Is chatgpt making good points about renormalization and qft here?

Factual Questions
1

Input: Simulate a debate between 2 physicists arguing over the validity of renormalization in quantum field theory

Physicist 1: “Renormalization is a crucial part of quantum field theory and has been extremely successful in predicting and explaining a wide range of physical phenomena. Without it, we wouldn’t be able to make precise predictions about the behavior of particles at high energies.”

Physicist 2: “While I agree that renormalization has been useful in some cases, I think it’s important to also consider the limitations of this approach. It relies on the idea of infinite quantities, which is problematic from a mathematical standpoint. There have also been alternative approaches proposed, such as perturbative quantum gravity, that don’t require renormalization and might be worth considering.”

Physicist 1: “I agree that there are some conceptual issues with renormalization, but it’s important to remember that it has been extremely successful in practice. It’s allowed us to make predictions that have been confirmed by experiments to high precision. Until we have a better alternative that can make similarly accurate predictions, I think it’s still the best tool we have.”

Physicist 2: “I agree that the predictive power of renormalization can’t be denied, but I also think it’s important to keep an open mind and consider other approaches. The concept of infinity is something that has caused problems in other areas of physics, such as the black hole information paradox, and I think it’s worth exploring whether there might be a way to avoid it in quantum field theory as well.”

Input: Now an observer states that he heard there is more than one way to choose renormalization constants and asks the physicists if that makes the whole idea more suspicious

Physicist 1: “It’s true that there are multiple ways to choose the renormalization constants, but this is actually a feature rather than a flaw of the theory. The constants are introduced to absorb infinities that arise in perturbative calculations, and the choice of constants is somewhat arbitrary. Different choices of constants correspond to different renormalization schemes, which can give different results for physical quantities. However, it can be shown that the scheme dependence of the predictions goes away in the limit of infinite orders of perturbation theory, so the predictions become independent of the choice of scheme at high enough orders.”

Physicist 2: “I see what you’re saying, but I think it’s important to recognize that this scheme dependence is still present at any finite order of perturbation theory. It’s not uncommon for different schemes to give significantly different results for physical quantities at lower orders. This can be a source of uncertainty and makes it more difficult to compare predictions made using different schemes. It’s also worth noting that some schemes are more convenient to use in certain cases, such as the minimal subtraction scheme, which is often used because it simplifies calculations.”

2

The input is me, everything else is chatgpt.

3

Well, the first point is that renormalization doesn’t actually depend on infinities. You can get the same results by assuming that there’s some sort of upper-limit cutoff (usually, at very high energies, but it can be very high something-else in other contexts). This cutoff must be very high, or we’d detect direct evidence of it, and we don’t know where it is, but it’s finite. And then you realize that there are some perfectly valid mathematical things we can do without needing to know what the value of the cutoff is, and so we do those things, and stop worrying about the cutoff’s value.

Now, renormalization is sometimes talked about casually in terms of infinities, but that’s really just a shorthand for “very large and the precise value doesn’t matter”. It’s like when a physicist says “massless string”, what they really mean is “string whose mass is negligibly small compared to all of the relevant masses in the problem”, or when they say “frictionless”, they mean “friction that’s negligibly small compared to all of the relevant forces in the problem”. “Infinite renormalization energy” just means “energy that’s much bigger than any other relevant energies in the problem”.

4

The deeper question is more a matter of how wrong the AI is, given it has had some amount of supervised training with what we assume are good commentaries on QFT. Clearly it could be as trivial as a bot regurgitating snippets of training conversations that it massages in terms of tense and structure to address a broad question. In that respect it is going to be limited by the training conversations. If the supervised training was just renormalisation = cancelling infinities, that is what you will get back. So you could perhaps better ask - was the training given to the bot reasonable? Then ask - did the bot mangle things enough to get it badly wrong?

It looks like - yes - to a point the training was reasonable. And - no it didn’t mangle it to the point it was was outright wrong. But I would venture to say that the output is very clearly still just pattern matching against limited training data.

In some ways this does underline the strengths and weaknesses of the bot. How much one believes things have advanced from the days of Eliza and Parry is a different question.

5

Send the bot some simple math questions, formulate the questions slightly ambiguously.
Then tell me it is “smart”.

My 10y old is smarter.

Extrapolating from stuff I can check myself, the chatbot is a tool to generate low rent bullshit. Sounding just enough like a real boy to fool an idiot in a hurry. It wil be a enormous boon for phishing, selling junk and other fun stuff.
If you think the internet is a wasteland of clickbait now, wait until chatbots have saturated the fiber.

6

The question also depends on what you mean by “making good points”. It’s making points that a human would make. And, if you’re starting from the usual introductions to renormalization, they’re good points: If someone tells you “Well, we just subtract off all of the infinite parts, and what’s left works out”, you would be right to be suspicious of that process. But those being good points really just means that the explanation of renormalization was wrong, not that the process itself is.

The bot, like most humans, may well never have been exposed to a good explanation of renormalization. Which really isn’t its fault. But it is a limitation.

7

While not directly on point with the OP I think this is relevant enough:

(I know…New York Post…but no paywall…you can Google for more reporting on the same story.)

I saw some discussion that one limitation is the AI cannot cite sources. If asked to it seems to make them up (from what I have been told).