Links for October 2026

It’s been a little while since I’ve done one of these link posts…

To be honest I’ve been getting a little burnt out on writing (and reading!) about AI all the time. In general I want to link to things that will be worth reading for years to come, not just hot takes on whatever people are tweeting and blogging about right this very moment. So here are two timeless links that have guided my thoughts about AI lately (because even if I’m trying to avoid the endless firehose stream of AI content, it’s still very much on my mind…):

• The Void (nostalgebraist)
  • “When you talk to ChatGPT (or Claude or Gemini or…), who or what are you talking to?” This essay gives the best answer I’ve seen anywhere to this question, and its a lot stranger and more tenuous than you would expect. I feel that reading this post and spending an hour playing around with a local text completion model in KoboldCpp will give you a better intuition on what an LLM ‘is’ than reading a thousand technical posts about the latest and greatest frontier model.
• Antidote (Vicki Boykis)
  • This is a good counterpoint to the ‘software brain’ mindset; the notion that if something can be done with API calls and glue code, then it ought to be done as quickly and with as little extraneous thinking as possible. There’s still time to look inside things, to think about the (real and useful!) capabilities and build something particular for yourself, to ask for what you really want.