• 0 Posts
  • 107 Comments
Joined 2 years ago
cake
Cake day: October 20th, 2024

help-circle


  • LLMs have been getting better at code generation over time. The problem is that AI boosters think that AI is and will get exponentially better over time, but in reality these increases are slowing down and increasing in cost, both in terms of training and inference (actually generating the output). Each advancement is smaller and costs much more than the previous one, creating a natural limit over time.

    I’m not sure where the theoretical limitations are when it comes to code generation, but there’s also an economic limit. You can keep throwing resources at the problem, but the curve is exponential. For example, you can increase the size of the model or the context window by throwing more RAM at the problem, but there’s only so much you can cram into a system, assuming that we can even manufacture enough.


  • Thank you! Been trying to explain this for ages.

    nVidia’s current stock price reflects the idea that they’re going to continue selling ungodly amounts of GPUs for ludicrous prices. Once neoclouds fold and there’s a glut of them on the market, and hyperscalers pause AI data center construction because they’ve overextended their capex, sales slow down and so does their stock price.

    They’ll continue to sell GPUs but it’ll be far fewer of them for far less money and all those people excited to own their stock will stop buying it and the price will stagnate, if not drop, staying depressed for years.

    Not to mention all their investments in these AI companies and neoclouds will drop to zero as they begin to fold. And all the circular financing will come back to bite them. They’re in for a rough time, possibly for decades.



  • I remember when Clinton got elected in 1992, and it felt like a breath of fresh air after a dozen years of Reaganism. His Centrist campaign protected him from cries of “bleeding heart liberals” and felt like a safe way to move the country slightly left after Ross Perot split the conservative vote.

    “It’s the economy, stupid” was the catch phrase that cemented so many economic conservative policies in place while providing enough of a bare minimum safety net that people stopped pushing for real change. In the meantime Newt Gingrich pushed the party harder to the right and really helped develop the attack-at-all-costs demonization that has gotten us into partisan hell these days.

    Finding someone middle of the road in modern times gave us Biden, who made some real changes that proved fleeting and easy for MAGA to sweep away after Trump turned real anger at the stagnant state of affairs into a second victory.

    Maybe we need someone controversial that is able to push real and substantial changes to convince people that their lives are getting better instead of slowly slipping into neofeudalism. Someone middle of the road is likely to continue to preserve the status quo that is suffocating the lower classes, which will lead to another authoritarian populist fascist down the line.



  • Apollo analyst Torsten Slok said that while the Magnificent Seven has seen real profits from AI, the profit margins and revenue per worker of other companies remains flat, with no evidence that the technology is enhancing productivity.

    At the tech firms, revenue per employee has jumped from an average of $550,000 in 2022 to almost $800,000 this year, Apollo’s analysis shows.

    Has revenue per employee really gone up in a meaningful way? Or has most of the Mag 7 performed a series of layoffs that has boosted the average based on the remaining employees.

    They also claim that profit margins are way up at the Mag 7 and that’s true on paper, but I think there’s a bunch of financial shenanigans that are muddying that picture, from stuffing debt into SPVs to changing depreciation schedules. Not to mention that they’re reporting changes in investments into other AI companies like OpenAI and Anthropic that are paper gains until they go public.

    I would argue the only company actually making money on the AI boom is nVidia, and while everyone trots out the line about selling shovels during the gold rush, they don’t really account for the fact that nVidia’s stock value is based on their ability to sell greater and greater amounts of shovels in the future. Once the boom ends their inflated value will plummet. I suspect nVidia will have the same trajectory as Cisco after the dot com bubble burst - the company was still around and selling lots of network equipment, but their stock price only recently (in the past year) hit the same level that it did at its 2000 peak.












  • If I had an employee that fucked up as often and confidently as LLMs do, I would terminate their employment. I don’t have the time or patience to redo all their work because I can’t trust them. If a brand had quality issues at as high a rate as LLM output, I would avoid buying anything related to that brand.

    A human being or company can make some amount of mistakes and that’s completely understandable, and I think it’s perfectly fair to hold LLMs to that standard. That said, with a human being I can stress how important some task is and they don’t double or triple check their when it is that important, I can not and will not trust them. If I tell a company “hey this is important to me”, they will usually make extra effort to make sure it’s correct, and if they fail to do so, a decent company will make an extra effort to make things right. An LLM will offer an empty apology and spew more bullshit at me that I still have to verify.

    The fact that AI gets things right often enough for you is great. Maybe I have higher standards, maybe I need higher quality output, or maybe I have trust issues. Everyone has a different threshold. Just because yours is much different than the people here doesn’t mean their opinions are invalid, and the way you dismiss everyone’s opinions and insist that it’s good enough makes you seem just as insufferable and ignorant as you accuse others of being.

    The fact that LLM output has improved is great, but it is still so far from meeting much less exceeding my threshold that it’s beyond frustrating to keep being told over and over that “no bro, it’s good now”. I’ve been hearing that for over two years and it’s still not close to where I would trust it to answer questions, much less perform actions on my computer with access to my personal data.

    You need to learn that some people are different than you and understand where they’re coming from if you want to have an “interesting discussion… about this topic”, rather than getting frustrated that people don’t agree with you and rage quitting.