daxfohl

37 ms·

601.
by daxfohl 2y ago
I was originally thinking about it as like a bazel plugin for large codebases. Each module would have its own LLM context, and it might make it easier to put whole modules into the context, plus summaries of the dependencies. That could wor
602.
by daxfohl 2y ago
Would there be any way to distribute RAG across multiple smaller models? Rather than one giant model handling your entire document base, have it be more of a tree where the top level classifies the docs into top-level categories and sends i
603.
by daxfohl 2y ago
Though "Air Magnus" rolls off the tongue better than "Air Nepomniachtchi".
604.
by daxfohl 2y ago
My hunch is it won't be much different, even when we can simply ask a machine that doesn't have a cached proof, "prove riemann hypothesis" and it thinks for ten seconds and spits out a fully correct proof. As Erdos(I thi
605.
by daxfohl 2y ago
Oh, good to hear! My knowledge is about 7 years out of date at this point. I know typescript and mypy are pretty good at inferring which options of an ADT are impossible in various contexts and letting you do stuff like this, so great to he
606.
by daxfohl 2y ago
Main thing I dislike about errors as ADTs is it forces indentation in both cases. I miss go's "iff err..." otherwise continue as normal. Yes monads can allow that, but they come with their own problems.
607.
by daxfohl 2y ago
I wonder if we'll start seeing a shift in compute spend, moving away from training time, and toward inference time instead. As we get closer to AGI, we probably reach some limit in terms of how smart the thing can get just training on
608.
by daxfohl 2y ago
Yeah but we can only use 10% of our brain too, so.
609.
by daxfohl 2y ago
Oh, right, the CAD thing was sketchier. IIRC there wasn't any feature of OpenGL/DirectX that was nerfed, it was just an agreement with CAD companies to reject the GPUs that didn't have the workstation bit fused. For this perf
610.
by daxfohl 2y ago
The first one is talking about the quantum realm. So, absent of measurement or some kind of collapsing of the wave function, pure quantum states are 100% reversible. So the question is, where does the transition happen from the quantum worl
611.
by daxfohl 2y ago
I mean, the paper you linked below describes the situation fairly well. We know the current semiclassical understanding of gravity is incomplete, but near the event horizon of black holes, QFT+GR are sufficient to model the physics. It&#x27
612.
by daxfohl 2y ago
Correct, that's why I explicitly said "classical" in that sentence. Classical relativity means our understanding from Einstein's field equations, prior to the information-theoretic approaches.
613.
by daxfohl 2y ago
At steady state, a classical black hole is fully described just mass, charge, and angular momentum. So there are no individual particles. Which itself was disconcerting to physicists because in the quantum world, information is supposed to
614.
by daxfohl 2y ago
The article is missing on a couple points. The analysis assumes all Alice's light is emitted radially exactly outward from the center of the black hole. In reality, light is emitted in all directions, and anything emitted at even sligh
615.
by daxfohl 2y ago
This is inaccurate. The object can fall in. Where the argument fails is that it relies on the idea that an observer will see the object redshift forever. But that applies in the classical GR realm only. However, when combined with QFT (whic
616.
by daxfohl 2y ago
I used modelsim free edition years ago. Also got a $200ish test board just to see things work on real hardware. Never did anything super complex but I think I got my money's worth of tinkering around out of it.
617.
by daxfohl 2y ago
Verilog Having done all the functional, lisp-like, logic, stack-based, etc., "non-mainstream" languages, the one that most blew my mind so far is Verilog. Everything executing all the time, all at once, is just such a completely d
618.
by daxfohl 2y ago
I wish there was a linter that could limit it though. It's like coders' subconsciouses say "the feature exists, so I have to use it to the greatest extent possible". Or, "assigning intermediate values is now an anti
619.
by daxfohl 2y ago
Yeah but C# is. Or, compiled to IL and JITted. Unless there is an AOT thing now that can truly compete with c and rust. Been out of the ecosystem for a while.
620.
by daxfohl 2y ago
Both have "unsafe" escape hatches, so all code is possible to write efficiently. Just some cases are harder to prove correct in the type systems of each.
621.
by daxfohl 2y ago
So why all the interest in rust, comparatively? My assumption is since there is a GC, and it is not native code, there are too many use cases where it can't apply, but rust can. Once there is a way to have it compete with rust in every
622.
by daxfohl 2y ago
I understand GP's point, but to yours I'll add they have also open sourced and mainstreamed some of their key technologies. So it's not all bad.
623.
by daxfohl 2y ago
Given that we've been throwing cash at every conceivable idea for the last ten plus years, yet when speaking of unicorns we still have to refer back to airbnb and uber, seems like we're well past "peak unicorn" and well
624.
by daxfohl 2y ago
One odd thing that never occurred to me until now is that two black holes that pass each other at near light speed, if they intersect by even a nanometer, will be guaranteed to merge. That nanometer has to fall into both singularities due t
625.
by daxfohl 2y ago
But there's still two singularities until one falls into the other.
626.
by daxfohl 2y ago
I was there three years ago. Most (possibly all) of what was running on bare metal was Windows/Hyper-V. Some Azure service backends ran on Linux VMs; AKS backend for instance was on Linux I believe. Most were Windows/C# though. I
627.
by daxfohl 2y ago
It foreshadows a degenerate form of AGI, where AI isn't nearly smart as humans, but is convincing enough to bring human intelligence down to its own level.
628.
by daxfohl 2y ago
Article writer is probably also an LLM.
629.
by daxfohl 2y ago
Same reason male seahorses get pregnant.
630.
by daxfohl 2y ago
This is what I found hard about Haskell. It's so tempting to DRY to such a phenomenal degree, but then the slightest requirement change somewhere in the middle breaks some abstraction and bubbles all the way out. It's almost like

More ›