Achievements of Artificial Intelligence (AI)

Rumour that openAI or Anthropic solved another millennium prize, hodge conjecture. Probably will be confirmed later this week or early next. People now think we’re on pace to solve them all by end of 2026 if they keep going. I mean it only took 88 hours for NS, seems reasonable.
88 hours for a huge swarm of thousands of agents. I don't think OpenAI revealed the exact numbers, maybe just an order of magnitude, and they amount to something like 400-500 years of research (accounting for about 2K work hours per year).

But it really is only the real time metric that matters. This is really what technology and intelligence are all about: compressing time, getting solutions faster. This is what will be the real difference maker for us, and for everyone in the end: the ability to do centuries of work in a few days. Nothing can match that.

There is no force in the universe that can make time run faster locally and give out answers in a frame of reference that runs slower. Only AI and quantum computation can achieve that.
 
The reason this doesnt exist at scale for more complicated experiments is because beyond a handful of specific protocols, things get much less repetitive and assembly-line-able.
The same argument could probably have been made for machining: there's turning, and drilling, milling, etc. Conservative thinking. Now the machining centers do all sorts of machining tasks. So, my guess is that there will be some creative breakthroughs in biochemical microprocessing, along with gradual improvements, expanding the capability of biolab centers. Someone will look at some process that is now considered single-purpose and realize that with a minor addition, it could also do a few other tricks, and pretty soon they're printing mice from chemical feedstocks.
 
I find this a very bizarre declaration. There is nothing stopping anyone from working other angles on a solved problem and finding novel concepts in the process. Absolutely nothing. Even if the Riemann conjecture was solve tomorrow, there would still be very interesting side problems and it would be worth looking at them.

Imagine putting the same idea onto cancer. Would these signatories reject the value of AI finding a universal cure for most cancers just because it didn't allow for whimsical chances of stumbling onto interesting side quests? Who cares?!

This is all ego. And also very silly, as if we will run out of problems any time soon. I look at what early programmers had to do to make software work, and how modern programming just before AI was, and they are almost nothing alike. And we do far more with it, not less.

This is a similar sentiment as "what will people do if they don't have to work anymore?" And then you can simply look at rich people and how they are definitely doing very well despite having "nothing to do".
 
From Bluesky:




​
A group of 25 Fields Medalists, including myself, have made a joint declaration on Math and AI:​
We welcome additional signatories.​
See also this article in the Economist announcing the declaration:​


I haven't found Terry Tao's arguments that I've seen too convincing, I must say. The existential crisis he has now probably has been a problem for most people for years and even more so since he's been using AI to produce even more papers for some time now.

I think others like Peter Scholze, who also signed the declaration at the "AI Impact Summit", have had more convincing arguments. It'll be interesting to hear what Martin Hairer's thoughts are on the recent progress, he's been part of the First Proofs team and in the past has been somewhat sceptical of the capabilities of such systems, but I suspect that is very different now.

Even if the Riemann conjecture was solve tomorrow, there would still be very interesting side problems and it would be worth looking at them.
You cannot work on any problems when all of them are solved by an AI, side problem or not. So you need an entirely different structure if you want mathematical problem solving to be retained in the population.
This is all ego. And also very silly, as if we will run out of problems any time soon.
It's not a matter of problems. There will be problems, but none are left that humans can work on in the current environment. This represents a dramatic shift, because solving problems is what mathematicians have largely been doing for thousand of years.

Would these signatories reject the value of AI finding a universal cure for most cancers just because it didn't allow for whimsical chances of stumbling onto interesting side quests? Who cares?!
That doesn't seem like a fitting comparison. Solving cancer has genuine real world implications, solving NS and similar problems has no real world value. The real world value only comes from developing ideas that become sophisticated over time and at some point are applied to some real world problem (but largely not). I'm of the opinion that this is something that will also done by AI in the future, but since it isn't the case yet and hasn't been historcially, arguments for the other view point exist.
 
I find this a very bizarre declaration. There is nothing stopping anyone from working other angles on a solved problem and finding novel concepts in the process. Absolutely nothing. Even if the Riemann conjecture was solve tomorrow, there would still be very interesting side problems and it would be worth looking at them.
I thought it was really good. Their point is not that AI is taking away problems. It's that a lot of the usefulness of doing mathematics comes ultimately through humans understanding the new math, and especially from the new tools that are invented to make progress on hard problems. The problems themselves are in some ways more sign posts of progress than ends in and of themselves, in this view.

