clarification

HyperHat's founder shares a design sketch for benchmarking whether AI agents can identify contradictory task requirements and ask clarifying questions rather than blindly executing, pairing broken tasks with satisfiable controls to prevent trivial gaming of the eval.

The paper introduces CLEAT, a framework that co-trains a clarification-value controller with an LLM task policy using paired rollouts to estimate residual query value, enabling agents to ask useful clarifying questions that adapt as the policy improves, achieving top scores across four benchmarks with Qwen3-4B.

The paper introduces a leakage-aware framework for conversational task disambiguation over tabular data, decomposing agents into asking and solution policies, and presents AmbiTab, a benchmark suite unifying six ambiguous text-to-SQL datasets. It also trains an asking policy with reinforcement learning, improving disambiguation metrics on all six datasets and task success on five.

该论文提出PlanPool方法，通过将澄清计划外部化为可变问题池，解决agentic Text-to-SQL系统中过早终止澄清的问题，从而减少静默失败并提升歧义覆盖能力。

OpenAI researcher Sebastien Bubeck clarifies a dispute with Anthropic employee Levent regarding authorship and collaboration on a math breakthrough involving Euler and Navier-Stokes equations, emphasizing the intention to celebrate achievements.

This paper introduces the Memory–Clarification Boundary (MCB) benchmark to evaluate how LLM agents decide to persist, verify, or clarify memory updates, finding that models verify changing facts more reliably than they ask for clarification.

This tweet discusses the importance of asking clarifying questions in system design interviews, explaining how questions drive better architecture than rote diagrams.

This paper investigates using LLMs to rewrite fragmentary dialogue utterances for improving frozen discourse parsers, finding that zero-shot clarification is unreliable and that error repair through rewriting has a practical ceiling, suggesting rewritability prediction as a key missing capability.

A tweet highlights that Anthropic conducted large-scale reinforcement learning using Slack conversations, with Andrej Karpathy emphasizing that it is not a trivial Slack bot feature as commonly misinterpreted.

Proposes a goal-oriented clarification framework using Information Gain Reward to train LLM agents to ask effective clarification questions under underspecified user instructions, improving task success rate by 3.7% with minimal interaction overhead.