900 karma · joined September 1, 2025
I feel like this is a modern version of believing in souls. You are matter, not data. If you find a way to simulate yourself on a computer, this will not prevent you from experiencing death. And if that's the case, what's the point? Stroking your ego with the knowledge that a simulation of you will stick around for some time after you give up the ghost?
"Bohr (1937), Heisenberg (1947), Frank (1936) and others explained carefully -- but did not prove -- that the theory makes no assertions concerning autonomous, i.e. observer-independent, things: that all its statements are about experimental situations. (This is why Bohr, and initially also Rosenfeld, stated that no special theory of measurement was necessary: they believed that quantum mechanics was already a theory of measurement.)" From Mario Bunge (1979) "The Einstein-Bohr debate over quantum mechanics: Who was right about what?"
"They surveyed hundreds of fish and chip shops in Scotland to find out if "the delicacy" was available and if people were actually buying them. It found 66 shops which sold them, 22% of those who answered the survey. [...] Annie Anderson, from the Centre for Public Health Nutrition Research at the University of Dundee, used to send her medical students out into the city to see if they could find somewhere that sold deep-fried Mars bar. "It was not much of a challenge in Dundee," she says."
1. You can infer the output from the mechanism. (Because it is implemented by it).
2. You can't infer the mechanism from the output. (Because different mechanisms can easily produce the same output).
My point here is 1, in response to the parent commenter's "the mechanism of operation dictates the output, which isn't necessarily true". The mechanism of operation (whether of LLMs or sunflowers) absolutely dictates their output, and we can make valid inferences about that output based on how we understand that mechanism operates.
Where does the output come from if not the mechanism?
So the name of the game is quantity over quality?
Re: "I'm baffled that someone in CS, a field ruled by applied abstraction, has to be explained over and over again that abstraction is a thing that exists". Computer science deals with models of computation. You are making a classic mistake in confusing models for the real things they are capable of modelling.
What I'm interested in is evidence that supports that "The more you try to look into the LLM internals, the more similarities you find". Some pointers to specific books and papers will be very helpful.
However, if you can point us to some specific reading on mechanistic interpretability that you think is relevant here, I would definitely appreciate it.
Otherwise, if "the parts are all different, and the construction isn't even remotely similar", how can the thing they're doing be "the same"? More importantly, how is it possible to make useful inferences about one based on the other if that's the case?
That's not really "little". Back in the day computers could only do a fraction of this. Today you have it all in your pocket