CMU Researchers Unveil TAHI, a Human-Agent Interaction Framework for Expert-Grade AI Output
EchoShao8899 · x · 2026-09-11
Zora Wang, a CS PhD student at CMU, introduces TAHI, a Test-time Adaptive agent framework through Human-agent Interaction, targeting the gap where "pretty good" AI output isn't enough for professionals staking their reputation on it.
Key features:
- Test-time adaptation via context (memory, skills) and weight training
- Efficient adaptation to individual expertise within tens of tasks
- Comprehensive rubrics for "non-verifiable" tasks
- Analysis of shared community guidelines vs. personalized tacit expertise
The companion Augmented Mind podcast episode (EP06) covers Agent Workflow Memory, tools vs. skills, human vs. agent work, and the future of human-agent collaboration.
More from coding & agent
- Okibi Launches on YC: Turns Your Product Into a CLI for AI Agents — ycombinator · 2026-09-11
- Claude Code desktop app lets you pop out diff and terminal panes into their own windows — ClaudeDevs · 2026-09-11
- DHH proposes video-game badges for sub-minute build times — AnushElangovan · 2026-09-11
- Redditor Wants LLMs to Debate Each Other Instead of Manual Copy-Paste — crua9 · 2026-09-11
- GitHub Copilot Day live streams showcase Copilot App and VS Code updates — martinwoodward · 2026-09-11
- Cerebras Fast Inference Flips Agent Workflows: Fewer Parallel Agents, Same Output — MatthewBerman · 2026-09-11