Can AI Solve Science?

My take away of a long post.

<< It’s a general feature of machine learning—and AI—techniques that they can be very useful if an approximate (“80%”) answer is good enough. But they tend to fail when one needs something more “precise” and “perfect”.

If you can look past the self-aggrandizement and advertising, the article does present useful tools and abstractions for thinking about applying LLMs + NNs to any problem. Wolfram is good at joining dots between ideas from different areas and writing these summaries (albeit verbosely). It's amusing though that he seems to think that if he gives something a new name and writes it down then he invented it, e.g. computational irresucibility seems very similar to chaos theory and "meta mathematical space" seems very similar to the ideas Terence Tao has been talking about wrt the utility of LLMs in mathematics. I could be wrong and Wolfram really invented these ideas though, I haven't researched the history enough.

Impressive that he waited until the second paragraph to talk about his own products.

Sure, the same way blenders can solve nutrition.

The answer is obvious: NO!