A Bear Case: My Predictions Regarding AI Progress — LessWrong

[rcjordan]

worth a read, though Debbie disagrees on the job loss issue.

If it's TLDR, just skip down to the 'Real-World Predictions'

https://www.lesswrong.com/posts/oKAFFvaouKKEhbBPm/a-bear-case-my-predictions-regarding-ai-progress

[Travoli]

I am hopeful that they significantly advance biotech. The protein folding research is pretty incredible and potentially enormously valuable. I agree it doesn't seem like LLM technology is going to achieve true AGI.

[ergophobe]

I kind of skimmed that, but there is some interesting stuff and the comments are more interesting than the article. The problem is that I don't know enough to know whether the people who really seem to know their stuff actually know their stuff.

Like..

Human brain holds 200-300 trillion synapses. A 1:32 sparse MoE at high compute will need about 350 tokens/parameter to be compute optimal[1]. This gives 8T active parameters (at 250T total), 2,700T training tokens, and 2e29 FLOPs (raw compute GPT-6 that needs a $300bn training system with 2029 hardware).

There won't be enough natural text data to train it with, even when training for many epochs. Human brain clearly doesn't train primarily on external data (humans blind from birth still gain human intelligence), so there exists some kind of method for generating much more synthetic data from a little bit of external data.

I think the "what kind of bubble is this?" question is still to be answered.

In other words, is it like the internet bubble, a small growing pain in a massively impactful technology, or is it like.... nuclear fusion?

So far, it has been more like fusion. But then again, heavier than air flight was also like fusion until it wasn't.