CharonY

But may chatbots will eat you? I am not quite sure whether LLMs are really collective intelligence as such. They process collective intelligence into something artificial.

Absolutely fair. I think that goes towards the argument made in the article. From a mathematics rather then physics perspective, developing and approaching questions, even if they do not have any real-world counterpart can be relevant for the field itself, but only if the work is being demonstrated and can be read and understood by humans. I think OpenAI offered to publish with Buckmaster as author. However, there was a second mathematician (Alpoege) who worked with Buckmaster which they didn't want to have on the paper. The reason was because Alpoege works for Anthropic. In addition to the maths and the issues of potential negative AI influence on science we also can add conflict of interest and the perhaps the negative impact of companies on scientific integrity....

As you may have heard, Open AI claimed to have solved the Navri-Stokes problem https://openai.com/index/navier-stokes-solution/ However, there are multiple discussions around that claim. For now, the full solution has not been presented so there are a few uncertainties regarding the validity of the claim: There is also another controversy: https://www.nytimes.com/2026/09/10/science/tristan-buckmaster-openai-math-navier-stokes.html?unlocked_article_code=1.AFE.CA12.DVpcZwQn9Yl6&smid=url-share That opened up a completely different question. Let's assume that a mathematical problem is solved without human input but at high cost (the estimated computation time was valued at around 15 mio) what is ultimately worth? I also like the final point made in the last paragraph here: Often, AI compute time is seen as free, but it ultimately is expensive. And I wonder whether 15 mio in form of grants to further mathematical research group would have yielded more insights, for example. I also see parallels in the use of AI in biological research where (as also mentioned in the article) the actual work starts after hypothesis generation. https://www.nytimes.com/2026/09/22/opinion/artificial-intelligence-ai-danger.html?unlocked_article_code=1.DVE.uGAZ._2SuQ4ojyQES&smid=url-share

That really sounds like a pellicle then. If you ferment for 2 weeks do you get something like 5-6% (v/v) of alcohol? Concurrent acetic acid fermentation will carve out a percentage depending on oxygen availability but with CO2 in headspace it can be very low. But I would have expected a fairly thin pellicle layer at most. Wild yeast also forms pellicles alone, but due to the lack of cellulose, they tend to tear easier and are not really rubbery. There are members of the Acetobacteraceae that are able to produce cellulose even under microaerobic conditions. IIRC Gluconacetobacter would fall under this category.

Is the stuff only on the surface (like a pellicle), are there gelatinuous blobs throughout and/or does the rest of the liquid seem more viscous? Surface is smooth?

I think based on what has been described how the hack happened, ants, bees and other eusocial insects are actually poor models. Paradoxically because of why they are good models in bionics. One of the reason why they are used as model in these areas is because individually ants are considered to be very simple and can be modeled with a limited set of instructions, the colony as a whole has limited complexity in terms of roles (e.g. often the colony can be modeled with limited granularity and required number of roles), and each role has somewhat limited flexibility. I.e. the idea is that we can understand the emergence, because each element is very simple, not very flexible and is mostly governed by limited instructions and feedback loops. AI swarms on the other hand have more flexibility in each role (and they invent new roles) and as such are quickly almost like a chaotic system that cannot be understood from first principles. I always have some issues with these overarching narratives. They make things feel like they are easy to understand, but they often clash with the minutiae of things we understand/don't understand.

It is a bit unclear, as functional characterization is missing. The targets are among the usual suspects that can modulate pathogenicity and/or virulence but from a skim, nothing really unique.

Or, for that matter XY chromosomal functions in humans, based on this framing: There is a reason why in men X-linked traits are dominant (and not it is not because of feminism). Women have two functional X chromosomes, whereas in men one is a reduced version, as KJW mentioned, which is why it looks more like a Y in a karyogram. I.e. men are lacking bits of a whole X-chromosome is a slightly more accurate framing. Slight clarification, the Y chromosome does have more genes (though estimates vary). Beyond encoding the SOX-9 protein, it also encodes the SRY protein, that activates the expression of the Sox9 gene, for example. IIRC a full X chromosome has over 1k genes, whereas estimate of the Y chromosome are around 100-ish.

Many if not most animals are able to learn and increasingly we have identified examples of abstract thinking and certainly of cognition. Humans are not unique in that regard and animal behaviour is not determined by DNA (or at least not more than in humans).

There is evidence that ants are are able to change fairly sensitively changes in humidity and atmospheric pressure. But I don't think folks have been able to show accurate rain prediction behaviour. But again, I don't really see the connection to AI here. It really seems to be an attempt to find tenuous links without really outlining what the actual question is. For example, emergent properties of ants and other social insects has been long investigated, though we certainly do not fully understand them. They have already been used in a lot of other applications, including neuronal networks and there are experimental applications to traffic control. I am sure there are some thoughts for AI usage somewhere but OP has been so vague that I am not sure what it would be. Conversely some of the examples given in this thread are rather strenuous links which, in my mind, make it less clear what the idea behind this is.

