Opinions
This is an opinion page. It is not necessarily a source of truth.

What do I think about generative AI?

I think (1) most important human activity is process-oriented rather than results-oriented and (2) LLMs blunt creative expression rather than enhance it. Therefore, I think that generative AI tools rob people of opportunities to understand the world and better themselves. Increased productivity at the cost of blunted learning is a misguided trade-off.

Do I think LLMs are making life worse?

For me, LLMs have made my life worse. People I know have had to find new jobs because of it, all websites have to deal with a constant well-funded DDOS attack caused by scrapers, the Internet is now flooded with slop, and academic cheating has been accelerated to light speed. Meanwhile, the only noticeable benefits for me have been that (1) research is easier now, (2) translation is better, and (3) I can get automatic summaries. At least for me, that trade-off hasn’t been worth it.

However, for those who resonate with AI tools, LLMs have undoubtedly made their lives better. We now have previously unthinkable devices possible including glasses that can turn visual information into speech for the Blind or turn audible conversations into captions for the Deaf. Similarly, many programmers seem to genuinely enjoy AI-assisted programming even if I’m personally not a fan of it.

As a result, I think it depends, and I frankly don’t think it’s responsible to make blanket judgments about the morality of AI without considering those who it’s helped.

Do I like AI?

Generally not. Most relevant AI tools have been made by evil megacorporations that I’d rather not support, so I avoid using them when I can. The main exceptions to this are Kagi’s Translate, Quick Answer and Summarize tools because they’re often useful enough for me to put my morals to the side.

There have been interesting developments done by KL3M and some other independent folks who have helped wrest AI away from the hands of large companies and make models that people can run on their own hardware. That stuff is great, and I don’t have a problem with it, but it’s a relatively niche sector of the community.

Anti-AI extremists

The vast majority of anti-AI people just don’t want to use the technology and want to spread awareness about its consequences. However, I acknowledge that some people are extreme with it, and I don’t think that’s productive.

For example, a few projects on the open-slopware list generally do not allow AI use, but they get listed nonetheless for not virtue-signalling hard enough against the use of LLMs. The authors explain this in the list, but putting projects like WINE and NetBSD (neither of which allow LLM contributions) on a “slop” list alongside fully vibe-coded projects is misleading and irresponsible.

LLMs for drudgeries

I don’t see a problem with using AI when (1) you don’t care about expressing yourself and (2) you think (either legitimately or arrogantly) that you have nothing to learn from the experience.

LLMs for software

Modern LLMs are really good at programming and working through relatively complicated codebases these days, and I find that even the free version of ChatGPT can solve relatively difficult problems, so I can understand why people use them for it.

The vast majority of AI-assisted software seems to be produced carelessly because the tools make it easy to wave through large amounts of code without properly understanding it. For that reason, I think that rejecting AI code contributions is an understandable policy for quality control.

LLMs are great at code review, suggesting tests, etc., and I don’t see any issues with using it for those purposes.

I personally don’t enjoy the process of “vibe coding”, so I do not use LLMs in my own software. Unlike most programmers, I also think that code is art.

LLMs for accessibility

I’ve heard testimonials from Blind people about how AI has empowered them. This is great. As a result, I think a hardline anti-AI position is unethical to take because taking away the technology would disempower many people.

However, I don’t think it’s ableist to hate generative AI. The modern AI industry has made the world worse in many way, and it has ethical issues, so people are allowed to have gripes with it even if it eventually resulted in good accessibility tools.

LLMs for translation

They’re great, and I don’t have a problem with it. However, there’s been an epidemic of people using AI to rewrite all of their messages because they think their English is bad or unprofessional, and communicating with these people is a pain in the ass.

LLMs for learning

They’re great, but learning how to improve your knowledge with nothing but raw learning materials is a necessary skill as far as I’m concerned. Books don’t always explain themselves, but neither does the world in general.

LLMs for chatting

Generative AI is great at this. At the core of things, LLMs are text completion programs, and adding a box where you can “talk” to them (by feeding a chat log as the input) was a simple but brilliant design stroke.

In addition, unlike many other uses of AI, it is relatively easy to train and run a model capable of chatting — even with data in IRC chat logs and whatnot, so it seems like one of the few uses of AI that can be ethically used.

I am, however, concerned about chatbot psychosis.

Various relevant pages