Post #2469473

Let me tell you the story of a company, which I will refer to as Company X (I am not talking about Twitter). Company X makes their pitch to other businesses. Their promise is that if your business pays exorbitant subscription fees to Company X, and integrates Company X's software into every part of your business, then it will magically generate profits for you. Company X's software harms every regular human who interacts with it, making the products of Company X's customers worse. Company X's software also doesn't work. Company X's customers do not see a single dollar in additional revenue or profits as a result of using it. They may even lose sales as their own customers try to avoid Company X's software. Every business who pays Company X is, objectively, getting scammed. There are other companies who sell scams similar to Company X, but Company X has marketed themselves as the "premium" product in their category. They charge more, but they promise more magical profit generation than their competitors. This reputation for being the "best" isn't based on any facts: Company X's product does not work. It has never worked and will never work, but through marketing hype they managed to convince CEOs of other companies to pay them for it. Shills for Company X insist that their product is "here to stay." You'll have to live in a cave and never use a computer to avoid Company X's product. By now, you may have realized that Company X is Denuvo. Oh what's that? You thought I was talking about AI? Nope! This was all a discussion of Digital Restrictions Malware! Every game publisher who pays Denuvo is, objectively, getting scammed. They all hope that Denuvo will magically generate additional sales, despite ample evidence it won't. Some gamers will avoid buying games with Denuvo in it, and many others won't care. But no gamer is thinking "I wasn't planning to buy this game, but now that I know it has a rootkit in it I'll happily pay for it!" Denuvo has stayed in business for 12 years because CEOs are the most gullible people in the world. Denuvo can tell publisher CEOs "Hey, we have a product which costs a fortune and gives you absolutely nothing. Buy it and demand for your project will magically increase." And CEOs fall for it every time. DRM has been around for decades before Denuvo. It doesn't work, has never worked, and will never worked, but it doesn't appear to be slowing down. The biggest threat to Denuvo right now doesn't appear to be the "DRM bubble" popping: it is operating system vendors building even more invasive DRM directly into their operating systems. Now, let's turn our attention to AI. Much like DRM, LLMs promise corporations that they are magic generators of infinite profit. Whereas DRM promise to magically increase sales, LLMs promise to magically eliminate the cost of wages. Much like DRM, LLMs do not work, have never worked, and likely will never worked. But much like DRM, the not-working doesn't matter, because the target audience, CEOs, are the most gullible people on the planet. Now, optimists like to point out one key point of distinction between the two scams: the big AI companies are currently losing money due to high operating costs. Surely, once the AI bubble pops, there won't be a way to fund AI anymore, right? But the high operating costs are not actually needed for the scam. Companies paying for AI aren't doing it because they care about the details of transformer architecture neural networks. They are doing it because they heard it's a magic machine that means they don't have to pay wages. Even after the AI bubble pops, the AI scam can stick around. All that is required is for the most gullible people to believe someone who tells them "there is a magic machine that can save you the cost of your entire workforce." Sure, a 200 billion parameter LLM isn't financially sustainable, but a 20 parameter LLM could train and run on an embedded device. "But a 20-parameter language model wouldn't be 'large!'" It doesn't matter, because the entire thing is a scam to begin with. For the most recent crop of big AI products, the companies aren't even revealing how many parameters their "large" language models have. If GPT-7 is just a wrapper around gunzip used as an autocomplete, no CEO will even notice the difference. And that is why I think AI really is here to stay. Not because it works (it doesn't). Not because it has some legitimate uses (it doesn't). And certainly not because it's not a bubble (it very much is a bubble). It will stick around because, at it's core, it is a scam, and there will always be people willing to fall for a scam.