-
EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
Paper • 2608.28363 • Published • 10 -
UndoBench: Separating Task Competence from Recovery Capability in Tool-Using AI Agents
Paper • 2610.05622 • Published • 12 -
The Verifier Tax: Horizon Dependent Safety Success Tradeoffs in Tool Using LLM Agents
Paper • 2603.19328 • Published • 1
Tanmay Sah
TanmaySah
AI & ML interests
AI agents, LLM reasoning, ARC-AGI, verifiers, agent harnesses, recovery, model evaluation, and trustworthy AI
Recent Activity
authored a paper about 19 hours ago
A CPU Inside an LLM: Zero-Gradient Compilation of Exact Circuits into Pretrained Weights upvoted a paper about 22 hours ago
A CPU Inside an LLM: Zero-Gradient Compilation of Exact Circuits into Pretrained Weights submitted a paper about 22 hours ago
A CPU Inside an LLM: Zero-Gradient Compilation of Exact Circuits into Pretrained WeightsOrganizations
None yet