91–100 of 111 posts
Earlier quoted context omitted.
The amortised cost including the panels and labour is nowhere near "almost free".
It is over a couple of years
Earlier quoted context omitted.
> of circular fake money Oh it gets worse than that, the money which caused all of this by OpenAI was taken from Japanese banks at cheap interest rates (by softbank for the stargate project), and the Japanese Banks are able to do it because of Japanese people/Japanese companies and also the collateral are stocks which are inflated by the value of people who invest their hard earned money into the markets So in a way…
What is the difference between "hard earned" and not?
Maybe you can argue that yakuza is making hard earned money but imo, they are doing illegal activities within the law and are doing something more closer to extortion.
Ironically, in a sense, what AI did in a sense is also an extortion.
One is just legal (barely, I am not even sure how or why), the other isn't. That was my intention to highlight when I said hard earned money.
Despite the shortage, RAM is still cheaper than mathematicians.
I don't know, I think if you weighed up the costs of AI related datacentre spend vs. the average mathematics academic's salary you could come to a different conclusion.
We will not see memory demand decrease because this will simply allow AI companies to run more instances. They still want an infinite amount of memory at the moment, no matter how AI improves.
I'm not sure that's infinitely true as long as AI costs to the user are proportional to the cost it takes to run the model. Even if user costs are heavily subsidized by investment, as long as they are non-zero and go up when models cost more, there will be at least some pressure for cheaper models and not just more capable ones and that pressure will go up with costs. AI is a crazy industry, but it's not totally immu…
The drop in memory stocks seems counterintuitive to me. The demand for memory isn't going to go down, we'll just be able to do more with the same amount of memory.
This paper was released a year ago and was probably part of how google got to 1m context before other labs.
I mean, since GPT-4, I believe the RAM is no longer creating the miracle that the LLM performance scales directly with the model size. At least ChatGPT itself convinced me that any decent-sized company can create a GPT4 equivalent in terms of model size, but limited by service options, like memory cache and hallucination handling. Companies buy RAM simply to ride the stock hype. I am no expert, so this is a shallow t…
Earlier quoted context omitted.
It's already possible. Post-training is vastly more important than model size. (There's bigtime diminishing returns with increasing model size.)
Is there a size cutoff you would say where diminishing returns really kick in? My experience doesn't disagree, at least. I've been using Qwen for coding locally a bit. It is much better than I thought it would be. But also still falls short in some obvious ways compared to the frontiers.
No idea yet. But also it's obvious that making LLMs without MoE is stupid.
Earlier quoted context omitted.
I don't know, I think if you weighed up the costs of AI related datacentre spend vs. the average mathematics academic's salary you could come to a different conclusion.
Raising, nurturing, training, and mentoring an expert mathematician is not cheap; it never was, perhaps the first time in history when we can witness that rule to change - spinning up a bunch of math-savvy agents, each smarter than Ramanujan maybe will get too cheap.