This seems unlikely. The current model of explicitly training AIs will eventually be superseded by AIs that train themselves by reading and watching stuff just like humans do. This of course also depends on what you mean with future, I assumed not the near future.

Can you tell me what you mean by AI that trains itself? Will the AI have root access to its own data center? Is there any technology you can name that will lead to this notion of an AI improving under its own cognizance?

> Is there any technology you can name that will lead to this notion or an AI improving under its own cognizance?

Add a supervising GPT-4 instance that decides which data to LoRA-train on?

That always results in bad actors hijacking the learning process and turning the AI into a nazi for shits and giggles. ahem 4chan ahem

I was not thinking of an AI learning just like large language models "learn" today by adjusting their weights with training data. I was thinking of an AI that has the ability to learn build into it. One way I could imagine this could be done is as follows.

The AI has some memory, essentially just a big byte array. It will answer questions just like a current large language model, it will feed the input and the content of its memory [1] into a neural network and produce some response. In addition to this there would also be a neural network that generates memory update operations from the input and the current content of the memory in order to memorize information. And here I would imagine that this neural network will eventually become smart enough to decide what is worth memorizing and what should be discarded.

As far as I know we do not currently have such systems and it is not clear when we will have something like that. While what I described above seems more or less doable with current technology, it is not clear that it could actually work, that there is for example a realistic way to train something like this. Human brains, I would assume, neither do gradient decent nor explicitly update some memory cells, so maybe we are still lacking some key insights. But I am sure that large language models are not the final word on artificial intelligence.

[1] If the AI would have a gigabyte of memory, you could of course not easily feed the entire memory into a neural network at once. This would have to be done in chunks or the neural network itself would have to generate addresses of pieces of memory it wants fed into the neural network.

It will be wayyyy more expensive for current events. Matt Levine's piece on positional goods is relevant here. Or just positional goods in general.

why would it be significantly expensive?

The word "superintelligence" is doing a lot of heavy lifting in these assumptions.

But if there was a model that could make better-than-human-with-spreadsheet predictions about the moves of the stock market, that could make a lot of money for its users, so you could charge a lot for it.

Unless, of course, you had competitors with an equally good model who were giving it away for free.

the better the superintelligence, the more significant the advantage. Whoever is offering the advantage can charge whatever they want, it will be worth it.

For example most stock market tickers you see online are delayed by at least 5 minutes or more. Realtime streams are far more expensive.