Sign in Subscribe

Sunday Links: Xaira, Data Centre Geopolitics and Search

Steven Willmott

• 3 min read
Sunday Links: Xaira, Data Centre Geopolitics and Search

Even later this week, but here are the most interesting links I came across this week:

  • Xaira, an AI drug discovery startup, launches with a massive $1B. The AI+Science trend continues with another massive investment. The company raised initial funds from well-known Silicon Valley funds, then operated in stealth, and is now launching publically with another large funding round. The size of the fundraising is somewhat understandable, given that getting drugs to market takes time and money, and so will the GPU power needed to find them. With investments like this, existing healthcare companies will probably have to make even more phone calls to NVIDIA.
  • Microsoft doubles down on Malaysia. With $2.2B of new investments, Microsoft is expanding data center operations in Malaysia as one of its Asia hubs. The company is also investing heavily in skills training. The geopolitical dimension of AI (and tech more generally) is fascinating now. Are data centers the new geo-political chess pieces rather than airfields and ports?
  • The Future of Search is Open...? Michael Spencer has an extensive piece on the signs that OpenAI's planned ChatGPT-based search engine will, in part, disrupt Google. (Some parts of the activity are real, at least.). I think Spencer's discussion is interesting and detailed, but I'd quibble with the headline - OpenAI entering the search market won't make it open. OpenAI isn't likely to open its search engine more than Google. You also have Meta entering the search space with Meta.AI (not available in all countries). For search to be open you would need an open search index that many could use - this infrastructure would be very expensive and takes huge know-how to maintain. It would need to be a large crowd-sourced effort, which isn't really on the horizon. For now, we'll likely see more companies trying to build their own indexes (like Perplexity is). What the OpenAI search effort (and Meta's, for that matter) does, I think, say, is: 1) OpenAI and Meta are very confident they can build an AI-based search product that will get good enough to compete with pure index-based search, 2) OpenAI (and again Meta) are betting on the consumer side of AI adoption. In other words, they are both competing for the "personal assistant" position in consumer AI usage. That's a huge prize and one that could ultimately eclipse pure search. Google and Apple clearly have horses in that race as well, but they are less cross-platform than OpenAI and Meta. Hopefully, the future of search will be more open, but that's not (yet) what this news signals.
  • Perplexity raises another $63M. Also, in this category, still my favorite search tool, Perplexity, raises another round of funding just a few months after the last one (probably with more to come). Perplexity's new product is enterprise-focused search/research, so it's unclear if it's playing for the same audience as Google, Meta, and OpenAI in search. That's probably smart. My feature request is "shared" search spaces where multiple people can research a given topic with the AI in the room.
  • ChatGPT gets memory. This feature has been available to select users for a while, but it's now rolled out to all Chat GPT plus users unless you are in Europe of Korea. (You need the Plus subscription to get access.) In essence, it allows you to ask ChatGPT to remember certain things about past conversations and use them to help guide it for future queries. There's clear value in this, and it's a step forward for a personal assistant. Our life context (name, location, the job we do, things we wrote in the past, etc.) all affect the kinds of things we're looking for. It might need a lot of work to stop "Memory" from becoming the gateway to creating a mishmash of half-baked outputs, though. As has been pointed out in reviews, it's not always clear when it will draw on a memory or not. How much of a result is truth? How much hallucination and how much from my own misguided prior opinions?

Since this week has mostly be on long-term tech speculation, here's a final harder science link. One of the limitations of the recently released Llama3 was the context window being smaller than some of its competitors. Thank you research and open source – now that's boostable from 8k tokens to 80k.