Hanno Reinhannorein@mastodon.social
Oct 22, 2026, 1:55 PMI just saw this new paper on the arxiv:
AI and The Enclosure of the Scientific Commons
Very interesting and worth a read!
https://arxiv.org/abs/2610.27251
Replies
François Rinconfrincon@sciences.reOct 22, 2026, 2:48 PM@hannorein See also
this blog post"You would think that a machine powerful enough to outsmart humans would not require such levels of opacity. That it would be able to rush ahead without training data from the last few years of research. That the prompts and any available information on its path to discovery would be made available for the benefit of its future users. That its outputs would be readable and compatible with formalized proofs, and that by solving 400 problems it would generate 1000 conjectures, as humans would have done. Clearly, the machine is not there yet.
But, for the purposes of capturing scientific production, maybe it doesn’t need to get there at all."
https://proofsandprompts.com/2026/10/21/artificial-intelligence-and-the-mathematical-community/
Boudboud@framapiaf.orgOct 23, 2026, 8:53 AM
Keith Böhlerkeithnator3000@universeodon.comOct 23, 2026, 2:46 PMThis is a good read.
> The founders of the Free Software movement furthermore had the brilliant insight that IP restrictions also have a role to play in creating and defending information commons as a whole
This is a good bit too. I am curious as to how these ideas might evolve to keep up in this new environment. I have read in online spaces many complaints that copyright and patents were outdated to how they are actually used.
Given that many things are getting de-compiled and re-implemented with AI; I think that we are going to see some landmark case. Like the thing with Art Craft; I don't see Adobe taking it sitting down and they do have money for a dream team of lawyers to figure something, or at least to lobby for a favorable change to laws.
- Hanno Reinhannorein@mastodon.socialOct 23, 2026, 3:31 PM
@keithnator3000 Yeah, I have no idea how things will turn out. For me as an academic, I think very soon anyone will be able to reimplement/vibe-code the software I write for astrophysics from "scratch" using AI. But of course, "scratch" really is "all human knowledge" including all my previous software. So I definitely think of it as theft. But I suspect scientists who just care about the end result and consider my software just a tool will no longer cite my papers.
Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 12:55 PM@hannorein @keithnator3000 And here open science has a role to play. If people want to do this they will have to show the whole thing. More so, they will have to own the mistakes too. The reason not to use existing tools, or build upon them, is to scoop up free credit and credibility. However, the reverse should also be true. If you do this and you are caught out you should lose credit twice over.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 12:56 PM
@hannorein @keithnator3000 Sadly academia has never been good at applying moral standards, rather than just use "merit" ones.
- Hanno Reinhannorein@mastodon.socialOct 24, 2026, 2:11 PM
@koen_hufkens @keithnator3000 I get caught off guard by this so often. A lot of people who were up in arms about some moral issue just recently (gender ratio, DEI, whatever) seem to have a complete blind spot when it comes to AI. All I hear them talk about is "oh, I was skeptical initially but now it's so good at doing XYZ". To me this sounds like "oh, I was skeptical about Hitler initially, but his autobahn is really convenient".
- François Rinconfrincon@sciences.reOct 24, 2026, 1:03 PM
@koen_hufkens @hannorein @keithnator3000 i have a lot of colleagues appalled by AI who now think open science/data/OSS has signed its own death sentence by simply naïvely being open and thinking it was good for humanity.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 1:05 PM
@frincon @hannorein @keithnator3000 I don't think so. What the community consistently does not do was call out any freeloaders and anyone doing open-washing. I'm against any sort of purity tests, but by and large it has been a generational movement with seniors grandfathered in kicking and screaming. And boohoo if they were called out for it.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 1:07 PM
@frincon @hannorein @keithnator3000 In short, the raping of the commons was commonplace before open science even matured and wasn't/isn't limited to AI. With bean-counters and seniors lining up to take the biggest cuts of that pie.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 1:09 PM
@frincon @hannorein @keithnator3000 And finally, and this I've told many times, academics should have learned about IP and copyright. Although AI waltzed over this too, at least within their own circles this could have been a stick to beat people with.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 1:18 PM
@frincon @hannorein @keithnator3000 And I'm going to add one last thing here - before I stop this rant. There is this misconception that data and code is apolitical (especially within academia). This has led to a lot of these issues. Where power dynamics with respect to code and data are not acknowledged, honoured or abused. Even collecting data is political and not until recently fully acknowledged. The political dimension of doing science is HUGE blank spot for a lot of academics.
- François Rinconfrincon@sciences.reOct 24, 2026, 1:20 PM
@koen_hufkens @hannorein @keithnator3000 yes i agree with this.
- François Rinconfrincon@sciences.reOct 24, 2026, 1:09 PM
@koen_hufkens @hannorein @keithnator3000 yeah, but when you realize how its been leveraged by the AI industry now, it increasingly looks like blue sky discovery of nuclear physics in terms of side effects.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 1:12 PM
@frincon @hannorein @keithnator3000 I think this is a societal problem, not one limited to open science. AI is the greatest theft in history. Grandmas were sued for supposedly downloading three mp3s in the '00s and now we can hoover up all digital content and use it for free? The issue is not with the content, it is with consent and enforcement of it in the face of capital.
- François Rinconfrincon@sciences.reOct 24, 2026, 1:18 PM
@koen_hufkens @hannorein @keithnator3000 the issue i see here is that with humans as long as something can technically be done, at some stage it looks almost inevitable that it will be done, whatever the consequences. Law is good but reality is stronger.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 1:21 PM
@frincon @hannorein @keithnator3000 Sure, but this doesn't mean it should be free of consequences.