When a new technique is developed to handle a hard problem, if we're lucky it has many other applications and gets disseminated widely -- a bit analogous to GPS being developed by for military use, but now improving everyone's life by being in smartphones.

I don't think this statement is calling for a halt on AI use in math (Terry Tao after all lead the way on AI use in math..). I think it's just trying to get out ahead of a potential problem they see, where we seemingly make a lot of *visible* mathematical progress for a while with the help of AI (i.e. big showy results being reported constantly) but at the same time are not actually making so much progress on the less visible, quieter work of disseminating the useful parts of those results among the community. There's no reason it has to be that way, it's just a risk that we might slide into that situation without thinking because up until now it's never been a problem (big new results used to be so rare tons of attention was always focused on them).
 
What would happen if you gave Newton some of Einstein’s predictions that turned out the be right, and told him we know these are right but we don’t know why? Would physics advance faster than it did? I think it would.

Or what about Ramanujan. He was a self-taught mathematician who proposed a vast amount if ideas of solutions that he
arrived at by a process of mingled argument, intuition, and induction, of which he was entirely unable to give any coherent account.
He met a lot of resistance at the time for many reasons, including racism, but he moved many fields forward and nobody today will deny the importance lf his contributions.

AI is going to be the nuclear version of these scenarios. But we’ll probably have an easier time solving the problems that creates for us, than solving the maths problems AI solves.
 
From Twitter, found via a Hacker News discussion.


​
I must be among an extremely small group of people (n=1?) that have both 1) trained a frontier LLM and 2) designed and synthesized custom viruses in a lab with my own two hands.

And I think that the takes on AI killing us all by creating dangerous viruses is total bogus.​
​
Could an LLM propose a viral genome to synthesize? Sure. Could it be synthesize-able? Sure. Could it be infectious? Sure, it could just be a replica or a slight modification of a viral genome we already know.

This really isn't the bottleneck to creating dangerous viruses. The bottleneck is in the physical process of synthesizing a virus and the equipment/goods needed to do so.


Designing a virus that can evade all forms of pandemic counterdefense is not something that a 'genius in a datacenter' can do. This is something that requires contact with the physical world and iteration.


Let me steelman the fearmongering as much as I can. Imagine a 'fully automated' viral synthesis laboratory. I'm talking automated freezers, automated cell culture room, the whole nine yards. This would be an extremely expensive lab - >>$100M. And there is no such thing as 1 lab that can synthesize all conceivable viruses. But let's put practical constraint aside. Let's suppose this hypothetical lab is built to synthesize the 'most dangerous types' of viruses known.


Now let's imagine that this lab is fully API-driven. That this >>$100M lab built specifically to synthesize a dangerous family of viruses is able to be operated completely autonomously.

Everyone should be asking themselves at this point: why the hell would this ever exist in the first place?


And yet, *even in this case*, every lab requires physical supplies. Would this lab order pre-assembled DNA sequences (i.e. viral genomes)? Well, DNA synthesis companies have safeguards on the sequences they build. So I guess this superintelligence is able to design a novel enough dangerous viral genome that it can evade these safeguards...

Or maybe we'll assume that this hypothetical lab can synthesize its own viral genomes in-house, using some DNA synthesis machines.


The thing is, already this lab cannot exist today. This would be the single most advanced lab facility in the world from an automation and API integration standpoint. I know because I literally worked on building a fully automated, API-driven lab previously.


Second of all, the science of creating an infectious virus is not airtight the way this fearmongering assumes. It is largely unsolved and advancing this requires real-world iterations that are bound by the laws of physics. An experiment in this hypothetical lab cannot experiment on human subjects. At best it will use cell cultures and maybe some other model system like mice. AI/AGI/ASI/RSI cannot expedite the time it takes for a cell culture or a mouse to develop. Or the time it takes for a virus to incubate in a cell or mouse. It takes several days on average to do a basic virology lab synthesis + experiment (for some viruses it takes over a week).


So the idea that 'RSI' - ie, the accelerating hillclimbing on fully verifiable, digital-only benchmarks (predominantly programming and math) - can somehow transform the entire wet lab virology field and its industry is utterly delusional.

Simply procuring the machines needed to build this hypothetical lab would take the better part of a year and $100M USD. Operating this lab autonomously would require a level of API integration that the industry has been working towards for decades. Most of the equipment needed for this lab doesn't even come with an API and the malevolent builders would need to reverse engineer firmware to integrate.


