• 2 Posts
  • 422 Comments
Joined 1 year ago
cake
Cake day: July 10th, 2023

help-circle

  • In 15 words: deep learning worked, got predictably better with scale, and we dedicated increasing resources to it.

    Are you sure about that Sam? Because one, you’re the snake oil salesman writing this and I wouldn’t trust you as far as I can throw you, and two, yeah maybe it scales predictably but the prediction is that training the next generation for marginal improvement will cost an exponential 100 billion (and that is taking your Microsoft discount for compute into account). You’re hitting a wall hard and the profits are still not in sight. This avenue of progress is a dead end and Sam knows it, because OpenAI is selling PPU’s instead of stock and looking to Saudi investment. Don’t get stuck with the bag folks, the few thousand days Sam claims to need aren’t survivable.















  • I’m not about to call myself the end all be all expert on LLM’s, but I’m a 20 year IT veteran in system administration and I keep up with tech news daily. I am the perfect market for new tech: I have a lot of disposable income, I’m tech obsessed and always looking for optimisations in my job as well as in my personal life. Yet outside of summaries (and even there I wouldn’t trust them) and boilerplate code that I could’ve copypasted from stack overflow I can’t think of a good reason to burn as much energy and money as the purveyors of LLM’s are. The ratio between expense and gains is WAY out of whack for these things and I’ll bet the market will correct itself in the not too distant future (in fact I have, I’m shorting NVDA).

    I understand what these plausible next word generators are and how they work in broad strokes. Have you considered that you can’t tell what someone does or doesn’t understand by a comment?

    By the way, you’re smarmy enough to tell me I shouldn’t be asking LLM’s for advice, but in the same thread you’re asking how to run a local unrestricted LLM to ask for not-entirely-legal advice? Funny that.