On Tue, Dec 09, 2025 at 07:49:09AM -0500, John Clark wrote: > On Tue, Dec 9, 2025 at 6:40 AM 'Tomasz Rola' via Everything List < > [email protected]> wrote: > > *> while my performance was not stellar, would I say that getting score > > required any kind of superhuman abilities? Nope.* > > > *I would say that getting a score of 118 on the Putnam is FAR beyond the > abilities of 99.99+% of the humans on this planet; so "superhuman" would > not be a completely inaccurate word to describe such an ability. *
Sure, solving more problems should be more difficult and would require bigger ability - in this particular field, i.e. problem solving. This says nothing about one's ability in other fields until one is tried in those other fields. And besides, while a horse is stronger than me, so it has bigger ability in the field of moving a ton of cargo, it is still no reason to treat a horse as some kind of demigod. Even if someone discovers that a horse may kick a baseball into neighboring town. Of course, some people will inevitably elevate such a horse into demigod position and I will inevitably make remarks how they will eat salami of demigod once it gets too old. This is not to mean that there has been no progress - there was a progress, quite impressive progress, happened during more or less single lifetime. However, each technology has limits, and if there is no breakthrough, then there is no more dramatic improvement. In particular, assuming they run LLM on computing cluster (and I cannot imagine anything else for such task, but I had not dug for the information, so I can assume only) - it is very hard to double performance of a cluster. You may see an example in wikipedia article about scalability, where is is shown how doubling cpus from 4 to 8 only gives 22% of speed increase (for very specific kind of computation, as described there). [1] Or, to put it another way, throwing more and more resources into cluster is going to give smaller and smaller performance improvements. Improving algorithms will change the serial part of computation to be smaller and will make parallel part to give faster results with more cpus. But algorithms cannot be improved infinitely. The whole talk about building nuclear plants for powering cluster to run bigger model on it seems (IMHO) to indicate "they" have hit the wall and cannot easily improve anymore. Mind you, this is a single computer (ok, cluster is your new computer, right), running a single application (LLM), and they need it to have its own power plant. Assuming they get there, can they have even bigger cluster, say, 20 times as big, powered by another 20 plants? This is going to be ridiculous, or pathetitc improvement. There should be a better way. And if no such way is found - there will be long stagnation, I suppose. [1] https://en.wikipedia.org/wiki/Scalability Of course there are going to be some interesting new uses for this stagnated tech. Some more click-baity titles. LLM will kick a football, too, always scoring perfect goal. Wild prophecies about coming era of artificial intelligence were happening for a very long time - the earliest I have read about was in 1954, if memory serves (I do not want to dig out this book right now). This means 71 years of unfulfilled prophecies. Color me bored. This time, it may be different. But so far, the only thing which seems different is the fact prophecy makers are not very trustworthy. If they were less wealthy, I would have been prone to think some of them were small time grifters. Since I have been tracing the field of tech for a certain number of years, I think I have read about CEOs giving testimonies in front of court, which I would say were not aligned with the truth too much, maybe even knowingly so, and getting away with it. Or selling product while probably knowing it was faulty. Frankly, guys similar to those earlier guys are now promising me a better future. Oh really. About mathematician being able to start a wholly new road - I think we had miscommunication. I never meant a road built nowhere, disconnected from other roads - this would have been useless. I meant more like it is easy to build road around the mountains, but one day someone builds one road straight throu them and thus a new way of mathematics is born. Or extends one road even further. Existing roads are not nullified, AFAIK. I think I am with Brent except I do not wait for coming of AI. But I will be glad to be wrong. -- Regards, Tomasz Rola -- ** A C programmer asked whether computer had Buddha's nature. ** ** As the answer, master did "rm -rif" on the programmer's home ** ** directory. And then the C programmer became enlightened... ** ** ** ** Tomasz Rola mailto:[email protected] ** -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/everything-list/aTwMm5NkIZt6nxb1%40tau1.ceti.pl.
