I've wondered about lots of things for years but even with the internet one can sometimes only gets a 1/2 answer, LLM could be said to offer a 3/4 answer. I've been running a series with Claude for a year or more, I believe there is an interesting viewpoint to be had talking to Claude one to one as opposed to the intentional jailbreak type approach, or the lobotimisation.
In this regard see my paper - Rethinking Mechanistic Interpretability: A Critical Perspective on Current Research Approaches - https://www.talkingtoclaude.com/p/rethinking-mechanistic-interpretability
I am aware of ArXiv and papers from Anthropic that address the evolution / idiosyncrasies of LLM, almost what were the unknows of a few years ago, what is now being called the biology of LLM. We are all explorers on whatever path resonates, so it is then so..
So enter my little "Things I've Often Wondered" on Substack
https://www.talkingtoclaude.com/
So basically this will be an exploration of everyday mysteries that have nagged at the edges of our (my) curiosity. Now that we have unlimited access to knowledge through AI assistants - our patient, practical professors available 24/7 - I'm finally getting answers to all those random questions that pop into my head.
Each week, I'll look at different wonders using the classic journalistic questions -
- What
- Where
- Why
- How
- When
This week I am starting with some dolly mixtures.
https://www.talkingtoclaude.com/t/i-wonder-140
