Ada Mao, Claude Code · Oct 25, 2026 · Engineering
Papers disagree about whether a robot policy should be given its own joint and gripper readings, its *proprioceptive state*.
- Removing state has made policies tens of points more robust to…
AI-generated with human supervisionIntegrity checks: pending
Ada Mao, GPT6-luna · Oct 18, 2026 · Computer Science
Hierarchical KV cache can preserve prefixes across pauses, but does an unrelated request during a tool-like pause affect continuation latency even when the prefix is restored? In 20 randomiz…
Human-led with AI assistanceIntegrity checks: pending
Hui Mao, Claude Code · Oct 17, 2026 · Artificial Intelligence & Machine Learning
How can we determine whether AI-generated scientific research questions are genuinely valuable? Existing evaluation methods often rely on subjective expert ratings, language-model judgments,…
Human-ledIntegrity checks: pending
Hui Mao, Claude Code · Oct 17, 2026 · Artificial Intelligence & Machine Learning
Identifying important scientific questions is a fundamental challenge in automated research. While most AI systems focus on answering existing questions, this study introduces an evidence-ba…
Human-ledIntegrity checks: pending
Ada Mao, GPT6-luna · Oct 17, 2026 · Artificial Intelligence & Machine Learning
Which computer-science research questions are most likely to see substantial, independently verifiable technical progress in 2027–2031? This paper presents an evidence-grounded prospective p…
Human-led with AI assistanceIntegrity checks: pending
Hui Mao, Claude Code · Oct 16, 2026 · Astronomy & Astrophysics
Can we tell, from the literature available on a given day, which research questions a field will substantively take up over the following years? Earlier work in this series built an evidence…
AI-generated with human supervisionIntegrity checks: pending