AI is well suited to solve math problems, as they are well-defined systems. Biological systems are not. So that particular example was not particularly strong, given the current level of evidence.

https://www.cnn.com/2026/09/18/politics/us-military-ai-false-intelligence-china-ship So this is now a story how AI errors could have disastrous consequences if there are no safeguards between AI generated analyses and subsequent action. This follows a host of other pushes to use AI-based methods in e.g. law enforcement and other areas which will likely already have real-life issues. The big question here is what kind of safeguards would be feasible in the first place and would folks be willing or able to implement them?

My guess is that the solution introduces a bottle neck with a pouring lip to make pouring easier. There are also paint spouts which are just bits of plastic that you can clip to your can for the same purpose. There are also full lids that have a similar setup as those in OP that you can buy. However I mostly just get these clip-on spouts as they are pretty universal in size and usually fairly cheap. Edit: If I want to feel like a real genius, I would design one in Autocad, 3D print in the fanciest way possible, spent much more than getting the stuff from the store and forget to file down some crucial bit so that the paint goes everywhere when I pour it. I haven't actually done that, but it is possible that my wife made a convincing argument against that route.

Let's be clear about that, in its current state, it really can't. There has been a recent paper using AI trying to design viruses. Specifically bacteriophages (i.e. viruses which only infect bacteria). While it could be a consequence of how it was trained and perhaps not using the right models, but the outcome was rather poor. Fundamentally they created thousands of variations of existing viruses. From the many sequences, researchers tried to re-create some and most ended up to be junk. Only 16 if 285 that they synthesized did something. There was some recombination going on, but the backbone remained conserved. I.e. if some students had just sit down, down alignments, kept conserved domains and randomized the rest, the result would have been relatively similar. So as a whole, the existing evidence (for now) does not point to much biological advances here (again, could be the wrong approach/model/whatever, but I have not seen any other studies attempting it). And of course, these minor changes are a huge difference to modifying humans. For that we already have tools, probably not in the way you might think about it, but in the area of gene therapy. And this is already done with Lentiviruses. Not quite sure what AI immediately brings to the table except perhaps speed up some existing workflows.

This is the issue if one forcefully tries to connect random things without being able to explain the connection. For example countries with the lowest suicide rates include Egypt, Syria, Jordan, Lebanon, Saudi Arabia. Those countries also have the lowest pork consumption. Therefore, we should stop eating pork to reduce suicide rates.

That's not fair. There is a broad range of uncritical ignorance. I think it goes beyond then. There is also a strong resilience to learning or aligning with facts. Spin is fundamentally presenting information in a way to influence perception. But these arguments are entirely decoupled from facts and information with zero engagement in critical questioning.

Won by three seats. The good news is that SD lost votes, so there is some hope. But things remain volatile.

A poll doesn't determine need, it determines wants. And often the wants are decoupled from reality. For example, if folks believe immigration is detrimental to the economy, but are unaware that for the most part the opposite is true, especially over longer time periods, how would you want to enact a policy that removes immigrants, but somehow doesn't crash the economy? I.e. there are decisions that should be done with some knowledge on the matter, yet if you just go by feeling, pretty bad things might happen. There is a reason why populism has a negative connotation in traditional political discourse. You do not understand representative democracy, then. A part of it is that representatives are supposedly gatekeepers of the laws and constitution. In your proposal you seem to suggest to remove all guard rails. In your model you seem also remove separation of powers by mixing at least judicial with executive powers. The proposal pretty much bypasses centuries of experiments and thoughts on democratic systems. And especially at a time, where the public is increasingly prone to misinformation with critical thinking skills in decline, this is seems like a really good way to accelerate our demise. Meanwhile, I am looking forward to a 0% tax rate with doubling of all services while no one is working to provide them.

Explain who would design those referenda? How would those referenda be executed? What happens if results are contradictory (e.g. demand an increase in spending, but also reduction in taxes)? What if folks vote to replace fiat currency with crypto and then the economy crashes? IOW, what would be safeguards for really stupid decisions? There are some in most democracies and the ultimate control is the ability to replace the government. If the government is, as you seem to describe, mob rule, who is responsible if institutions are randomly dissolved simply because the majority didn't know what they were good for?

That is why, unfortunately, a lot of quantitative data is collected in artificial and controllable environments. However, it is well-known that those are prone to artifacts and in many cases do not represent what is happening in nature. I am sure the discussion is fairly old, but at least in the last decade or so it seems that the warnings are a bit more prominent than they used to be (certainly compared to when I was studying ethology as an undergrad).

