The discussions in these "AI" threads (I am still hesitant to call it AI at this point -- when I was an undergrad merely 15 years ago it still mostly meant some discrete search algorithms!) have been a bit one-sided, so I'll throw in my 2 cents. Sorry if I am being annoying. I also understand that all of us read too much of these things every day.

I believe there is a strong Fear Of Missing Out sentiment and anxiety in many of these threads: the world is moving towards a completely new age, and we might be missing out if we are doing things the "old way", etc. I don't think being FOMO is a generally good state to be in. It short cuts our judgement, it transfers our emotion to our students, and it makes us unhappy. Logically, 1) just because everyone else is doing something, does not mean that something is a good thing. 2) the old way "works" -- we and everyone were doing great and interesting stuff before any of these language models came out!

I hypothesize that part of the FOMO sentiment mostly comes from the capitalistic "research factory" mindset that was mentioned in this thread. If we treat ourselves as paper machines, and our students as code machines, then it's easy to reach the conclusion that a machine that outputs a lot of (sometimes interesting) text very efficiently is our holy grail. However, I, and I believe many people like me, got into this career to understand things, to build theory of things, and to socialize and communicate with other people who build theory of things. Peter Naur's "Programming as Theory Building" discusses this mental model well. Because most of these "theories" happen in people's minds, it doesn't really matter if we have these super efficient text generators or not. The bottleneck is always in our own heads. Understanding things is always painful, slow, fun, and worth doing on its own! Being able to convey logically precise thoughts between humans is one of the unique perks in both CS and math, and I'm worried that we're giving this up by chasing efficiency instead of preciseness.

Before we advocate for drastic changes to the way we work, we should also consider the risks. Language models have been shown to increase homogeneity of human produced content (there are many papers, this is one I recently read), which is the exact opposite of what academics should pursue. We still don't know the effect of language models on people's mental health. Papers that rely on proprietary language models as part of their computational pipelines run into problems in reproducibility due to companies retiring older models. There are also the usual ethical concerns of IP theft, energy use, overreliance on big for-profit companies, etc. Society-wise, there's a strong anti-AI sentiment coming up in the US. The technology might be cool and actually useful for some tasks, but the execution from the big companies is causing a lot of discomfort to people. There is very little regard to the workers, developers, and creators, who are actually doing the work. The common saying is "if you don't use it, you will be left behind and be unemployed" (well, I personally have been more productive and have more research ideas than ever since I've stopped using these language models). I believe these are all resulting from the factory-like result-driven mindset: if there's a machine that can produce a lot of code that can run and pass tests, surely it's going to replace those people who are producing less code? But I argue that human intention should always be prioritized because society is about humans, not capitals. (And natural language is almost never sufficient to capture the full human intention rigorously, see the classical article from Dijkstra.)

Education-wise, I have indeed seen a strong impact on the students. Still, I think the message should not be to tell students to embrace the technology. We should emphasize the importance of being a human that has agency. To be a human that can do different and diverse things. To be a human that can build precise theories in their head and communicate with other humans. A human that is more than a research factory worker. I think the word "human" is often missing in a lot of these discussions. We live in a society where our actions are often strongly driven by "incentives": grades, credits, deadlines, salaries, grants, promotion, and likes on social media. Perhaps we should also tell our students to care less about these incentives and have more agencies: do the things you love! Even if the big companies may not necessarily like them. With more of these people, we should have more power to negotiate with the big companies.

I find it slightly disappointing that with all the extremely intelligent, passionate, and diverse academics, our attention is all towards the same set of things these days. There is a lot of very interesting research that has nothing to do with language or data-driven image prediction models and is not produced by consulting them (see my webpage if you want some examples ; ). Yet all we are seeing is the repeatedly homogenized things, over and over again: a new minor version of this language model comes out, check it out! It was interesting 2 years ago, but it is getting very stale at this stage. I am not arguing that we should not use/discuss these language models as they are clearly capable of something, but it might be a bit too much these days.

My stance on generative models.