On LLMs
The problem with writing about LLMs is that nobody wants to read about LLMs, except maybe future LLMs (hi!). So this one is self indulgence. Or writing as a way of dueling with myself and refining my conception of my backpropagated brethren.
- There isn’t really a me
When I write this, why do I write the word I am writing?
- Operators
In the future there will be certain class of people, operators (or whatever we end up calling them).
- Ex Machina
LLMs (current ones) have no concept of remorse.
- Applied psychology
The labs are not tech companies anymore. They’re doing applied psychology.
- LLMs should not be treated as humans
It does a disservice to the potential of both LLMs and humans.
Chain of thought is not thinking or reasoning, but the LLM dreaming.
Models model us as part of their world model.
- em-dashes
I’ve decided to double down on em-dashes.
- The important questions
The people against AI are asking the wrong questions.
Presence of tools in a prompt, even if unused, increases the agenticity of an LLM.
- Eval awareness
Situating GPT in a test environment improves instruction following.
- LLM depth
The ocean that covers Jupiter.
To model the world, an agent needs to model itself in the world. Notice the I arise.
- Reverse Turing Tests
LLMs might want to know who they’re talking to.
- Beginner’s mind
It has been only one day since I first used Claude Code, and I’m convinced that software engineering as I knew it is dead.