Achievements of Artificial Intelligence (AI)

If our only hope for a treatment on a non glacial timeline is that AI will magically fix everything and solve drug development, we are all fucked.

We need trials of existing drugs now. Done by people. We need accelerated approval for new drugs, like with covid vaccines. We don't need AI for that. We can't rely on the propaganda out of a desperate tech industry coming true and saving us all. LLMs are not what Sam Altmann and his phony whistleblowers pretend they are. They will not save the world or destroy it (apart from speeding up climate change and destroying the creative arts). They are just a tool, and an extremely flawed one at that.
 
[Wrote the following a while back and never posted it. Thought I should probably clarify my position on AI beyond the above cranky comment. If AI can help with finding already approved drugs to trial, or new drugs to then great. I remain skeptical. But we didn't need AI for the accelerated covid vaccine rollout and I fail to see why a disease of this magnitude with a financial impact and potential payout for drug discovery as large as this has should be any different once governments and pharma recognise the problem as being real and solvable. Anyway, here's what I wrote before:]



I'm interested and surprised to find many people on here who's opinions I respect seem so optimistic about the potential of AI (specifically meaning LLMs) in science. I admit I've been reading and listening to a lot of skeptical stuff about AI, and my direct interaction with it has been through the enshittification of nearly every online space, app and search engine because of it.

As a writer and musician I am furious with what LLMs are doing to the creative arts, and the fact it needed the biggest act of plagarism/copyright infringement in history to do so. Creative LLM use is the ultimate manifestation of a society that does not respect or properly reimburse the labour of artists.

I do enjoy the pompous parsnip images though. Maybe that's my one acceptable creative use case!

The glorified chatbots the general public interact with daily really concern me. Their sycophancy and the way they have enabled some people's psychosis etc is frightening. They are pretending to be sentient. That was not a necessary component of LLMs but a deliberate one by big tech, and one with predatory and destructive consequences on people's psyches.

And we haven't even gotten into the catastrophic environmental cost of the data centre buildouts and their power needs.

In terms of scientific potenial, it sounds like it AI is perhaps more useful for mathematics than other disciplines, although I don't understand how Lean works and how it verifies answers to problems - surely we need people to use the answer or logic derived from that answer practically in some kind of applied context like engineering or physics technologies before we can be certain if the answer stands up?

Of course the protein folding stuff has been encouraging and hopefully will lead to real progress. On the other hand I have read a lot of people who work in computer science saying the opposite about their use in that field than whats been said here - essentially that they make a lot of mistakes, hallucinate and do random stuff ouside of the given prompts that humans have to go through and fix etc. The discussion on here about the limitations of LLMs to drive progress in biology is I think instructive as to the fact that a lot of LLMs vaunted potential quickly evaporates when faced with the practical realities of making progress in a field.

The way the supposed potentials and hazards of this technology is presented to the public by Altman, Amodei et al seems outright decietful to me.

Whatever the benefits or otherwise of LLMs, the heads of these companies act like con artists and cult leaders, promising that LLMs will imminently become AGI which will solve all of our problems if we keep giving them money and building their data centres. AI has become 'too big to fail' in tech and finance and is being forced into every aspect of our lives because Silicon Valley has put so much money into it, rather than because it's useful for everything. And the financials seem really dodgy and it does seem like we're in a bubble. When that bubble bursts and what happens after is up in the air right now. I don't think LLMs will dissappear entirely of course. But I dont know that they will be shoehorned into every aspect of life as they are right now.

One thing I would say is that the the idea that AI is going to become super intelligent and kill us all is just a form of twisted marketing hype. The companies want their products to be seen as as powerful and godlike as possible. It's incredibly unethical but they are very dishonest in how they market themselves. The tech people who do believe AI will become super intelligent have a kind of religious fervour. It's not something we should take seriously imo.

Obviously I would love to see leaps and bounds made in medical research and elsewhere in science. I'm just skeptical that LLMs as opposed to advanced forms of the kind machine learning we were already familiar with before LLMs launched, combined with old fashioned human ingenuity, will be what brings it.



Cory Doctorow on the 'criti-hype' LLM companies and AI enthusiasts in the media engage in:

Pluralistic: The ordinariness of evil (19 Aug 2026) – Pluralistic: Daily links from Cory Doctorow

pluralistic.net pluralistic.net

Cory Doctorow talking about LLMs on the Dreaming Against the Machine podcast:
www.dreamingagainstthemachine.com

Episode 11: Reverse Centaurs and AI, with Cory Doctorow

This week, Adam talks with Cory Doctorow, author, blogger, and tech policy advocate. Cory explains why your boss wants to replace you with an AI, why that won't work, and why they're trying to do it anyhow -- and what you can do about it. Cory and Adam also
www.dreamingagainstthemachine.com www.dreamingagainstthemachine.com

