior-aware verification with explicit attribution enables more reliable and sample-efficient harness evolution under constrained interaction budgets."