no that’s just my name :)

article’s a year old. can be ignored.

the future will be competing visual metaphors for coding agent labor camps

liking this bc I’m delusional enough to think I’m one of the special software girlies who will thrive in the new world

felt weird to like this one, Amber

the long shadow of ChatGPT: every LLM output is determined by the entirety of the context window, not just the last thing we asked it. but chatbot interfaces don't highlight that

github.com/JD-P/miniloom
this is the latest implementation I know of but it's not quite what I had in mind; being able to prompt the same model, different times, and have all the responses displayed in parallel was what I was thinking of

adversarial dating chatbots requiring more sophisticated countermeasures to avoid bad actors. zero knowledge proofs to see if you have the same hobbies

id like to see more Loom-like interfaces, where we can parallelize our prompting over the same starting places and see where the trajectories go. that, to me, more accurately captures the idea that this piece is reaching for (which i generally agree with)

one thing i felt this was missing was the role of the context window. the LLM isn’t working in this pure space of idea projection, where you have a cube that you rotate. as long as inquiry proceeds in the same window, everything is influenced by everything before it

Turns out when you direct the universal function approximator to predict the encoded outputs of human minds at the scale of 'the entire human corpus' the resulting network has to get mind-shaped internally to succeed well at the task

yee

im out of touch w the masses bc the notion of “explaining tool use” never even occurred to me. duh they can use tools now,

i told a friend about chain of thought prompting and she got mad at me. “why would that work. how.” girl,

technical writing is the programming language by which you direct an LLM!!

i understand