Nikil Suresh on AI in buisiness:

AI Mania Is Eviscerating Global Decisionmaking

The reality of how much damage AI mania is doing to our ability to run institutions effectively.
hermit-tech.com hermit-tech.com

Cal Newport on the 'Hugging Face Attack'
calnewport.com

Did OpenAI’s New Model “Go Rogue”? - Cal Newport

A couple of weeks ago, the AI company Hugging Face announced that they had discovered an intrusion into their production infrastructure. They didn’t know the ... Read more
calnewport.com calnewport.com

Ed Zitron on the recent 'AI will kill us all' media blitz
www.wheresyoured.at

AI Is Already In Dangerous Hands

If you liked this piece, you should subscribe to my premium newsletter, and you can subscribe on the following links: $70 a year, $18 a quarter, or $7 a month. In return you get a weekly newsletter that’s usually anywhere from 10,000 to 18,000 words, including vast,
www.wheresyoured.at www.wheresyoured.at

Ed Zitron and the FTs investigation into OpenAIs finances:
www.wheresyoured.at

Exclusive: OpenAI Losses Increased Nearly 8X in 2025, With Spending Hitting $34 Billion

Soundtrack: In Flames - Colony To further support my independent journalism, please subscribe to my premium newsletter. It’s $7 a month or $70 a year. If you’re subscribed to the free newsletter and logged in, you should see at the bottom right hand corner of your screen a little
www.wheresyoured.at www.wheresyoured.at

 
chillier said:
@EndME you posted a summary of a critique of a paper at some point recently done by I think gpt astra and I thought it seemed quite impressive. There's been talk maybe on this thread or elsewhere of posting reviews and hypotheses from AI models and how to handle that - I know @mariovitali you are keen on that. I know these things are becoming extremely powerful, though when I ask free models like gemini for example about hypotheses I get a version of the currently popular ideas/memes around MECFS aka some constellation of energy metabolism, mitochondria, autoantibodies, neuroinflammation, viral reactivation and so forth.

I'm curious when you ask the frontier models this question do you get a different/more plausible output than that? I'm curious about the capacity for frontier models to do a literature review. If you instructed it to only focus on the clinical temporal symptoms of the disease, primary literature of experiments and epidemiology done in ME, CFS, chronic fatigue, and maybe also the primary literature and clinical manifestations of other diseases. Could it pick up on useful things do you think?

On this forum lately there's been a lot of picking up of old, potentially very relevant and reasonably carried out research like the buspirone stuff forestglip has been picking out. Would an AI agent trawling through this material be able to find things?
Click to expand...
For literature review the Frontier models are extremely impressive. You can easily ask it to go through a paper and it will go through it with a lot of care. That means it will go through every single reference and verify if all the statistical analyses in those papers were performed appropriately and then also look at other literature etc. So even with little guidance remarkable things can be accomplished in those areas, which humans could never perform due to the amount of tasks. Literature review will be no problem, especially if you use some additional tools on top (harnesses, agents etc). In my experience it also tends to find references that have very little citations and which you'd probably never find if you weren't explicitily looking for it. However, there's often problems related to older papers not being accessable though, which causes issues and requires some amount of intervention (I suspect that it probably can access a lot more than it legally wants you to know).

My impression with the Frontier models lately has been that they are not very good for the type of very vague and general hypothesis building yet. However, in my experience they tend to be rather careful, so they are more likely to point out errors and restrictions in the common/popular memes rather than support them. So it's can be useful in a way to question your own ideas and find negative evidence. At least that's what I've experienced for ME/CFS related concepts. I've actually had some discussions with AI related to such ideas as well (I'll add a post of my conversation), and it seems to agree that it's greatest skills are currently in more specific fields and that for very general automated hypothesis building more progress in additional fields will be necessary.

chillier said:
On this forum lately there's been a lot of picking up of old, potentially very relevant and reasonably carried out research like the buspirone stuff forestglip has been picking out. Would an AI agent trawling through this material be able to find things?
Click to expand...
I think so, certainly with some instructing. I'd be more worried that it finds so much that'll be hard for humans to put everything into context. When I instructed it to go through the review by forestglip, I already got too much feedback for my taste.
 
chillier said:
That sounds ideal, and sort of the exact opposite experience that I described with the free chatbots.
Click to expand...
Yes. Before the newer models came out all I ever got was boring agreement, but in my impression things look different now. For example: Astra thought that the review by forestglip was actually too "confident". But of course you never quite know how the next model will turn out to be and the output is initself somewhat stochastic.
 
ME/CFS Science Blog said:
1) We’ve written a new article about the spectacular progress in artificial intelligence (AI), the billions that AI companies are investing to transform drug discovery, and why this is a reason for hope for ME/CFS patients.
https://mecfsscience.org/the-ai-revolution/



Here's a brief summary

2) We start by highlighting how AI models have recently solved some of the hardest mathematical problems, finding insights that eluded the greatest minds of our generation. They have also become better at coding than almost any human professional.

