We are already on the cusp of fully automated reasoning, and once we have fully automated reasoning, OpenAI and Anthropic can just dedicate part of their compute towards generating new high quality novel output, which will then be fed as training data during pretraining of subsequent models.

▲

Nasrudith 2 hours ago | parent | next [-]

I don't believe that to be possible in general. Because we've already had Millenia of philosophers attempting to make discoveries through sheer reasoning and with the small in the grand scheme of things exception of formal logic failed to do so. Which leads me to a principle: No matter how smart you are, you still need the real world as a reference.

Once again LLMs will have to be bound to a source of entropy or feedback of some sort as a limit. Sure you might be able to throw terawatts of cycles at say music production but without examples of what people already like or test audiences you cannot answer the question of whether it is any good.

	▲	geremiiah 2 hours ago \| parent \| next [-]
		Well, yes, that's why the rest of science was invented, no? I did not mean to imply that AI would restrict itself to philoshical thinking and formal logic.
	▲	energy123 2 hours ago \| parent \| prev [-]
		It's proven to be possible in narrow areas like Go. There is no entropy or feedback or whatever. It just keeps getting better.

▲

qsera 2 hours ago | parent | prev [-]

That is like saying we can get unlimited data compression by feeding the output of a data compressing program into its own input..

▲

geremiiah 2 hours ago | parent [-]

Not it's not like saying that at all.

	▲	qsera an hour ago \| parent [-]
		Yes, that is exactly like it. Generating new training data from existing data will only generate patterns that already exist in the training data. It might help LLMs to capture it if has not already. But it can never generate novel patterns. For example, Imagine some kind of neural network architecture that can do OCR. You might be able to generate variations of the letters it already know using some technique and use it to better train the recognition of already known letters. But it would never be able to generate letters that it does not know.