Skip to text

Milan Andreew

Contents

Reading notes

Scientific American, 1959

It's inconsistent, but they're trying a lot of stuff, and a reasonable amount of it lands

Poem

Richard Khan

In Massachusetts did RMS
a stateful automaton decree

Collection

LLM Exercise Submissions

Notebook

Underneath the foundations

There are tons of different systems that can serve as a foundation to formalize everything else.

Now

I'm still mostly doing nothing, just reading.

Reading

How to Think About AIRichard Susskind
Knowledge GraphsAidan Hogan et al.

Thinking & trying

An interactive mathematics laboratory

I've been thinking about what one would look like, following Farmer's proposal. I'd like to make one for type theory and formal logic, or for category theory or cryptography, mainly because it would help me learn mathematical domains that I find interesting.

AI-generated prose and design

One of the biggest weaknesses of current SOTA open-weight models is that the quality of their prose and design is abysmal. I'm unsure how one could remedy this, but I'm certain that there have to be ways. My first thought is to give the model examples of high-quality prose or high-quality UI/UX, but that does not seem like a very good strategy.

Jev

I've been playing around with Jev, TypeSafe's model for fast typed decisions, to see how it can be useful in agent harnesses.

AI-targeted SEO

I'm seeing it become more and more common, and am considering solutions to systematically defend against its effects.

A Poem

Napped half the day;
no one
punished me!
- Kobayashi Issa