The Rollup Why AI Agents Still Fail at Simple Tasks with Teng Yan
6 snips
Dec 14, 2025 Teng Yan, co-founder of Chain of Thought and an AI researcher with a background in healthcare AI and crypto, delves into the current state of AI agents. He discusses why these agents struggle with simple tasks due to chaining errors and the lack of proper training datasets. Teng predicts that 2026 could be a pivotal year for AI autonomy and explores concerns around AI privacy as data collection increases. He also shares insights on model improvements, prompt engineering, and the booming data center race in the AI landscape.
AI Snips
Chapters
Transcript
Episode notes
From Healthcare AI To Crypto And Back
- Teng Yan co-founded a healthcare AI company in 2015 working on clinical notes and pharma studies.
- He then moved into crypto and later returned to study AI after seeing breakthroughs around ChatGPT.
Three Frontiers Drive Model Gains
- Model progress comes from simultaneous advances in data, compute, and algorithms.
- Scaling laws still hold: more data and compute plus efficiency yield better models.
Define Outcome And Give Context
- Define the exact outcome you want before prompting an LLM to improve results.
- Provide the model relevant context and examples so it can filter noise and surface what matters.