I struggle to believe that AI R&D is ingenious
This discussion provides practical, anecdotal counterpoints to the hype around fully autonomous AI research agents, highlighting the gap between narrow optimization and genuine research taste.
Reported by Hacker News ·
What happened
In a LessWrong shortform post, Arjun Panickssery shares observations on the limitations of autonomous AI research and development across several personal projects, including building game engines and formalizing philosophy proofs. While noting that AI excels at hill-climbing tasks with fast feedback loops when guided by a human, he argues that models still show poor basic judgment in literature review, experimental design, and identifying novel ideas.
Who it’s about
- Arjun Panickssery
- LessWrong
- Claude
- Epoch AI
Source
Hacker News
I struggle to believe that AI R&D is ingenious