Some interesting stories this week:
- New Insights for Scaling Laws in Autonomous Driving (from the Waymo Team). TL;DR Tesla's strategy of turning all its cars into roving data capture machines was probably the right call. Not that Waymo isn't saying that Lidar is useless (Waymo uses expensive rigs whereas Tesla works with vision only), like Elon Musk is. In a sense, they are saying the opposite; most likely, data from multiple sensor types is still better as long as you can get a lot of it.
- LLM Visibility Tools: Do SEOs Agree On How To Use Them? The new era of the AI web will dramatically change how web pages are found and consumed. However, it's not clear that there will ever be such a thing as AI SEO (Search Engine Optimization). There are really two reasons for this: 1) getting included in AI results can happen in two ways: via training in the AI build process or in web searches that the AI itself does. The former is extremely hard to predict because so much processing of the data goes on; the latter is easier, but it's unlikely AI search engines will look very much like human search engines (https://exa.ai/ is one of the first AI native search tools). 2) The second problem is that users aren't anywhere as likely to jump out of an AI conversation to go to a webpage as they are from a Google search (where clicking a link is the only choice). It's likely "SEO" may turn into something that looks more like affiliate marketing, where AI services draw links from specialist indexing services and charge for leads for certain queries.
- Meta Launches Superintelligence Labs in Major AI Restructuring. Meta is pushing hard to expand its AI team. The question is, why such a huge push? Likely, a big reason is that the daily online user experience is being redefined. The more minutes ChatGPT wins, the less we'll spend on social media. At the very least, Meta would like that AI usage to be on its own AI and not a rival service.
- Lovable on track to raise $150M at $2B valuation. Rocketing to a $2B valuation in 6-9 months? Even for AI, that's an incredible jump. Having used Lovable, though, I can think there is likely a good case for this. Lovable and a number of competitors enable web app creation purely via AI prompts. The experience is near magical and is likely to redefine the whole category of web applications. There is competition, no doubt, but as these apps go deeper and get better, they are complex enough that there will likely only be a few winners. Lovable seems like it could be one of them.
- Pay up or stop scraping: Cloudflare program charges bots for each crawl. This news has been a little under the radar this week, but it's actually quite important. Cloudflare is a Content Delivery Network that many websites use to "sit in front" of their content in order to improve performance and protect against security risks. The company covers vast numbers of websites (20% of the web according to Ars Tecnica). This week, they began to give sites the ability to charge for access to AI scraper bots looking for training data. These AI bots are now blocked by default. This suddenly gives anyone who wants the tools to block AI scraping. AI companies will need to consider paying. It puts power back in the hands of website owners. On the negative side, though, anyone who has already trained an LLM on public data now has a bigger advantage.
Wishing you a great weekend!
