Lobsters: Selfish AI

All of this discussion, from Jeffrey's video to 99% of what shows up in forums or Mastodon discussions, is about how all of this will impact "me."

What I almost never see is the impact of AI code on our society.

Where on Fedi is this person hanging out?

Do you think he has never seen pivot-to-ai.com ?

I feel the author's pain. Other than people I met online, I don't think anyone I know avoids AI on principle because it hurts people.

But I don't think that is surprising. Generally we have a culture of pretending personal choices don't have consequences. It's socially acceptable to be outrageously, unsustainably selfish, as long as one does it legally,* and as hard as you might try to be a good person, you'll just be giving more rewards to selfish people, more words for LLMs to steal, more useful work for HFT companies to tax...

Larry, if you read this, I hope you don't give up. But things are not going to change unless we start punishing selfish behaviour. Or sabotaging it. Get hired by an AI company and deliberately do a terrible job, maybe?

*And only barely socially acceptable to abstain—just ask any vegetarian—even though you're doing everyone else a favour.

The author is wrong in many details and ironically choosing selfishness over collective action.

What I almost never see is the impact of AI code on our society.

Well, you're repeating incorrect memes. Guessing you've not actually read any of the relevant court opinions in Google Books, Kadrey v. Meta, or the discussion in ongoing Bartz v. Anthropic. A quick listing instead of a fisking:

LLMs can be trained without mass scraping
Copyright infringement isn't necessarily felonious
The courts have said that LLM training is fair use
LLMs do not fundamentally require human labeling of data
Datacenters are being built according to demand projections, as usual, with the unnamed exception of Stargate
Back-of-napkin math with relevant numbers suggests that datacenter construction is not outpacing renewable energy deployments
On top of which, many datacenters are hydro-powered
The water shortage in the Western USA is driven by farmers wielding exceptional inherited water rights, not datacenters paying market rate for municipal hookups
On top of which, datacenter water usage has become slightly more efficient as air-conditioner technology has improved
On top of which, some datacenters do (re)use grey water as opposed to always using fresh potable water

I'm also not impressed by anything that smells like culture-war histrionics. Datacenters use more water than bottled water? Okay; they also use less water than golf courses, both per acre and in aggregate.

Now, that may seem mean, but you have to consider that when the author says:

You consider Amazon unethical? … Walmart …? … Uber …? You don't want … a car? Don't want to put your small business on Facebook …? Don't want to use Apple or Android …?

Well, I don't use Amazon; I don't order packages from them, I don't use AWS, and I recommend to employers that they make better choices. I don't shop at Walmart. I don't take Ubers. I don't own a car; I sold it like half a decade ago because it was unaffordable. I don't post my small business on Facebook. (There is somebody on Facebook who claims to be me; they aren't me, please don't get scammed.) I don't use Apple products. I do use Android; in fact, I have a basket of half-broken phones that I fix for my community, including jailbreaks and mods.

So, maybe it'll surprise the author, but I don't use generative chatbots either. I know how they work, I know how to analyze them, and I know how to make recommendations about them to employers, but they're not part of my workflow. Moreover, I know how to resist employer pressure; Google wasn't able to get me to switch from Firefox to Chrome when I was there and that's the most pressured I've ever felt.

All this is to say that the author is struggling with their own doomerism. They've accepted an overly-bleak, conspiratorial, and frankly incorrect view of the world, decided that that world can't be fought, and decided to lie down and give up while bemoaning that the only thing they can do is resent other humans. "It is what it is," they say, but nobody else seems to be saying it. If you want a better world then you must choose to build it rather than sneering that "[they] do not forgive [them]," whoever they are. What the author should learn is that we cannot fight the proliferation of generative chatbots by exercising copyright.

You write that

LLMs can be trained without mass scraping

And

LLMs do not fundamentally require human labeling of data

I thought, those things were necessary to train LLMs. Do you have a link or something where I can read up on this?

I've suggested a "rant" tag because the phrasing in the linked post sounds quite aggressive, and the post itself starts out with "This will be a bit more ranty than my usual articles".

It's hard to pick apart what is basically an indistinct wall of objections and recriminations, so I'll just offer a thought to "Where do we go from here". I don't think code generation will be forever offered as a pay-per-token service; small language models are progressing by leaps and bounds, and running your own will soon fit in a workstation.

small language models are progressing by leaps and bounds

Like which ones?

Qwen, Kimi, Nemotron, GLM ..

I know that smaller models from these shops exist but I have never heard of anyone using them.

I do use Qwen3-30B-A3B, locally on Ryzen's iGPU (a miniPC bought used to be a buildbox), to slop over slog coding chores. It is good enough to be useful for some tasks (conveniently, for the most draining-via-boredom ones).