Post

Since it's come up again, I'll try to succinctly state my view on "AI" as we think of it in culture & tech these days. I think "Big AI" products (ChatGPT, Google Gemini) are mostly terrible for society, because they've been foisted without consent on millions, with lies about what they can do.

The impacts on trust in public conversation, sustainability of culture and creators and media, and their systematic attacks on labor are egregious. Worse, being intentionally built without safety means they are a literal threat to children, for everything from self-harm to mental health.

Much has been written about the environmental and ecological impacts of the energy demands of AI; I think those are severe as well, but may be the part of the threat that is most likely to be mitigated as some of the commercial bubble subsides. The other negatives are less likely to fade. _Then..._

Today's LLMs are the latest in a 50+ year history of "AI" technologies that all have their uses. We would have been better off if they had been described like prior technologies like bayesian networks or Markov models instead of over-hyped as the next giant bubble that every VC has bet on.

I work in Health AI where the debate is very different and elite nuanced. (Nobody with cancer boycotts AI) AI needs criticism. It needs better criticism.

hey so are you casually sweeping under the rug that AI-useful-for-cancer is very likely classical machine learning based aka extremely NOT what the current AI boom is about (LLMs) or...?

Oncology and the life sciences are replete with LLM tech.

Because hundreds of millions of people are already using these tools and finding utility in them. And they're not crazy, they are reasonable. Telling them, "no, you have to do without", will make them think, "Oh, I guess the critics are crazy and the AI bros are the reasonable ones."

I was with you this far. I lose some respect for everyone who uses those chatbots, and I tell them. It is crazy to use them, for multiple reasons - not least of which is having the same effect as bringing a robot to the gym to lift weights for us... But with the goddamned MIND and language instead.

Well, here's a real-world scenario to ponder: millions of people have terrible healthcare. And many, when they get a confusing diagnosis from a doctor, will want to give it to an LLM tool and ask for a plain-english explanation that they can better understand. That... seems reasonable to me.

But if we say, "You know you can't trust big tech. Maybe you should use AI that isn't so bad for the planet, or your job, and doesn't tell kids to kill themselves, and isn't owned by the worst people in the world. And it's coming soon." *They will listen.* The proof? You're here. On an alternative.

Read this article this morning and found it worth the time. Doesn't directly address your point, rather I'd say it is adjacent. Companies are already looking past genAI to what will be better in terms of ROIhttps://transmitter.ieee.org/generative-ai-2026-companies-looking-for-business-value/

Okay, what is the alternative software I should install on my computer?

This presupposes that a "Big AI" alternative that checks the boxes you've mentioned is inevitable. Do you think that's the case? Have you seen anything pointing to this?

I tried to write a bit about this at the start of the year; IMO the correct framing is to oppose "machinery hurtful to commonality", as the Luddites put it, and the answer is to try to create machinery that is helpful rather than hurtful roblog.co.uk/2025/01/the-... & roblog.co.uk/2025/08/ai-s...

Some quick footnotes: * A summary of the majority view from insiders in the tech industry on what's really going on with AI www.anildash.com/2025/10/17/t... * A reflection on what a "good" AI system would have to look like www.anildash.com/2025/05/01/w...

The problem is that most of us don’t have that choice. Our company says “use this”. There is no moralizing about the ethics of Gemini, ChatGPT, etc. They make me use it because they can buy a license that says BigAI won’t throw all of the company IP into the general AI bucket.

Today my boss (who knows how to use a spreadsheet) took a CSV file and ran it through a chatbot to fake some some simple aggregations. Each pass through the BS machine took a few minutes. Afterwards, he asked me to “verify” the numbers. I’ll do the actual work and the chatbot will get the credit.

Bluesky is a *marginal* improvement from Twitter. It's still run by horrible people making horrible choices, but the closest alternative is a literal Nazi. AI "alternatives" need to do a fuckload better than that.

And despite the fact that it's better, it's still marginally used (by percentages.) I don't think just making something that's better will cut it. It has to be so much of a radical improvement that it can't be ignored. And that has problems that are literally insurmountable (training corpus.)

This also presupposes that there is such a thing as a GenAI mode of usage that is not cognitively harmful, which thus far has not been shown to exist. I can show you a dozen papers and articles that say GenAI *is* shown to be cognitively harmful.