Here are this week's AI Links. More strategic plays from ChatGPT come into view, and how to get a nice chunk of shares in OpenAI:
- The Death of Search (Paywall after a couple of articles). Matteo Wong, in the Atlantic, has written a couple of articles on the changes AI could bring to search, and in the latest, he talks about using the current generation of AI search to do article research. He focused on accuracy (rather than the occasional lack of it) and that AI search could lead to companies aiming to keep users "in their ecosystem" and not sending them out to the open web. I agree with him that AI will fundamentally change the way we use the Web and is potentially a very big threat to the open Web; I think he gets causality fundamentally the wrong way around. Current search incumbents are extremely happy with the way the Web: millions of people publish content, they help discovery and monetize via ads. It's not Google trying to make a change; it is user behavior and convenience forcing a change. It is entirely unclear what will happen to the "millions of people creating content" part of the web under this change. We should definitely worry about it, but we can't wish user preference away. Instead, we should figure out how to empower more content and commerce. One way would be to have more different AI search entry points, and another would be to build more open search indexes that many people can use but flow payments back to original sources.
- Also... OpenAI just launched ChatGPT Search and probably just bought the "chat.com" domain name. You can use Dharmesh's tweet to try to figure out the sale price. What's key here is that OpenAI clearly wants to keep playing to be one of the search and information entry points that people use most. Google has everything to lose in that race; OpenAI and Meta are the fastest pursuers. Apply is wondering how to use the device to stay in control. Large destination sites like Amazon, Alibaba, and many others are probably wondering how this will all play out for them.
- Artificial Intelligence, Scientific Discovery, and Product Innovation. Via Ethan Mollick, this MIT paper is one of the clearest studies yet showing how AI support can boost innovation in science. What's more, it also suggests the gains are evenly distributed. From the paper: "Investigating the mechanisms behind these results, I show that AI automates 57% of “idea-generation” tasks, reallocating researchers to the new task of evaluating model-produced candidate materials. Top scientists leverage their domain knowledge to prioritize promising AI suggestions, while others waste significant resources testing false positives.". That makes a lot of sense to me: AI expands the search space, but experience is needed to triage the options. It also makes me wonder if gains will be capped: how do you have AI explore every new part of search space? The answer is probably, again, that expert researchers need to guide the ideation process using their intuition.
- First artwork painted by humanoid robot to sell at auction fetches $1m. Ai-Da's portrait of Alan Turing is one of several painted by the AI powered robot. There has been a lot of AI art, but far fewer that use an AI physical process to create the art. This doesn't really seem like a landmark technically - all of this has been possible for a while. It does seem like a landmark culturally, though, or at least a smart investment. By paying over $1M, the buyer effectively guaranteed the status of those particular paintings as a piece with historical value. No doubt galleries will desire it; maybe it'll have soup thrown at it. I'd be willing to bet that in 10-20 years, its value could have soared due to fame, not necessarily brilliance. Also, think of the T-shirt sales.
- Chinese researchers build military AI using Meta’s open-source Llama model. On the flip side of open, militaries around the world are working hard to determine how to best use AI in military operations. They will undoubtedly use open-source when and where they can. One can argue that open-source should not be shared for that reason, but I think that's relatively futile. I doubt this news reflects anything close to the full sophistication of what is being built. Knowledge is already out there, even without individual models being shared. We are better off focusing on how we can enable humans all over the globe to be more prosperous, productive, and happy with AI so that we reduce the desire to go to war in the first place.
It's hard to end this letter without commenting on the result of the US Election. It's hard to see America opt for a President (and perhaps more so a party) with such a toxic view on personal civil liberty. It feels like a huge backward step. It's also hard to see, though, that the choice was (in part) economic options(*) vs. cultural progress. There was no reason to split the options in that way. All we can hope is that institutions survive intact and that discourse moves back to a more sensible center.
Wishing everyone a happy Sunday.
(*) Not that I think Trump/GOP economics are better, in fact they could end up far worse. However, there may be an opportunity for improvement, and Democrat policies have been ignoring a lot of economic pain.
