How quickly will AI improve? • Hacker News Board

Re: How quickly will AI improve?

#11

Post by nprateem » Fri, Jan 24, 2025, 4:16 AM UTC

jasonhanley wrote: post #10

tumnus wrote: post #9

Earlier quoted context omitted.

Kurt Vonnegut did it first.

You know what's wild? Not ever having heard of him, I can find out who he is and how this relates to the conversation in milliseconds. For others' benefit: "Kurt Vonnegut explored themes of humans being observed by extraterrestrial beings in a zoo-like setting in his novel Slaughterhouse-Five. ... While this scenario involves humans being placed in a zoo by aliens, it does not specifically depict artificial intellige…

Great book. Well worth a read.

Re: How quickly will AI improve?

#12

Post by nprateem » Fri, Jan 24, 2025, 4:18 AM UTC

joshka wrote: post #6

grajaganDev wrote: post #2

There is no reason to think that LLMs will improve at all. They may degrade to due lack of clean training data.

The only reason to believe that statement would be that training data is finite and cannot be meaningfully synthetically generated in a way that is useful to the LLM model. If you can agree that there are certain things which can be qualitatively measured by deterministic logic (e.g. "does this build", "what is the cyclomatic complexity of this", "does this pass the unit tests", "what is the performance characteristi…

He said "may". There are plenty of other reasons to agree with the statement.

Re: How quickly will AI improve?

#13

Post by nprateem » Fri, Jan 24, 2025, 4:20 AM UTC

jasonhanley wrote: post #8

grajaganDev wrote: post #2

There is no reason to think that LLMs will improve at all. They may degrade to due lack of clean training data.

As far as I can tell, the training is still going pretty well. Once humans learn enough, they are able to start coming up with and evaluating their own ideas. This ability isn't 100% apparent with current public AI models, but I strongly suspect that this is happening behind the scenes. Certainly researchers are already using AI extensively to improve AI, and that really has the potential to go exponential.

I just don't buy this at all. I've had no end of circular conversations with AI models because they fundamentally don't understand anything.

No amount of training data will solve this. Intelligence isn't a word prediction game.

Re: How quickly will AI improve?

#14

Post by thedevilslawyer » Fri, Jan 24, 2025, 12:11 PM UTC

nprateem wrote: post #13

jasonhanley wrote: post #8

Earlier quoted context omitted.

As far as I can tell, the training is still going pretty well. Once humans learn enough, they are able to start coming up with and evaluating their own ideas. This ability isn't 100% apparent with current public AI models, but I strongly suspect that this is happening behind the scenes. Certainly researchers are already using AI extensively to improve AI, and that really has the potential to go exponential.

I just don't buy this at all. I've had no end of circular conversations with AI models because they fundamentally don't understand anything. No amount of training data will solve this. Intelligence isn't a word prediction game.

I've rarely seen this, mostly get answers correctly, and am able to use it for generation and reasoning. Can you share any link of a chatgpt/gemini conversation where this kind of circular conversation happened?

Re: How quickly will AI improve?

#15

Post by joshka » Sat, Jan 25, 2025, 7:58 AM UTC

nprateem wrote: post #12

joshka wrote: post #6

Earlier quoted context omitted.

The only reason to believe that statement would be that training data is finite and cannot be meaningfully synthetically generated in a way that is useful to the LLM model. If you can agree that there are certain things which can be qualitatively measured by deterministic logic (e.g. "does this build", "what is the cyclomatic complexity of this", "does this pass the unit tests", "what is the performance characteristi…

He said "may". There are plenty of other reasons to agree with the statement.

The core of my argument was against "There is no reason to think that LLMs will improve at all." This is a falsifiable statement if there is one or more reasons to think that LLMs will improve. I provided one such reason. I don't disagree with the part of the statement that "They may degrade to due lack of clean training data." as instances of this have already, and will in future happen. However this is immaterial to the totality of the statement.

Your argument is that there are some reasons to agree with the statement. To show that my argument is false you actually need to show that there are no reasons to disagree with the statement. In effect you're attempting to argue that because you saw some red cars means that another person's statement that all cars are red is true.

Meta argument aside, there are many other reasons to suggest that LLMs will continue to improve, the easiest of which is they have done so recently so far.