Oct 3, 2026, 3:34 PMThe more I read about AI, the more it seems like everyone lands in 1 of 4 camps:
Booster - it's the next industrial revolution, get on board or get left behind
Technologist - useful in certain domains, but the financing/business side is irresponsible
Skeptic - overhyped, a cheap imitation of what people actually produce
Abolitionist - built on stolen work and used by the wealthy to hold down wages
Which is closest to you? If none, reply with yours.
@HobbyHero I consider myself a skeptic.
It can do things, definitely, but knowing how it works and seeing it never generate the same thing twice makes me deeply distrust it.
The pitch for it being largely based on personal anecdotes and experiences doesn't help, either. It makes it feel uncomfortably similar to alternative medicines or psychics.
@MutedTrampet And of course, the essay The LLMentalist Effect is a modern classic on the similarities between LLMs and psychics. https://softwarecrisis.dev/letters/llmentalist/
@ljwrites Yep I remember reading this some years ago and still agree with it now.
@MutedTrampet @ljwrites I never really understood this view. I could see it for a while, when gen AI wasn't producing anything particularly useful, but we're at the point at which it's generating executable code and theorem proofs. If psychics could find proofs for theorems by doing whatever it is they do... well, that'd change my view on psychics I guess.
@modulux @MutedTrampet As explained in this declaration signed by Fields Medalists, https://mathandai.org/ machine-learning and generative models are in fact not producing anything particularly useful when they simply spit out solutions without comtributing to the primary goals of insight and education.
As also pointed out in the declaration, these models also raise attribution and plagiarism questions, turning the psychic comparison into something even more sinister: a con who rifles through people's papers and the trash to steal credit and pretend they have superpowers.
@ljwrites @MutedTrampet Are we reading the same declaration? The first sentence says: "Over the last few months, the mathematical capabilities of LLMs have improved dramatically, to the point that they can solve major outstanding problems in many fields of mathematics." Is this true for psychics?
Further on the declaration, it says: "building on a vast body of previous human work, AI systems are becoming increasingly capable of producing the results of such work directly." Is this true for psychics?
Consider the unit distance problem solution. As far as I know the approach was completely novel.
Yes, there are serious issues with the way these systems are being pointed at problems and producing outputs that are either not explanatory or plain impossible to understand by humans (the 13 million line proof of FTL comes to mind). This is a genuine problem, but it wouldn't be a problem if the systems weren't capable of finding useful solutions in the first place.
The attribution issue is more debatable, theorems are true or false independently of who states them, and mathematics itself has had such a problem, if it is a problem, for a very long time, consider Stigler's law of eponymy, also known as Boyer's law.
@modulux @MutedTrampet Did you read this part? “But solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Forgetting this in the world of AI may turn the tool against the primary goal. Indeed, the mass production at faster and faster pace of "true/false" statements could destroy fertile ground instead of breathing life into new ideas.” This means AI has the potential to be counterproductive, as in the opposite of useful.
You appear to be missing the reason I brought up credit-stealing issues in the first place: I'm not debating the truth or falsity of AI solutions or even the ethics (though that is a related and serious issue), but rather pointing out that to the extent AI's "achievements" are based on taking human processes uncredited — and we know that an AI knows absolutely nothing that has not been entered by or stolen from humans — it has no power of its own, only a conglomeration of human work often without consent and at direct detriment to humans who actually did the work. Hence the comparison to a con person who steals information: Not only is it unethical and harmful, my main point is that the "AI" has no intelligence or skills of its own, and any appearance of such is an illusion comparable to a psychic's con aided by thievery.
@ljwrites @MutedTrampet I did read that, and I agree with it. Pointing an LLM at unsolved problems and getting incomprehensible yes/no answers can be counterproductive, though there are cases where having an answer from an oracle can itself be useful since it can ground further work.
I agree I was missing the reason you bring up the credit issue, though I cannot accept the framing of it as stealing given my views on intellectual property (it should not exist). But let's set that aside.
If I understand what you're saying correctly, forgive me if I'm reading your claim as stronger than you mean it, you're stating that any AI output can only embody knowledge that has already previously been produced by some other human and is present in its prompt or training corpus.
If so, I don't think this holds. The declaration also seems difficult to reconcile with that position. In particular, it says: "AI offers the potential of enhancing and accelerating genuine mathematical study and understanding." It's hard to read that, together with its claim that AI is increasingly capable of solving problems directly, as claiming AI cannot produce novel results.
Consider the unit distance result.
The result itself was new, externally verified and surprising. The conjecture was widely believed to be true. Naturally everything an LLM generates is based in some way on its corpus, but the particular construction to solve this hadn't been produced by a human before. If deriving a novel result from previously existing mathematics doesn't count as novel, then I'm not sure what does.
Incidentally, this is also something psychics can't do, unless they're also mathematicians; in which case, the analogy holds because they're mathematicians, not because they're psychics.
@HobbyHero I have no idea how much this will help anyone, but I was just having a random thought. I have been using Claude to help me with local administration and configuration tasks. It's been increasingly good, including for bash, python, golang code generation. I tried Gemini for the same thing recently and it failed immediately at bash.
It's still very uneven, and narrow experience can lead to false impressions. fwiw.
