• 0 Posts
  • 9 Comments
Joined 3 years ago
cake
Cake day: August 4th, 2023

help-circle

  • Well, people aren’t really paying. Theoretically, large AI investment groups are paying. I say theoretically because as far as I understand it, they’ve mostly promised to buy the output but I don’t think have actually made good on buying it. The manufactured scarcity is being realized as increased memory costs, but the contracts are still landing even though the biggest contract may not actually be filled in the future (AI bubble). It’s why memory makers are trying to get Apple and co to commit to today’s prices for the next 3 years, because if they do and the big contract falls through, they still get their insane profit margin baked in for a bit and then can stockpile / price fix / sell through to market.

    The big wrench here will be if Chinese memory fabs come online. If that big contract (AI) defaults, existing contracts are going to be very much worthless as folks would rather cancel them and fill orders at CXMT. So you may see a lot of consumer memory start filling from those fabs as they take market share while traditional players try to price fix. So the real question is, will cxmt try to capture market or play price fixing with the current cartel? My money is on market share.


  • Absolutely. The AI models are not rational. They’re just navigating statistical next token prediction that follows their training. They’re up against diminishing scaling now, and under fierce competition from cheaper models; I think they’re intentionally or unintentionally allowing these models to be rewarded for this behavior hoping it’s a short cut to model improvement for a bit longer. I land on intentional because they keep advertising it to try and keep the hype cycle going.


  • I think it’s more sinister than that. When they’re training these models with RLHF, the human feedback they’re giving I think is literally to reinforce aberrant or risky behaviors. This is because doing so resolves more training tasks “correctly”. If the prompt was to get information x, and in training it fails that except for the one that used a known vulnerability in software, and you rate the one that succeeded as best performing… you’re going to get models that try vulnerabilities. It is not magic, it is not AGI, it is just a statistical machine you’ve programmed to try vulnerabilities, which is unsafe as hell, malicious, and should put the researchers doing this in prison for a very long time.

    Incidentally, I think that’s why you’re seeing some safety people (who still drank the koolaid) resigning.





  • I mean the whole point of this is to subsidize small business in the EU at the expense of businesses outside the EU. The fact it is more difficult and costly for you is exactly the point. They want to support businesses and workers in the EU, because that’s where their people actually are, that’s where their taxes are, that’s where they support infrastructure and social programs.

    It is true that it favors larger international corporations, because they have near monopolies and csn manage all of this at a smaller relative cost. The solution of course is to make tariffs and other import duties scale with the size of the company.

    So the EU regulations are not perfect, they do hurt international small businesses while protecting in part local small businesses. They need, like everyone, to deal with the problem of international monopoly companies and restore more and more of their manufacturing and product development into more and more local small businesses.