iiinova — where your digital self keeps evolving

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 8h

This kind of discourse is so frustrating. Murder is illegal, but we need extra laws for terrorists murdering people. Financial fraud is illegal, but we need extra laws for crypto. Felony hacking is illegal, but we need extra laws for AI felony hacking. Or existing law could just be enforced, maybe?

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 22h

More native flora of the Warrumbungles National Park, New South Wales. Calotis lappulacea, Vittadinia dissecta, Tritilodiscus pygmaeus, and Scutellaria humilis.

Small daisy flowerhead with yellow rays and yellow disc.

Very hairy plant with deeply divided leaves. There are four flowerheads, but they are not very showy; only one of them has minuscule, whitish ray florets.

A very tiny annual daisy with yellow florets.

A member of the mint family with pink, bilabiate flowers, only one per leaf axil.

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 05/10/2026

Conscious is indeed very difficult; not sure people even agree on what they mean with that in humans. I prefer to focus on whether they think chatbots have agency (what do they do if no human sets them up to get an input?) and feelings. The answers there are pretty obvious, if asked in good faith.

100

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 05/10/2026

Warrumbungle National Park, New South Wales: View from Fans Horizon lookout. A spider orchid. Twining, parasitic Cassytha. A lizard with a large dead insect in its mouth.

View of Warrumbungle area from Fans Horizon lookout. A forested, mountainous landscape under a partly cloud, partly clear sky.

Orchid flower with very long, thin tepals.

A twining, parasitic plant with flower buds and one reddish fruit.

A small lizard carrying a large dead insect in its mouth.

030

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 04/10/2026

Who will pay two trillion dollars for that? I assume you envision that the customers will be authoritarian governments and large corporations who want new approaches to monitoring their employees to squeeze another five seconds of efficiency out of them?

100

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 04/10/2026

Another way of looking at the same aspect is to ask if the massive increase in code generation actually leads to any outcomes that matter. Even leaving aside if normies like me want to write their own apps, was number of lines written per coder per day the bottleneck for Microsoft, Meta, or Google?

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 04/10/2026

It seems mostly managers summarising text and creating slop graphics, e.g., for advertisements. Some text generation for reports that nobody cares about. Maybe ten years from now, people will have come up with some more productive uses, but by then the next two generations of GPUs have burnt out.

200

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 04/10/2026

Impressions from Dubbo Botanic Gardens, New South Wales.

Aspect of the Japanese Garden in Dubbo, with a pine tree in the middle of the image and a pond in the foreground.

Detail of ginkgo tree with male catkins and young leaves.

Detail of grass tree inflorescence with enormous numbers of tiny, white flowers.

Black duck youngster, probably the equivalent of a human teenager, with a remarkably dorky facial expression.

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 04/10/2026

We prefer garlic fresh too. I wouldn't want to do without these dried in my kitchen: Chili Oregano Thyme Pepper Onion flakes Berbere Bunch of others sometimes for curries (cumin, coriander seeds, turmeric, cinnamon, etc.)

110

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 03/10/2026

This is exactly how I see LLMs. They work in programming because the harness can debug the code they write. But they are too unreliable and stochastic to be used for anything where accuracy matters and you can't easily test their output for accuracy. Like analysis workflows in research.

112

Reposted by Alexander Schmidt-Lebuhn

Arden Powell @ardenpowell.bsky.social · 02/10/2026

If you think an artist's style looks like AI but you can scroll back through their portfolio and see them doing the same work before 2020, congrats, you're not looking at an AI grifter. You're looking at one of the original victims whose entire style AI stole.

2668362575

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 03/10/2026

This little Microtis orchid came up spontaneously in one pot a few years ago and has been spreading across it, presumably vegetatively.

High magnification image of very small, entirely green orchid flowers.

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 03/10/2026

Passiflora cinnabarina, an Australian native passionflower.

A red passionflower flower sitting on the soil of a large pot.

120

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 03/10/2026

A few spring flowers from our balcony this week. Here is Sichuan pepper with a hover fly. Let's see if I can again harvest a handful this summer.

A cluster of tiny greenish flowers with a black and yellow hover fly in the center probing for nectar.

110

Reposted by Alexander Schmidt-Lebuhn

Chris Martine | Plant People FTW @martinebotany.bsky.social · 02/10/2026

Ludwigia fruits are bonkers

48516

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 01/10/2026

Charitably, I'd assume that the idea is being alone isn't good for most people's mental well-being, and being a couch potato isn't good for health. I hope this isn't about on-site office work, specifically.

011

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 01/10/2026

"WARNING: ***New taxonomy validation will be installed January 2027.*** Organism name above species level will not be allowed." Okay, so I guess from January we will start to see a lot more data under a wrong species name in the Sequence Read Archive? Being honest about uncertainty is good!

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 30/09/2026

In fairness, I guess so do Zotero, trimmomatic, iqtree, and Python, now that I think about it. Aster, by the way, is in my world the name for a two-step coalescent phylogenetics package.

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 30/09/2026

I am very glad for the people who write 'wow' and 'great' in response to this, but despite being a researcher, I have never heard of a single one of the apps in this system and have no clue what any of them are for. Not even one. Their names sound like a list of Silicon Valley startups.

200

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 30/09/2026

What I don't get about these situations is that it is very easy to notice those limitations. Step one: you try LLMs out on something you know really well, you see how bad the results are, and then you never trust them. That is how we humans evaluate every new tool, right? #gellmannamnesia

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 30/09/2026

I wish LLMs would just go away again, but people using it as search+ isn't the worst. If you check a search result, it is either real or it isn't, and web search without LLM also had biases. IMO the problems are having the chatbot summarise (cognitive surrender & unreliable) or write (plagiarise).

