Sunday Links: AI Books, Autopilot, and AI Talent

Very late this week due to travel this week - apologies. However, there are still plenty of exciting AI things happening in the world. Here are the things that most piqued my interest:

• Spines and the rise of AI book publishers. The newly funded startup spines has been roundly condemned by the book and literary industry for their plan to use AI to edit, proofread, and prepare books for publishing. The startup aims to empower authors by streamlining the process of taking a work from manuscript to finished book. The industry pushback centers on the idea that this "cheapens" the process and most authors will not achieve their goals. Clearly, the AI-fication of these processes can be seen as an accelerator for vanity publishing. This simply seems an evolution of the kind of thing that platforms like Blurb already do. How long until Blub and others have AI buttons of their own? AI seems inevitable here. It's likely true that net more junk content will be produced as books, but every step of reducing costs does open the door to a few more gems being created.
• AI eats the world. Benedict Evans' annual trend presentation this year unsurprisingly focuses on AI and has an even more unsurprising title. Perhaps nothing particularly new if you've been following this blog, but it's useful to have some numbers and projections. I do like one of the main points Benedict makes, which is "As automation becomes more ubiquitous, it disappears" (Slides 68 and 69). That will be true for many of the AI "features" we see today.
• Alibaba QwQ impresses at GPT o-1 model levels. The iterative reasoning in OpenAI's o1 strawberry models has been the cutting edge of GPT reasoning since its release a few months ago. However, once a principle has been proven, it becomes easier for others to apply the same logic. Now Alibaba has showcased its own "Large Reasoning Model" (LRM), and it shows strong performance. This is another piece of evidence that we're likely to see multiple cutting-edge models for a long time rather than a clear leader.
• New Zealand navy ship sank off Samoa because autopilot was left on, inquiry finds. Not a modern "AI Story" as such, but a lesson nonetheless: Automation is amazing unless you forget to turn it off or add in automatic cut-offs. As the crew of the ship frantically tried to correct their course and understand why controls were unresponsive, the activated automated was the real reason that they had no control. Why this was not obvious remains unclear. A disaster like this highlights that semi-automated systems can be just as dangerous as fully automated ones if the humans involved don't fully understand what is going on.
• Google, OpenAI Snap Up Startup Founders; AI War Games [Paywalled]. For AI startups, reaching escape velocity is hard. This story in the information covers several founders of high-profile AI startups leaving and going to join large players such as OpenAI and Google. It is hard to make any startup work even with 10s Millions of dollars in funding, but in some spaces, there are already many competing (funded) players. No doubt these moves show how critical talent is for large players if they are attempting to hire from the same pool of founders that just left them 12-24 months ago.

Wishing you a wonderful week.