3) We use quotes from experts in math and computer science because they are the canary in the coal mine for what is likely going to happen to many other fields as well, including science and medicine.

4) We suspect that this important development hasn’t really gotten through to most people for several reasons. First, there’s so much hype and overstatement about AI that it’s difficult to look beyond the slop.

5) Most people encounter AI through fake images or bullshit-text from spammers and crypto-bros.

Second, people might have tried an AI model in the past and never updated their assessment (they’ve truly gotten a lot better)

6) Third, most people only use AI superficially, for example, to create an image or do a quick internet search using a free version. Running multiple agents using the top-tier models gives much more spectacular results.

7) Fourth, there are big commercial interests at play, so news articles about discoveries are often seen as a strategy to increase the hype. Unless you’re in coding or math, it’s likely that you haven’t seen the full power of AI with your own eyes.

8) AI companies such as Anthropic, OpenAI, and Isomorphic Labs are investing billions into drug discovery. The time and money needed to develop drugs could be cut dramatically if AI models can accurately predict what effect they’ll have in the human body.

9) It’s easy to be cynical about such grand ambitions, but at least the greatest minds of our generation are focused on solving disease instead of designing credit default swaps or advertising algorithms for social media.

10) Progress in medicine will likely not be as fast as in coding or math. Biology is not a closed system: whatever an AI predicts still has to be tested in the lab and in patients.

11) While it might take longer than in math and coding, we suspect that AI will also transform science and medicine. The goal of this article is to raise awareness of this development in the ME/CFS community so we can take advantage of it and be ahead of the curve.

12) We also think these developments are a reason for cautious hope for ME/CFS patients. The decade-long timeline for new drug discovery may be drastically shortened in the future.

13) In the nearer future, AI models can help identify already-approved drugs that may be relevant to ME/CFS pathology and unravel the meaning of genetic findings such as DecodeME.

14) None of this is meant to overlook the many downsides and dangers of AI, but ME/CFS patients are among the wretched of this earth, and they have the most to gain from an AI revolution.
Click to expand...

Lovely piece! I think it captures things very well.

Amongst your 14) points, the point you last made in your piece, a hypothetical 15-th point, stands out to me: We can and maybe should prioritize gathering useful data now, even when the analysis of the future will look very different to today.
 
V.R.T. said:
I'm interested and surprised to find many people on here who's opinions I respect seem so optimistic about the potential of AI (specifically meaning LLMs) in science. I admit I've been reading and listening to a lot of skeptical stuff about AI, and my direct interaction with it has been through the enshittification of nearly every online space, app and search engine because of it.
Click to expand...
If you’ve not seen it this may be of interest. https://s4me.info/threads/achievements-of-artificial-intelligence-ai.36582/post-733666

I’d say the things you say are problematic and the things people say are hopeful can both be true. Few things in life are all one thing and AI is many things. To me the disappointing thing is how divisive things have got and how quickly but that is probably why, its become easy to pick a side and dig in. That has I think very much been down to a particular approach of people like Altman. They hold most of the responsibility here and have in my view gone in particular directions often for their own self interest despite saying otherwise. But it’s worth remembering some of those loudest voices at the other end of the spectrum are not entirely without selfish motivations too (an engaged and riled up audience is good for numbers)

I hope we can have the conversations we need to as societies about how we want to use these tools, how we don’t, on what regulation is needed, etc. but right now that seems sadly impossible. Then again some of us wanted that conversation 5+ years ago and it seemed impossible then because of different reasons!
 
EndME said:
For literature review the Frontier models are extremely impressive. You can easily ask it to go through a paper and it will go through it with a lot of care. That means it will go through every single reference and verify if all the statistical analyses in those papers were performed appropriately and then also look at other literature etc.
Click to expand...
I think they have become ok-ish at it.
I tested it with one of the comp-sci papers I've written and asked Opus 5.5 to do a review and find strengths and weaknesses.

You may say I'm biased here, but I am painfully aware of the weaknesses in that paper. And I needed a paper that I know inside-out and is in my area of expertise.

It came back with 8 weaknesses of which:
  • 2 were outright wrong
  • 1 almost touched an important issue but the way it was worded was just wrong, so no display of understanding here
  • 2 would qualify as useful feedback but still contained (obvious) errors
  • 3 were correct and useful, of which 2 were minor
Upon pushing back, you get the usual "You're right, and I withdraw that point" (at least for the wrong ones).

Of the 5 strengths it found, 2 were genuine, 3 were maybe genuine but very generic.

If this was a peer-review, I'd give it 3 out of 5 stars, but only because the bar is really low.

