> They clearly aren't writing for others to read their work.
I don't know, the prose and material is fine in my opinion. I do find the uncapitalized style a bit weird, but it's not something that stops me from reading it.
749 karma · joined January 5, 2026
I don't know, the prose and material is fine in my opinion. I do find the uncapitalized style a bit weird, but it's not something that stops me from reading it.
Especially as people grow more dependent on agents doing all of their computer work.
Coming from a non-EE background, this is the most interesting part to me. 10ms is an eternity in programming perf work, but it's already very constrained in how clear the signal is for reading the transmission. I just find it interesting that we want to do nothing to create a signal. Perhaps this makes it harder to exploit?
I've done similar work simply by giving a model a database of text, and asking it to do a unit of work for every passage. "Checking for spam" would be the model checking surrounding messages and user's history, likely an opinionated verdict, but would work. Difficulty comes in when those units of work become complex or have to work in tandem, which isn't the case here.
Endgame of AI + human gameplay is delegating everything to the AI. I'm not sure how you would have to design a game where you only delegate parts of macro/micro, and still make the game interesting.
I could see that square packing is more simple, but by shifting elements per row I don't see why it couldn't propagate at odd angles.
They don't really show it here, but supposedly a Cerebras' chip is sub-divided into a 2D mesh of what they call PEs (cores). So data flow generally flows in two axis, and maintains consistent timings to its 4 neighboring cores. They show the layers flowing from left-to-right being streamed by PEs, but I wonder how true that ends up being: maybe the communications are easier to stream via diagonals, or some other shape that has better topology for data flow.
Then, even from die shots of CPUs and GPUs we often see that cores are tiled as squares, and those don't have the same cross-core connection and streaming requirement. Now, I assume that some parts of what we call a core can be moved around more freely while maintaining whatever power/heating/timing requirements we have, so perhaps it's easier to 'force' a square by moving those loose parts around. But why do we confine those to squares? Is super-core layout just something not worth investing time into, like how interconnects are timed, non-convex tilings, and distances to I/O?
And for streaming architectures like Cerebras, why aren't other types of tilings considered, something like a hexagon grid? This could have 3 inputs / 3 outputs, which could allow more complex data flow (?). Similarly, we could have 8 connections with longer diagonals in a square tiling, etc.
I guess, what are the tradeoffs in place such that so many chip designs seem to reuse square tilings both locally and globally?
I think this could be achievable with LLMs if they had a much higher token rate. That way you'd have less downtime and could work on the problem more aggressively. How it's designed to work now, offloading cognitively, doesn't feel very good, even when you 'ship' more.
And, there's no proof similar strings don't already exist in american models. What if "{id}_{url}" works to send it to an arbitrary URL, not a preset XYZ one?
I wouldn't even grant that. It's a very one-directional problem solving approach where it's unknown to what extent it can build on its own findings.