Hacker News Reader

Comment by ck2 | Hacker News Reader

"AI" has limits in that it cannot invent knowledge, it can only distill and search for patterns in existing knowledge

not sure how many will get this reference but "AI" for science and math is like super-shoes for runners

at first we are blown away by the impossible improvements including sub-2-hour realworld marathon and every other PR/CR/WR is dialed down

but then the improvements slow and reach a stall point because of the limit of technology and the source of the achievement

ie. sub-2-hour marathon yes, sub-1-hour never happening (rollerblade inline-skate record is 1-hour marathon)

View on HN

I agree. The most likely scenario is that this is just a "new normal" lift that is percolating through human endeavors and will saturate at some point. For example, the whole cyber-security bruhaha should ultimately resolve into higher standards for code published -- we can now cheaply find and fix a whole slew of minor bugs that weren't worth our time before.

The fact we see a lift is not the same as evidence that the lift is unbounded.

The lift being finite is supported by the fact improvements have come at the edges: improvements from human feedback, improvements in harnesses, improvements on model compatibility with harnesses, improvements in inference efficiency with new architectures, etc. If we were just training better models from scratch that would be one thing, but we are just making better use of a tool we've developed.

Not sure what your first sentence means, or why you are quoting AI. Many of these problems individually were math at a level approaching the highest possible for expert human mathematicians. These are not simple combinations of existing ideas, or following of human intuitions, or implementing something following specific human instructions. Then again maybe you mean that math is not knowledge and that all math is simply extending the basic axioms using known patterns, to which I would not agree.

as much as I love analogy with humans using super-shoes, not everybody can be a world-class expert in their industry. There should be a place at the table for average people to take part; otherwise, it won't be sustainable.

As a programmer, I am mostly interested in whether my role is sustainable long-term and whether the models will get better. I don't feel in jeopardy yet, but two more years like this and the calculus of hiring software engineers could shift even further. QAs are already overwhelmed with work

well not every runner can run world-class sub2 marathon or even vaguely close, not even in super-shoes

but with super-shoes more and more runners are qualifying for boston marathon and even olympic trials marathon where it would have been impossible for them previously

and that's what "AI" currently does, it allows average people to immediately "pick the brain" of every expert in every field, in every scientific paper, without previously reading a single other google result, something that would have been impossible for them previously (super-shoes for the brain? too far?)

but "AI" isn't creating new knowledge, it's just stitching together existing knowledge from patterns that would have taken years by human hand if even possible at all, it's going to "hit the wall" eventually (in its current form)

How do humans "invent knowledge"? Is your argument that the answer to these questions already existed in the training set? Why didn't any human recognize that before?

humans discover a lot of things by trial and error (aka what I meant by "invent knowledge")

basically everything Benjamin Franklin did was trial and error because no-one understood what electricity was in the slightest

almost everything Edison did was trial and error too, he had his lab try thousands of materials for his long lasting lightbulb filament

even the most advanced "AI" today is just machine-learning going through everything already known trying to piece together previously discovered facts, admittedly at levels and detail impossible by human hands

but that means there are limits and it's not really "AI"

Have you seen how it debugs? It's trial and error. As the most reasonable hypothesis fail, new data is found, and it keeps iterating. It's not a human intelligence, but if it has an issue, it's not some mysterious creativity anima in the heart of man.

I have seen it produce tentative genetics for experiments, just like a scientist does. Then the data comes in and it can evaluate the data from the experiment just as well. One just had to give it a lab budget.

Reply on news.ycombinator.com