Notes
These short posts include "asides" from the my original blogs, a sample of "tweets" from Twitter, and more recently are often syndicated to my Fediverse profile.
SpaceX is doing amazing work, and is full of engineers I really want to celebrate.
I would find the usual regulatory capture antics gross, but that’s not why I cannot enjoy their achievement. It’s that every success of these brilliant people is directly empowering their actively-fascist oligarch. And I despise him all the more for this.
I have been trying models from 3 providers on 2 very different codebases and using 4 different agent/IDE toolchains. For any given task, all of these variables figure into which setup works best.
Anyone who asserts “this is the best model for coding” is oversimplifying, grifting, or has found an optimal configuration that eludes me.
I so tired of exaggerated ink traps.
There is a growing body of research into LLM self-awareness but I am particularly fascinated by the “evaluation awareness” research by Anthropic into the Hawthorne effect in LLMs. (h/t to @futureshape.bsky.social for the connection)
Today in “things my coding model said”:
This is a great example of premature optimization - the debouncing added complexity to solve a performance problem that doesn’t actually exist at the scale we’re operating.
(It wrote the debouncing, until I asked it to simplify.)
As I enjoy watching models explain their reasoning through work I’ve given them, I keep thinking about the recent paper _Chain-of-Thought Is Not Explainability and how much we should remind people that rationalization is not always insight:
verbalised chains are frequently unfaithful, diverging from the true hidden computations that drive a model’s predictions, and giving an incorrect picture of how models arrive at conclusions