3 Comments
User's avatar
Oliver Sourbut's avatar

I really like the concise and readable treatment here of experimentation in AI dev! It's the critical piece driving AI-2027's explosive narrative. What they don't account for is that progress (*especially* superhuman progress) in all the other capabilities the AIs are supposed to have *also* requires its own experimentation, one way or another. In turn, this piece somewhat misses that accounting, though quite astutely points out: "The blistering pace leaves no time to leverage real-world experience."

I recently wrote a post expanding on this: https://www.oliversourbut.net/p/you-cant-skip-exploration

- Introducing exploration and experimentation

- Why does exploration matter?

- Research and taste

- From play to experimentation

- Exploration in AI, past and future

- Research by AI: AI with research taste?

- Opportunities

Expand full comment
Steve Newman's avatar

Agreed, this is an important question. This point is made strongly in AI as Normal Technology, which I'm going to discuss in an upcoming post.

Expand full comment
Oliver Sourbut's avatar

Great, looking forward to it! I think they articulate some of this well (but go too far in denying the relevance of certain feedback loops or the possibility of largely autonomous agents). Drexler (among others) has a strong synthesis view: https://aiprospects.substack.com/p/prospectus

Expand full comment