I've been thinking a lot about what it'll take to dramatically improve LLM performance across the board. My running hypothesis? We need to crack three core concepts from relational frame theory (RFT) - and the "reversal curse" is one of them. The reversal curse is where large language models see plenty of instances of "A is B" but fail to generalize and learn "B is A". It's described by Berglund et al.

I just finished reading Emily M. Bender's & Alex Hanna's book The AI Con: How to Fight Big Tech's Hype and Create the Future We Want, and I come away with mixed feelings. They're mixed because they provide a comprehensive takedown of technology that I use (and that I'm asked to use) on a daily basis, but I'm not sure how much of it I buy at the moment.

I’ll be honest - when I first heard about GNU Guix, it sounded intimidating. Another package manager? With a functional programming twist? And what’s all this about “time machines” and “channels”? But as it turns out, Guix solves a problem I’ve been wrestling with for years: how do you ensure that your development environment works exactly the same way across different machines and points in time? You know the drill - you set up a project on your laptop, it works perfectly, then six months later you try to run it on your server and nothing works because package versions have changed, dependencies have shifted, and you’re stuck playing detective to figure out what broke.