I can’t stop thinking about something Rohit Krishnan said in a recent post about how we interact with LLMs. Just put together some thoughts and launched something new to explore one idea.
Check out the post below and browse around the Latent Library for one example of a different interface we might be able to have for interacting with LLMs at latentlibrary.xyzhttps://worksonmymachine.ai/p/solving-amazons-infinite-shelf-space
It shouldn't come as a huge surprise, but there's a lot to learn from linguistics about working effectively with LLMs. They are large LANGUAGE models after all.
In this post I delve into this association along with a small announcement about the 0.2.0 release of VSM - the AI agent framework we're building that explores these ideas through code.
https://worksonmymachine.substack.com/p/how-do-you-speak-pidgin-to-a-probability
I got some questions about vsm and airb last week and wrote up a little response to them about why I'm thinking the way I'm thinking, what this point in time means, and what comes after.
You can kind of look at the history of software as an eternal struggle between the Dijkstras and Kays, with each side trading off leadership back and forth over time.
Check out the latest post:
https://worksonmymachine.ai/p/safe-is-what-we-call-things-later
The System Inside the System
Announcing two new AI Ruby Gems: vsm and airb
At Sublayer, we've been exploring different ideas for how to build agents for a long time and I recently caught the cybernetics bug after reading Dan Davies' "The Unaccountability Machine".
Stafford Beer's Viable System Model jumped out at me as a structure that would be really interesting to explore as a framework for building agents, so I put some together and wrote up some more thoughts about it.
https://worksonmymachine.ai/p/the-system-inside-the-system
When working with AI agents, there's a lot of waiting around. I've found myself starting and working on a bunch of different projects in parallel lately.
Wrote up some thoughts about it and what we're building in to APM to help support it
https://worksonmymachine.ai/p/the-parallel-lives-of-an-ai-engineer
Refactoring is an important tool in your toolbox for effectively managing your AI agent's context window, but it feels a lot different than it did in the past (a year ago). Just wrote up some about what refactoring means now
https://worksonmymachine.substack.com/p/refactoring-in-the-age-of-unlimited
