Epistemic Status: You Get About Five Words; explaining my top takeaway from the post.
Of course this difference is to some degree not as much a difference in the model behavior training of either company as a difference in the documents that each choose to make public.
This sentence weakly implies the Five Words: "instrumental convergence of model design". If we take this principle further, we get the idea that Claude N, ChatGPT N and Gemini N are all likely to be very similar as N increases.
This has implications for alignment work that I'm not quite sure how to think about:
- Any current technique that only works on some SoTA models is brittle;
- If... (read more)
Survey completed!
I believe Duncan's posts are cross-posted from his Substack to LessWrong due to popular demand. In that sense, Duncan can still be viewed as somewhat active.
However, the "(inactive)" in the name serves a purpose, which (very confident inference) is to public declare Duncan's relationship with the LessWrong comments section, which (reasonable inference) is somewhere between "I don't participate in the LessWrong comments section" and "I find some comments valuable and some comments anti-valuable, but want to specifically discharge my usual social obligation to read and respond to comments on my posts because this would be net-negative for me."
[comment epistemic status: quick and somewhat imprecise to serve the purpose of quickly answering the implied question of Keenen's comment.]
Problems often persist not because they are hard, but because nobody's paying attention to them. These eight fictitious studies are sketches of what paying attention can look like in practice.
The Well-Dressed Intern
Aaron wanted to make a strong impression in his first week as an intern, optimizing his appearance to be taken seriously. He aimed to both fit into the office culture and subtly signal his status as an intern, making it significantly easier to ask so-called "stupid questions."
During his interviews, Aaron paid careful attention to the office's general dress code. He noted the typical button-downs and khakis, but also observed senior associates leaning towards slightly more formal attire.
Based on these observations, Aaron... (read 1986 more words →)
Agreed. I think a more measured phrasing is "I am confident that across a significant number of current AI companies, in the near future, the models are more similar than one would expect from reading the difference between (representative example) Claude's Constitution and OpenAI's Model Spec."
I think I am personally holding the strong conclusion (that you are pushing back on) as plausible-enough and important-if-true-enough that I want to be tracking how true I think it is as I get more information.
If even moderately weak forms of "instrumental convergence of model design" end up clearly false in O(5-years) time, I would be surprised but not confused.
If anyone with more context on modern AI models than me disagrees, please say so; me knowing multiple perspectives on all of this is useful and makes my world-models better.