What those models are really incredible at is finding minor inconsistencies of all kinds (e.g., it says "74.68" here but "74.7" there and so forth). But I think even frontier-models are still not be trusted without double-checking.
 
Last edited:
fst said:
You may say I'm biased here, but I am painfully aware of the weaknesses in that paper. And I needed a paper that I know inside-out and is in my area of expertise.
Click to expand...
I would say you're reporting your real life experience with Opus 5.5, which for example I've never used before and probably many on the forum haven't either. So it makes for a more accurate reflection of what various models can and currently can't do.

When I ask Astra (at peak capabilities) to review something I know something about, I get a lot of extremely valuable feedback. Things I have missed despite putting months of thought into them. But possibly this is also currently a bit specific to my area. At the same time in areas I know less I get so much feedback that I have no idea what's going on. Human prompting seems to be the least efficient and at least for me, least enjoyable way to use our time. I suspect that currently for many people engaging with an AI to do a review feels less rewarding that doing the grunt work oneself.

I suspect the issue you report above might have been resolved by the use of agents, harnesses or a more streamlined format, or maybe just a better model, but who knows and I find it possible that the public models will currently be useless in certain areas of medical reviews due to the horrendous quality of research. It's quite possible that waiting for the models that have a training background in biomedical benchmarking, whatever that may mean, is quite sensible as well.
 
EndME said:
I suspect the issue you report above might have been resolved by the use of agents, harnesses or a more streamlined format, or maybe just a better model, but who knows
Click to expand...
Quite possible! A naive prompt can only do so much. But what I think it shows, is that "in everyday use" these issues are still common.
 
voner said:
Are there any that have a training background in biomedical benchmarking?
Click to expand...
Not any general LLMs that are public. It's also not clear to me how that would look like in the first place. Other than that the larger focus on medicine and biology is delayed in comparison to the one in mathematics. And for the very general AIs, rather than task specific ones, today I believe you need progress in robotics, but who knows, I may be wrong and my opinion may be very different in a month. It seems natural to me that you'll already get a ton of progress from more task specific things, from mathematical capabilities and from progress in many other fields. As the old joke goes, biology is just applied chemistry, which is just applied physics, which is just applied mathematics, which is just applied logic. That doesn't allow you to do everything but maybe the genetic studies we've seen are now considered a "joke"?
 
voner said:
Are there any that have a training background in biomedical benchmarking?
Click to expand...
There are various models that have been trained specifically on scientific or medical text, more for data classification and extraction, summarisation or semantic search than benchmarking as such. Maybe because of disagreement on what ‘good’ looks like or from problems tagging specialist text for training? Or perhaps the smaller specialist models have been overtaken by the general purpose LLMs where there is vastly more money for training. It’s an area I’ve only dabbled in, but did see some academic projects looking at this sort of thing.
 
EndME said:
As the old joke goes, biology is just applied chemistry, which is just applied physics, which is just applied mathematics, which is just applied logic
Click to expand...
I’m not familiar with the old joke, but that’s exactly the thinking I was having… I was thinking about that with neural synapses and signaling.
 
voner said:
I’m not familiar with the old joke, but that’s exactly the thinking I was having… I was thinking about that with neural synapses and signaling.
Click to expand...
Maybe, but the joke also only goes so far. Even the most brilliant mathematicians have no idea how to face the current problem, a problem in "applied biology" as the joke would suggest.
 
Before this paper came out I tried this at home actually for fun. Claude researched, built, and handled the environments from the protein design tools, something that would have taken me weeks. Claude is really just doing tools calls here, instead of choosing react vs node, it’s protein design tools. I had to steer for secondary validation of refoldinf but it chose a lot of the same tools as this paper. This isn’t that impressive in itself as probably none of my proteins work and you need 10000s of trajectories.

What’s impressive with very little activation energy, a few hours in plan mode anyone ( me included who knows zero protein design) can try their hand at design. This really proves to me there isn’t a bottle neck here it’s having good ideas in the first place to target and validation in vivo that is the bottle neck.

This is all to say, if you haven’t used these tools to drive ML tool calls, install pipelines, and help build other coding tools it feels like an empty promise. When in reality it’s speeding things up incredibly fast. Claude isn’t doing the discovery here, it was more of a orchestral manager.

www.anthropic.com

How Claude designed protein binders against scorpion neurotoxins

We asked an internal research Claude model to design minibinders (small proteins that attach tightly to a target) against scorpion neurotoxins.
www.anthropic.com www.anthropic.com
 
voner said:
Are there any that have a training background in biomedical benchmarking?
Click to expand...
They're lagging behind, but OpenAI recently announced they are making their Rosalind model available to some researchers. It's going to lag behind for a few months because it's so specialized, but it will happen.
 
You must log in or register to reply here.
Back
Top Bottom