[SpicyLemonZest]

They may not be able to! It's pretty widely acknowledged, for example, that if there's some surprising plateau hiding around the corner they're both going to fail. But that could mean that they're overcharging for AI usage to get research money and sustainable rates are lower rather than higher.

[guax]

I think that for coding we're past the plateau issue. The frontier models of today are good enough and very valuable. The expensiveness in running them will eventually be solved by cheaper faster hardware.

I do hope that a day will come where you can buy the nvidia spark thingy for 5k that can run the equivalent of Opus 4.6 or 4.5 locally and that would be a massive thing.

[johnvanommen]

> The expensiveness in running them will eventually be solved by cheaper faster hardware.

How?

* Moores Law is almost over. The 5090 improves over the 4090 mostly because of quant improvements.

* even if the hardware improves, there’s a huge incentive to slow roll the next generation. Nobody wants to end up like Sun Microsystems. Sun’s used hardware was faster than its new hardware, once you considered price. Sun ended up competing with its own used equipment.

The most obvious place for improvement is RAM, network and storage.

If someone can bring more RAM onto the market, that will unstick things.

[Legend2440]

GPUs are not really the ideal architecture for running neural networks; they are heavily bottlenecked by memory bandwidth and struggle to keep all their tensor cores supplied with data.

There is significant room to make more specialized neural network accelerators with new compute-in-memory architectures.

If the brain can run 86 billion neurons on 30W it must be possible.

[akomtu]

Our brains run 86 billion neurons the same way a waterfall runs a fluid simulation with N quadrillion particles.

[VorpalWay]

There are already some companies doing specialised inference hardware, Cerebras Systems for example. Such designs are still early days and I wouldn't be surprised to see more innovation there. Though because custom silicon design takes time I expect a multi-year cycle.

For training, not sure. But even if training runs on GPUs, once you have the model the main cost is inference.

[CuriouslyC]

The whole hidden plateau hypothesis is kinda bunk, because we're already pretty far in a plateau for general knowledge/question answering, but there are many subdomains where we can push model capabilities, and as we saturate one subdomain we can just shift to another economically valuable one.

There isn't one AI intelligence S curve, there are thousands of them, and they're mostly invisible in the major benchmarks, but for someone trying to do work in that specific area of capability, the progress is transformative.

[SpicyLemonZest]

I'm skeptical of a hidden plateau, but I really think it's overconfident to assume there's not one. Remember that it doesn't even have to be a technical plateau; the effective plateau of e.g. car speeds is determined by regulations and road conditions, and far below what "frontier cars" are capable of on a controlled racetrack.

[wonnage]

That’s the scenario where we’ll all be using Chinese models