1,749 mentions · 790 chunks · 131 episodes
...and ones. Mechanistic systems could only handle black and white. Embeddings and LLMs are now fuzzy, squishy, grayscale. We can bring back the nuance! This moment is the time for us to reclaim our nuance and humanity. Before we were st...
...nder: sycosocial relationships are faux friendships with infinitely sycophantic LLMs. It is an indictment of the system that lulled the human into that relationship in the first place. That's what makes those kinds of relationships "s...
LLMs are good at things normal programs are bad at, and bad at things normal programs are good at. Chat is great for things that computers used to struggl...
AX will matter more than DX. If most code is written by LLMs, then Developer Experience at that layer matters less. LLMs are infinitely patient, and prefer things that are like other common things. Humans will ...
Specs are editable documents, so they are better for interacting with LLMs. A powerful pattern is editable inputs to LLMs Context that you intentionally factor out to be shareable in other tasks, vs implied context of "messa...
... of itself. Only the most highly motivated organized would bother. But now with LLMs it can provide extremely valuable context. You accumulate the information in one place for your own use… but also help the LLM-powered tools to under...
The better LLMs get, the less context they need to feel truly personal. They can read between the lines, expand beyond what you said. But the quality of that context...
I'm optimistic about LLMs potential for humanity, and pessimistic about the slippery slope ChatGPT is on. Not where ChatGPT is, but the drain it will circle. Every hyperscale ...
LLMs aren't just a steamroller that we have to freak out about and fear. LLMs are the most humanistic tech ever. What if we used it to be more human? We h...
...domain that you're an expert in. They got to set how technology worked. But now LLMs democratize the power to create tech. But now the experts can have that be unlocked without having to beg some random one-ply thinking technologist. ...
...not enough about task abstraction. That is, to lever the task to the point that Llms can reason about it
... not the busted ones. Use cases that are transformative and not possible before LLMs have more return than use cases that were possible before LLMs but could just be a little more efficient with LLMs. The assumptions of the systems ar...
...e is now cheaper so you can have languages that give formal guarantees but that LLMs can be patient enough to use.
...ative curiosity.[et] This distinction also gets at why Bret Victor doesn't like LLMs: it moves people to an outsourced thinking position.
Socrates critiqued writing because you couldn't talk with a book. But now, with LLMs, you can. It only took us a couple thousand years.
The disruptiveness of "work to rule" strikes implies that automation by LLMs could have a ceiling of effectiveness in complex domains.
LLMs are a mirror. Of society in general (the weights; the background awareness). Of the user specifically (the context; the questions the user brings to ...
...socratic dialogue from Geoffrey Litt[fq] about why chat is not the final UI for LLMs.
...downsides fundamental to chat itself? Those will help discover the ideal UX for LLMs.
If chat isn't a good UI for LLMs, why is it winning? To me it's entirely based on the novelty of LLMs, everyone's just experimenting with them right now and starting open-ended tasks...