Sometimes I beat myself up for not reading enough books, but I added 189 things to my reference manager last year. I think that counts as enough reading.

Just a few days after his 60th birthday, Dr. Yoshua Bengio gave a lecture on AGI Safety at the University of Tokyo, widely appealing for participation in this field. Today has become a memorable day that marks the kickstart of AI alignment research in Japan.

This is the conference I've been organizing for the last 8 months! We have some really cool speakers lined up covering a bunch of promising research directions in technical safety. Sign up to attend on our website! tais2024.cc

Very excited to have Prof Aoyagi at TAIS 2024. She coauthored with Sumio Watanabe who first researched singular learning theory, and her research makes meaningful progress towards getting analytic real log canonical thresholds for deep ReLU networks which would be huge

Did I organise a whole conference just to have an excuse to talk to the causal incentives group in person about their research? I'll never tell.

Manuel's paper "The Emperor's New Markov Blankets" did the rounds on lesswrong a couple of years ago, I'm hype to hear what he's been working on since. cambridge.org/core/journals/…

I've met Rob Miles a couple of times and he's as charismatic in person as he is online. Looking forward to hearing him speak!

Last weekend, the Technical AI Safety conference (#TAIS2024) took place in Tokyo. We are proud to have organised this landmark event as part of our commitment to advance the complex discourse on AI alignment. Thanks to all speakers, poster presenters, partners & attendees!

This is a neat paper! It's cool to see LLMs applied in a discrete domain where it's easy to check their work.