Scientific research on artificial intelligence published by Big Tech has to be read the right way. In particular, you need to understand the corporate message hiding behind it. I talked about this in a podcast that will be out soon, “Il libro open source”, a beautiful project, and I got the same confirmation reading “Empire of AI”, a book I’ll give a proper review to - wonderfully politically incorrect, by the way.
Watch out when Anthropic, Apple, OpenAI, Microsoft and Google publish research. Because beyond the paper itself - beyond the premises, the hypothesis, the methodology, the experiments, the conclusions they draw - behind these articles there’s typically a corporate message. There’s always a moral, oriented toward communicating something about what they sell, what they’re about to sell, or, as in Apple’s latest paper, about what they don’t sell.
Apple, or how to lose gracefully
Small example. I’m Apple, I fell behind on artificial intelligence, so I publish research engineered specifically to highlight the limits of these systems, and I tell the whole world: “I’m Apple. Look, language models have their limits, AI isn’t reliable. The reason we’re not competitive, the reason we’re not investing in this, isn’t that we failed to bet on it and we’re late. No. The reason is that we’re protecting you from the dangers of unreliable AI.”
Anthropic, or how to win by muzzling yourself
And then there’s the Anthropic message, with the AI that threatens to tell the programmer’s wife he’s cheating on her. All this stuff about alignment and safety is Anthropic’s way of saying: our AIs are so powerful that we have to keep them gagged. Look at what they can do. But don’t worry, we’re on top of it and we’re protecting you. Use artificial intelligence, because yes, there are risks, but we’re on it, and it’s so powerful it’ll solve a ton of problems for you.
The thing is, that exact same research could be done in far, far more measured ways. There’s no need to drag researchers’ wives and infidelities into the prompts. These are pathetic strategies, every bit as pathetic as Apple’s.
And I’ll stop here for today, because I’m getting old - though it’s hard to keep me quiet.
