Comment by mtlmtlmtlmtl

Don't often see the great Dang commenting in a non-moderation capacity, and that's really a shame, because this is one of the most interesting, well-formulated comments I've seen on here for a while. I fully agree with your frustration with the AI zeitgeist, but I won't pretend to have any input to alleviate that frustration. I would love to much less eloquently rant about lisp(and emacs) and other vaguely related things for a bit though, so here goes.

The tendency for the Lisp community to self-segregate into these bubbles is something that's perplexed me for years. You noted that you don't agree with it; I'd in be interested in why. I do agree with it, in the sense that this phenomenon does occur to an extent, and certainly to a larger extent in Lisp compared to other languages. The part I don't agree with is that this is a universal deal-breaker for Lisp. And there's still plenty of cooperation and sharing in the Lisp community. Emacs once again is the prime example. Yes, every .emacs is as unique as a fingerprint. But the reason GNU Emacs refuses to die is precisely its wonderful ecosystem of extensions that have in fact been shared and maintained by a wider community. And this ecosystem is in part possible because GNU Emacs is devoid of any sort of boundary between application code and extension code, which is a philosophy that can be traced back to the LISP machine era.

It's also interesting to tie this in with RMS(the author of GNU emacs, what a coincidence!) and his stance on Free Software really being about every individual user being free to modify the software they run. Of course, the deep implications of this philosophy, and its implementation in Emacs and other Lisp software, are clear to me, because I'm a Lisp programmer, just like RMS. And the Free Software movement originated in a time when most people with access to a computer had at least some amount of familiarity with programming. But as computing has become mainstream, RMS' original vision has morphed, and the focus has been on the importance of the wider community being able to modify software. And that's certainly important. But it's not really free software in the original sense unless you happen to be a programmer. Most software users today are not programmers and likely never will be, so regardless of how the software they're using is licenced, they're not really free in that original sense, are they?

Then LLMs showed up. And suddenly, I can see on the horizon a revival of free software in the original sense. And yet it feels quite far away. For this to become reality, we need a lot more than what we have currently, which is a pretty damn good search engine/meme generator for the "normies", and a pretty damn good boilerplate generator for the coders. I think we probably need an entirely new concept of what software even is. I think we probably need new foundational research. I don't necessarily see a role for current LLM architectures in this. I feel more like current SOTA is scratching at the surface of the real deal, and it may take a few years to understand how to make foundational progress. Maybe we even need another AI winter to force the capital expenditure into other avenues. Certainly we need foundation models that are open in all possible senses of the word. Being tied to proprietary, heavily censored blob running in a giant datacenter is a non-starter.

It seems like the majority of people in the field have some intense tunnel vision about LLMs and transformer architectures. I'd love to see more variation there. I know Lecunn is doing some unique stuff. I can't pretend to understand whether it has legs, but I applaud the effort. Certainly we need to address the issue of energy expenditure, the absurd amount of data and training iterations required, and the lack of online learning. Human brains are vastly superior at all 3 of those metrics. So like, how about addressing that instead of just building ever larger sandcastles filled with nvidia chips? Anyone?

Those are some of the thoughts percolating through my head. I can't really figure out what to so with them. Maybe someone else can. Please? Thank you.