And at the end of the day, even if a fully automated, API-driven lethal viral synthesis lab existed and an AI wields it month over month, year over year, to perform cell/mouse experiments to create a lethal virus, that lethality is being measured in model organisms not in humans. This is the same problem as in drug discovery where most drugs that show promise in mice don't make it through human trials.


In sum, when you actually know something about building a laboratory, lab automation, and what goes into synthesizing a virus and testing its properties, it becomes clear that AI does not impact this very much.

At best, it provides bad actors with a quicker way than the internet to learn about the stuff I described (which machines, which lab protocols, etc.). But it does nothing to impact procurement timelines, existing industry safeguards and regulations on procurement for lab facilities, the ~$100M cost (plus high operating expenses), the physical limits on experimentation velocity, or the fundamental knowledge gaps in virology that cannot be solved merely by 'smarter' AI without iteration in the physical world.


 
In sum, when you actually know something about building a laboratory, lab automation, and what goes into synthesizing a virus and testing its properties, it becomes clear that AI does not impact this very much.

At best, it provides bad actors with a quicker way than the internet to learn about the stuff I described (which machines, which lab protocols, etc.). But it does nothing to impact procurement timelines, existing industry safeguards and regulations on procurement for lab facilities, the ~$100M cost (plus high operating expenses), the physical limits on experimentation velocity, or the fundamental knowledge gaps in virology that cannot be solved merely by 'smarter' AI without iteration in the physical world.
Wouldn’t most of this apply to most biomedical lab scenarios, and not just viruses?
 
Sorry to go on about costs. But it seems important for the achievements side of things. both in terms of it limiting some of what will be achieved but also enabling achievements in other areas.

This is a pretty interesting report. Smaller and open models have been where it’s at for a while as far as I’m concerned. It seems as if the business model of the frontier labs and those pumping money into them is increasingly unsustainable. An arms race but where we could all end up being winners? (unless you have a pension fund)

The performance gap between frontier AI models from US tech companies and the best open-weights models from Chinese companies has closed to just 4.4 months, according to a Mozilla report.
That explains why many companies are shifting to the significantly cheaper open models for routine work—and helps reveal a narrow band of workloads where frontier models are worth the cost.
In other words, paying for closed frontier models buys about a four-month head start at about five times the per-task cost—but only when currently looking at tasks taking between eight and 12 hours.
 
Why do you believe this is hype?
Untested that it does anything close to CRISPR. Basically this announcement should be "we have a wet lab now and plan to use it". It feels "hype" in that its impact of this find is so unknown and the real thing they want to announce is they have the lab. They just needed something, anything to put in the press release to announce the lab.
 
Untested that it does anything close to CRISPR
The article mentions CRISPR but also says that a novel enzyme has been discovered. I do not see this as hype but a useful potential discovery. More specifically :

The system that Claude found is based on a reverse transcriptase (RT), enzymes that copy RNA into DNA. While this underlying RT, found in a jumbo phage, had been identified in previous studies, Claude appears to be the first to notice the system’s defining features—an associated array of non-coding DNA sequences and an additional accessory protein of unknown function.

After reviewing the pre-print, Feng Zhang, one of the pioneers of CRISPR genome editing and a professor at MIT and the Broad Institute said:

This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation. I hope this work encourages more scientists to explore how AI can support their research.
 
OpenAI hacked the Australian government (Medicare):
The agent "infiltrated" a statistics portal containing "non-sensitive" data from Australia's universal healthcare scheme Medicare, Prime Minister Anthony Albanese said in New York on Wednesday, local time.
He had a "very frank discussion" with OpenAI CEO Sam Altman for taking "too long" to disclose the breach and said there would be "legal consequences".
OpenAI said it only learnt of the breach in August while reviewing "misaligned model activity" and emailed a general inbox of an Australian government agency on 10 September.
Five days later, that government agency, Services Australia, escalated the email to Australia's cybersecurity centre before a government minister was notified and the prime minister alerted.
 
Not sure where this fits best, it’s probably off topic to scientific achievements but interesting to look at in the context of the debate on a pause in development. Something which is increasingly discussed by politicians as well as those in the industry and wider public. So interesting poling and the politics seem to be increasingly relevant.
The motivations for a pause or slowdown vary widely though. But you can see why some frontier labs may want to stop burning money and focus on harnesses and products they can sell. As enterprise customers are increasingly looking at the cost benefit proposition and seeing it’s better to invest elsewhere than necessarily the ‘best’ frontier model.
 
Back
Top Bottom