- Hanno Reinhannorein@mastodon.socialOct 24, 2026, 2:30 PM
@koen_hufkens @frincon @keithnator3000 How do you feel about open weight models? For me this mostly resolves the theft issue, but it does not address the consent one.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 4:33 PM
@hannorein @frincon @keithnator3000 For some re-use there is no need for consent, for others there is (given the scope of the re-use). e.g. using my mostly AGPL code is OK - if they not only share the weights, but arguably also the tooling, which would be implied by the license. Which means any API stacks, the lot. Again people play fast and loose on that part.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 4:41 PM
@hannorein @frincon @keithnator3000 There are very few truly open models around (which take into account consent, licensing, or only use public domain data). Most open models violate the spirit of open (science). The idea in most open models is that they "invested" in the stolen material (model training) which justifies the theft if giving you free access to the models, therefore "offsetting" the harm and lack of consent.
- Koen Hufkens, PhDkoen_hufkens@mastodon.socialOct 24, 2026, 5:07 PM
@hannorein @frincon @keithnator3000 To quote Aaron Swartz, all software is political. Big tech consistently, and intentionally, ignores the political dimension of their work. The assumption of move fast and break things is that if you can outrun the political dimension for long enough you get to set that agenda, too.
Ted Mielczarektedmielczarek@mastodon.socialOct 24, 2026, 3:54 PM@keithnator3000 @hannorein FYi I think the claims of “decompiling” or “reverse engineering” are overstated. Software like Artcraft seems to be nothing more than a superficial clone
- Hanno Reinhannorein@mastodon.socialOct 24, 2026, 3:58 PM
@tedmielczarek Fair enough. But I do worry that people think of slopware (is that a word already or did I just invent it?) as "good enough".
Rainer Zufallrainerzufall_le@mastodon.socialOct 24, 2026, 7:56 AM
LogicalErzorLogical_Error@fosstodon.orgOct 24, 2026, 8:59 AM@hannorein i wish the paper went into some detail on changing global structures to encourage those who care about the commons
game theory can be valuable here
Hendrik Weimerhweimer@fediscience.orgOct 24, 2026, 9:08 AMI believe I have issues with the claim that "Common wealth is being transformed into private gain at enormous scale." I think we are way past the point where the enormous costs of training foundation models can be justified in terms of the gains over open-weight models, i.e., over the commons. And I don't see a good way to build a rent-extraction scheme around inference alone (you can always switch suppliers at low additional costs).
To me, it seems that AI companies are highly stupid in the Cipollian sense. Sure, the commons are losing because they have to deal with slop spam, but the AI companies are losing even more.
This doesn't mean that LLMs will go away when the bubble bursts, but we might be stuck with the last generation of open-weight models for a long time to come. Maybe a bit like our ability to build heavy rockets after the end of the space race.
Oct 24, 2026, 6:20 PM@hweimer @hannorein it feels like an equilibrium will have to arise as the modus operandi is untenable.
I quite like the vision of the training frenzy dying down, and existing open LLMs getting modest updates over several months, while novel architectures are trained mostly for efficiency.
I think there's also a bit of a plateau in that the biggest frontier models can only really offer benefits for people able to come up with challenging enough tasks that are beyond midrange open models' abilities, but that's shrinking to be the "scientists with supercomputers" kind of thing we had before.
- Hanno Reinhannorein@mastodon.socialOct 24, 2026, 6:57 PM
@dain @hweimer There's a paper which argues that an inevitable model collapse will happen because future models will be trained on more and more LLM generated data. From a physical point of view, I think this makes perfect sense. https://www.nature.com/articles/s41586-024-07566-y
- Oct 24, 2026, 7:09 PM
@hannorein @hweimer that might be true in the recursive case, but empirically, especially small models benefit from synthetic output from larger models (if you want I can find the paper). Also, while post-training has been yielding most of the gains in the last couple of years, what also happened in pre-training is more and more 1) aggressive filtering of training data so slop, SEO fodder, repetition, etc are removed using previous model generations and 2) synthetic data is generated to fill in gaps.
So I think collapse is far from inevitable even with just filtering, but also there are massive gaps in data that are still being filled in. My favourite recent (open model) example is Kolibri that got much more German training data: https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/
Three plus or minus fiveThreeSigma@mastodon.onlineOct 24, 2026, 11:29 AM@hannorein
I dunno. It’s little more than an extended analogy, and gives very few actionable or analytic insights. Interesting, but offers little hope for changing the destructive path we’re on.
Maybe someone can take it another step or two
Oct 24, 2026, 4:24 PM@hannorein
This has been a key trend of post industrial capitalism. The last 40 years or so it has become the main feature of it. Things like privatization of essential government services, water, power, etc. "Selling"(read giving) of public lands to corporate interests. Late stage capitalism is focused on becoming a middleman not on creating anything. Why provide a service when you can just get all the people providing it to go through your platform?
Oct 24, 2026, 6:16 PM@hannorein Thanks for this reference. I concur, this is a great read. One of my favourite pieces in there:
"It is now clear that if you give people generative AI tools and a network, they will spam slop. Virtually every commons-based peer production process mentioned above is currently facing the equivalent of a distributed denial-of-service (DDOS) attack on its integration processes."
- Oct 24, 2026, 6:19 PM
@hannorein Most importantly the article makes clear that our most urgent problem is not what current state-of-the-art AI technologies can do or might be able to do in 2, 5, or 10 years, but how we can keep our commons-based peer production processes up and running (including not only open source software and digital public infrastructure, but science in general).
Everyone: go read this. And then take time off and talk to each other, instead of generating more content!
Zack Batistzackbatist@archaeo.socialOct 24, 2026, 6:53 PM@hannorein Pasting my reply to another post about this paper: https://archaeo.social/@zackbatist/117486138926354380
Al Pastormorgan@sfba.socialOct 25, 2026, 6:42 AM@hannorein I will read the pdf. Thank you.