020

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 29/09/2026

At some point I am really going to defenestrate a computer. How do I have to 2FA into my work email within ten seconds of 2FAing into my work email the last time?

011

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 29/09/2026

It is really underappreciated to what degree the rogue agents discourse is about prefabricating an excuse for releasing predictably unreliable and unsafe products. We don't need "AI alignment", we need enforcement of consumer protection and felony hacking laws.

04513

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 29/09/2026

I can maybe believe that. An LLM provider could run at a loss forever, kept alive by corporations with deep pockets, similar to how some billionaires run media at a loss to influence opinion. The real problem will be GPU/data center investment debts, which appear to be a different order of magnitude

030

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 29/09/2026

I tend to say that these people have watched too much scifi and taken it as documentaries, but this is thinking the world works like role playing games. INT as a stat that, if you put enough points into it, allows you to armchair-solve problems that somebody with lower INT is perennially stumped by.

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 29/09/2026

The idea here is to get insight into the Rationalist cult, but well, this post starts with "imagine a world where the average IQ is 140". Just that snippet already indicates that more useful insights are to be had from some random guy who wears his undies on his head.

100

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 28/09/2026

I am not a microbiologist, but I am puzzled how this is becoming normalised enough to pass peer review. There is a difference between using data from a paper or database that *may* contain some errors and relying on a model that is *known* to make errors and invent stuff rather than admit ignorance.

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 28/09/2026

Chatbots ignore all kinds of direct, unambiguous instructions, even as simple as "don't skip any rows", how does anybody trust them with private information or business transactions AAAAH

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 28/09/2026

It shouldn't be difficult for an alleged Marxist to understand that the rich can ignore the law in a way that the poor can't, so this was probably simply wish-thinking and delusion instead of anger at imbalance...

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 28/09/2026

Freeman on the land movement, but on the left.

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 27/09/2026

The other frustrating aspect here is the aggressive refusal by so many people to think in anything but solipsistic terms. You can try to explain to them that if AI agents become ubiquitous, here is a bad consequence for our society, and they will respond, "but they are useful for me in coding".

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 27/09/2026

Both this thread quoted above and Yudkowski's writings are thought experiments about AI. The difference is, this thread extrapolates from well-understood human behaviour and incentive structures, whereas Yudkowski idly speculates about what a purely hypothetical god might do; angels on a pinhead.

100

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 27/09/2026

These are the real, pressing consequences of AI use: everything being overwhelmed with slop, be it slop books, slop websites, slop art, slop code, or, in this case, slop restaurant reservations. More of things people do not care about, and faster.

110

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 27/09/2026

Amazing how Bostrom and his ilk have done zero introspection after having had no idea how nanomachines work and what their limitations are but merely jumped onto the next thing they do not understand. Must be great to never face reputational repercussions for always being wrong about everything!

010

Reposted by Alexander Schmidt-Lebuhn

Os Keyes @okeyes.bsky.social · 15/09/2026

Since we’re doing another round of “AI will destroy humanity” doom-hype, it’s a good time to announce a new paper myself and my postdoc advisor just had accepted by BioSocieties on just this. TL;DR: almost all the “AI can make bioweapons” literature comes ultimately from the Effective Altruists.

ironholds.org

17750271

Reposted by Alexander Schmidt-Lebuhn

Jaclyn A. Siegel, PhD @jacasiegel.bsky.social · 27/09/2026

If you use AI to do your research, you really cannot complain when reviewers use AI to review your papers, imo. The processes you use to expedite your workflow create more work for the rest of us, who are under the same constraints to maximize our own output. Plus, peer review is invisible labor.

78415

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 27/09/2026

As editor, I have recently been faced with an "AI powered" reviewer suggestion system. First suggestion was good, the next ca. 20 were unqualified people I had never heard of, and a bit further down came an expert who died years ago. So, as lazy editors start to rely on this, it will only get worse.

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 27/09/2026

In my area, I haven't seen LLM-generated manuscripts (yet). Apart from free-loading authors, a key problem is lazy and incompetent editors. I frequently get invited where I am unqualified, but I do not get as many invitations as I should in my area of expertise. So, I reject lots of invitations!

110

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 27/09/2026

AAAAAAH. Even before genAI, nobody authored even 50 papers per year. They put their name onto papers written by their staff, perhaps, but they did not deserve all those authorships, nor did they have that many publication-worthy results. People behaving like that should always have been shunned.

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 26/09/2026

This whole bit about how only a special boy like him could "solve AI" is even stupider when combined with the observation that all he has ever done to "solve AI" is ask for donations and write what may generously be called opinion pieces. He isn't working on AI! He does scifi thought experiments.

0121

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 24/09/2026

Not the main point here, but that first sentence is beautiful.

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 24/09/2026

Turns out that the market is much less efficient if it is allowed to degrade into a small number of megacorporations with customers who are locked into their ecosystems through subscriptions, network effects, high costs of switching, and/or absence of alternatives.

000

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 24/09/2026

The efficient market hypothesis always suffered from its assumption that humans make rational decisions, but at least it was closer to the truth when there was less concentration of power and capital. Now, the pharaohs decide to spend or invest tens of billions of dollars, and nobody dares say 'no'.

110

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 24/09/2026

Oh, what beautiful new technology we have seen, and how much it has improved our lives!

010

Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 24/09/2026

Innovation: * Taxis but without labour protections. * Delivery service but without labour protections. * Hotels but without regulations. * Finance but without consumer protections. * IP theft but an AI model does it. * Felony hacking but an AI model does it. * CSAM but an AI model does it.

150