| >> But in the real world, the Web is a pile of auto-generated and auto-assembled fragments of slop > There are parts of the web like that but your assertion seems to rely on this being universally true. It clearly and obviously isn't. In a sense, the whole web is like that and has been for a long time. Which is not surprising, 99% of everything is shit. We've just had tools that have been astoundingly successful at separating out the wheat from the chaff. With the advent of AI, there are two significant differences: (1) the scale is shit is vastly greater, we're at something like 99.999% shit and adding 9s steadily; and (2) every way of distinguishing shit from gold is being steadily overcome by AI advances. >> Also in the real world, people are decisively choosing the AI-generated summaries and fevered imaginings. > Are they "decisively" choosing it if it's turned on by default? If it were actually opt-in then we could measure this. As it is I don't think you have any data to rely on when making this assertion. I'm not referring to things embedded in browsers. I agree with your counterargument there. I'm talking about people using chat interfaces for search, and search engines presenting AI-generated results. There is hard data that shows users moving away from doing traditional searches and clicking on individual pages. From a quick (traditional!) search: https://www.statista.com/statistics/1454204/united-states-ge... https://www.gartner.com/en/newsroom/press-releases/2024-02-1... (Gartner representing a pre-AI source of made-up bullshit with questionable origins). >> Not for everything, but web search -> URL -> page visit is becoming a declining percentage > The same web search companies that own AI models they're trying to sell? Do you not suspect there could be a few confounding variables in this analysis? Undoubtedly. But what are you implying -- that you could start a no-AI search engine and be wildly successful today because people will use yours over the other engines that have AI summaries? That's kind of what you're implying for the browser (and one of the most common uses of a browser is a search engine). >> except that automatic translation is really nice > Which we already had and has nothing to do with language models masquerading as "AI". Er, the language models are wiping the floor with more traditional semantic machine-learning based translation. It's kind of sad, but also cool that it works so well. Oh. I guess you're making a distinction between LLMs and the not-so-large language models used for translation. I don't see much of a difference. They're transformer-based, they're intentionally stupid but work because they're fed huge amounts of data, etc. >> is fine and dandy but ultimately pointless unless you have an alternative that doesn't require the entire world to cooperate in turning back the clock. > An alternative to what? Tab renaming? Bad article summaries? Weak search engine algorithms? The article summaries may be bad, but they are very popular and widely used. The search uses are not bad or weak, the only criticism I see is that it doesn't need to be kicked off directly from the browser. But fortunes are made on eliminating a single click. I'm not claiming that having AI in the browser is super awesome. I personally haven't seen any killer use cases yet. I'm only saying that keeping AI 100% out of the browser is going to be a losing proposition. Not because it's so amazingly useful to have it, but because it's at least a little more useful and that's enough to choose one browser over another. People are demonstrating that they're going to use AI whether you think it makes sense or not. |