Oct 3, 2026, 10:21 PM@buckfiftyseven
Results really can vary wildly by what kind of task you're giving it, on what model, with what harness, and with how good of a job you did providing necessary context.
I seem to be in a minority that I actually had pretty good experiences with Gemini Pro for coding help when I had a free trial that came with my phone. I've now gone back to the work approved Copilot license and it is so much worse for what I want.
@HobbyHero
Unix Memory Vibemap – treemapping memory usage on FreeBSD and Linux
https://github.com/grahamperrin/umv
– my first vibe-coded app used Python, for what it's worth. https://www.reddit.com/user/grahamperrin/comments/1wtipep/unix_memory_vibemap_treemapping_memory_usage_on/
#AI #FreeBSD #Linux #memory #RAM #Claude #Python #PySide6
Cc @HobbyHero
This should tell you https://sightlessscribbles.com/posts/a-thermodynamic-failure-of-the-soul/ @HobbyHero
Oct 3, 2026, 9:18 PM@WeirdWriter Interesting read!
Oct 3, 2026, 10:16 PM@HobbyHero
Bluesky doesn't do polls, but I would be really curious to try to compare people there with here. On my feeds at least, I think I used to see on Mastodon a lot of skeptic and now see a bigger dose of technologist along with some abolitionist, while my Bluesky is still overwhelmingly very angry abolitionists.
Even on LinkedIn I don't see a lot of true boosters outside of the ads.
@Ryan this is very useful, thank you.
I had considered joining Bluesky, mainly for interaction with a few people whose content I love when I see it in Mastodon.
I have taken measures to keep anger away from me in Mastodon. Given what you describe: I'll not inflict Bluesky upon myself. Instead, I guess …
Cc @HobbyHero
Oct 7, 2026, 12:40 AM@fuzzy
To be clear, there is a lot that I do like about Bluesky - lots of interesting people, custom feeds, labels, block lists, better app UI, etc - and I do usually end up spending more time overall there.
But when it comes to "AI" conversations I mostly find it unhelpful anger where Mastodon does seem a little more capable of having a bit more nuance.
@HobbyHero
@HobbyHero Abolitionist, and add lack of consent and centralization of power to the list.
@HobbyHero Couldn’t pick one, largely because “technologist”, “skeptic”, and “abolitionist” aren’t mutually exclusive. The first two are pragmatic assessments while the last is a moral/ethical stance, and a person who believes “it can be useful in certain limited circumstances while still being one of the most overhyped techs *ever*” can fit both of the first two.
(I’m probably closest to a skeptical/abolitionist combo, but I am *absolutely* not going to dump on, for example, disabled people who have started using LLMs and other generative-“AI” to do some things when other options were clearly worse or even nonexistent.)
Oct 3, 2026, 11:50 PM@dpnash Fair point.
@HobbyHero in practice, I land somewhere between skeptic and technologist. I think that there needs to be more sensible research on how this technology can be useful, but I'm not totally convinced that there are any use cases where it's going to be the best tool.
Politically, I'm closer to abolitionist. Even if it can be useful, this tech needs to be torn down and started over on an ethical and sustainable foundation.
Oct 4, 2026, 1:15 AM@alter_kaker very true - LLMs are definitely designed to be jack of all trades, master of none.
@HobbyHero It’s a big, overhyped bubble that’s going to take the entire tech industry down in flames? I’ve never seen tech vendors shove a feature people don’t want so aggressively! It smells like desperation to monetize.
I’m getting flashbacks to the dot-bomb crash of the aughties. Overnight people realized that just because something was on the internet, it didn’t make it profitable. Next thing you know I was in an unemployment office filled wall-to-wall with unemployed software engineers. Like me. 😉 It was years before I found another job in IT and I had to side load myself in (library -> library technology-> IT).
It’s a tulip bubble.
Oct 4, 2026, 3:31 PMThat connects to what @HobbyHero is saying though, which is more subtle than that.
The "AI industry", by which I mean OpenAI, Anthropic, Google, etc. I don't think these companies are AI companies at all in the same way Google is not a search company.
Google is an advertising company that provides search.
The big AI companies are data collection companies that provide AI.
If you look at AI through that lens, you understand both the push and the money.
As for the technology- I think that we're only seeing the tip of the iceburg, and the future are smaller models, non-language models, etc. That's where things will be more subtle and interesting.
It's like how computers were in cars starting in the 80s, but no one noticed it because they were doing fuel injection and brakes.
@serge @HobbyHero My comment was intended as an expansion of my vote, but yeah. The poll options, while useful and interesting, are kind of reductive by nature and a lot of us are not neatly in one category.
But also… it feels like every tech product is rushing to announce that they have AI whether it’s useful or appropriate. “EXCITING NEWS we’ve IMPROVED by adding a feature you don’t want and are also RAISING PRICES!!!”
I’m particularly vexed by everything from wikis to word processors begging to write for me. There are serious consequences to that in my current field—creative writing—ranging from anti-AI witch-hunts (which I hate) to loss of publishing contracts (because “you didn’t write that”) to being blacklisted for awards to questions about copyrights. The last is unexplored legal territory.
(1/2)