Honestly, pretty much anything in biology requires a mind numbing amount of data. Looking at more recent paper specifically in behavioral sciences seems to add yet another slightly dimensions, which is basically learning. It touches on two of his questions, but they focus necessarily on population level effects in order to focus on the evolutionary context. As such, in most areas of biology we tend to look at general principles that can be extended over populations or species as the organizing unit. However, in my read, behavioural sciences have become more granular, and while they still deduce general principles, they increasingly acknowledge individuality in animals that are dependent on unique experiences. While close, they are in my mind not fully embedded within Tinbergen's framework and are probably most closely to what is considered to be ontogeny. It is a bit of an expansion of the more classic, reductionist view and opens up even more complexity. I am not yet sure where the field is going but I think it helps to a large agree to explain some phenomena that are often simply not measured in traditional ethological studies.

That is absolutely a weakness of most of these hypotheses. They tend to be modeling approaches to see if that would work in principle, but they are often based on very little biological data (if any). Most publications aim to build hypotheses so that other groups would to the actual work (and the former get most of the glory...;P). But I think this is generally true in biological sciences, modelling is important, but it always lags behind the experimental or observational data to support or refute them. This is why at least for now AI is making limited inroads into biology, as we have still a dearth of information and understanding of these systems. There are a few papers including on birds of paradise in my memory who have tried to measured some parameters. I don't recall them offhand, however as I believe the data set provided only limited conclusions. But maybe it is just my recollection- I have only skimmed those (outside my field of expertise). Edit: just saw a paper still in my files: Proc Biol Sci. 2022 Jun 1;289(1976):20220444 https://doi.org/10.1098/rspb.2022.0444 This is not exactly what we were discussing, but it is a meta-analysis of studies where the effect of sexual signals on predation are studied (though I cannot comment on the quality of those papers). They did find wide variation which they interpreted that the cost of sexual signalling is very context-dependent and might in some cases carry no cost or even reduce predation chance. There is also a follow-up discussion on predatory eaves-dropping by olfactory and acoustic signals, whereas visual signals showed no higher risk. But in summary, cost appeared to be highly context-dependent, making the cost (handicap) as the primary principle to assess honesty rather unlikely.

This is a bit the crux of the matter and the question is moving now a bit more from sexual selection to more complex behavioural sciences. There, the discussion is actually even more complex as behaviours might not be as deterministic as many evolutionary models suggest (or require). I think what that posted as well as some other papers are debating is whether the cost itself is necessary as a signal of genetic suitability and this is where the models fall a bit flat. What some papers focusing e.g. on mating behaviour in birds of paradise describe is closer to what is argued in the paper, i.e. elaborate courting is contingent on design features (e.g. is the signal visible and effective in courtship) and constraints (e.g. mechanical limits and resource optimization). I.e. having a high cost is not a signal per se and the trait does not e.g. converge to maximum waste that still does not reduce net survival (at least not prior procreation). The second part, whether the display is perceived as honest (i.e. an accurate display of mate quality) then requires a different model to explain evaluation. Two recent ones are the balance and trade-off model. The former argues that individuals need to allocate resources between mating signalling and viability. Fitness is ultimately determined by taken these factors together. Hence, fitness is determined by mating success as well as viability and individuals finding the optimal balance are ultimately successful. Here honesty is not determined by waste. I.e. imagine a low-quality individual who wastes already limited resources towards signaling to imitate a high-quality individual. The net allocation would be unbalanced and net fitness will drop. In some ways it is extension of the handicap model, but not looking from waste, but from a resource allocation perspective. The trade-off model is more generalized and states that individuals in different conditions face different fitness trade-offs. As the respective optima vary, the signals have to adaptive to these conditions. Under these parameters, deception becomes net unprofitable. I.e. we are looking essentially at a condition-dependent trade-off between marginal benefits and costs. As TheVat pointed out, the optimization of these parameters are honest by by pushing the trade-off to the individual optimum.

I think one should step back and contextualize this discussion a bit. It falls under the broader range of sexual selection, i.e. how can be positively selected due to mate selection. I.e., under this umbrella the trait in question also has to be heritable. OP's scenario falls under communication but not really under sexual selection, as it is not established that drinking is a genetically determined trait. The handicap principle is an attempt to explain how and why sexual selection of certain seemingly unfavourable traits occur. I am not going to repeat the assumptions that have already been discussed but I will point out that the model by Grafen was flawed and rather than deliberately wasting energy to create honest signals, they actually optimize investment based on condition. This explained in the provided paper:

Going forward I encourage to provide such details in the post plus putting in relevant sources as per https://www.scienceforums.net/topic/133848-policy-on-aillm-use-on-sfn/ I am not wearing a mod hat for this post as I am considering whether I should respond to the content of this thread.