dnautics

20 ms·

481.
by dnautics 8mo ago
no. I'm a pretty skilled programmer and I definitely have to intervene and fix an architectural problem here and there, or gently chastise the LLM for doing something dumb. But there are also many cases where the LLM has seen somethi
482.
by dnautics 8mo ago
no nulls is truly not a problem. Of all the problems I have seen, I have never seen Claude mess up with nulls in Elixir.
483.
by dnautics 8mo ago
I'm not 100% convinced that "plugging memory error holes" was right at the compiler level. Currently building out clr, which uses a heuristic (not formal verification) method for checking soundness of zig code, using ~"r
484.
by dnautics 8mo ago
they kinda tried that and it is my understanding that it sucks. One example of "direct linear mapping": https://pdfs.semanticscholar.org/5080/78a97f4778eee488e11032...
485.
by dnautics 8mo ago
we should 1) pay reviewers. 2) you can't publish unless a reviewer replicates your work. yes. It can be done. https://www.orgsyn.org/
486.
by dnautics 8mo ago
incorrect. having a higher caught error rate means that you consume more context on the way to your solution which makes for worse results, both by spending more time in the context danger zone, and by losing more on compaction handoffs. g
487.
by dnautics 8mo ago
i mean as a first order approximation context (the key resource that seems to affect quality) doesn't depend on real compilation speed, presumably the agent is suspended and not burning context while waiting for compliation
488.
by dnautics 8mo ago
how about not making the error in the first place
489.
by dnautics 8mo ago
you'd think, and yet its like 70% on autocoderbenchmark
490.
by dnautics 8mo ago
nah, the agents jump around files anyways.
491.
by dnautics 8mo ago
> I think the more you can shift to compile time the better when it comes to agents not born out by evidence. rust is bottom-mid tier on autocoderbenchmark. typescript is marginally bettee than js shifting to compile time is not necess
492.
by dnautics 8mo ago
what honestly is the difference between an mcp and a skill + instructions + curl. Really it seems to me the difference is that an mcp could be more token-efficient, but it isn't, because you dump every mcp's instructions all the
493.
by dnautics 8mo ago
you know what you could do? You could write a skill that turns mcps on or off!
494.
by dnautics 8mo ago
it proves that the algorithm is embeddable in a bigger transformer of ~similar architecture.
495.
by dnautics 8mo ago
symbol manipulation in transformers is fp arithmetic?
496.
by dnautics 8mo ago
and he has had several meetings change his opinion of the other party for no apparent reason (eg zelensky extrapolation is futile
497.
by dnautics 8mo ago
> Architecture wise and also just from a code quality perspective I have yet to encounter AI generated code that passes my quality bar. You should consider trying to using AI in a programming language that scores high in the AutoCoderBen
498.
by dnautics 8mo ago
> you've got silent data corruption showing up 3 months later on an edge case nobody tested I mean this happens in normal development?
499.
by dnautics 8mo ago
it's a good sign? We are having these debates, and it's aired in public?
500.
by dnautics 8mo ago
> pattern matches the "idea shape" of words in the "idea space it does much more than this. first layer has an attention mechanism on all previous tokens and spits out an activation representing some sum of all relations
501.
by dnautics 8mo ago
thats unlikely. but they are awfully lot like turing machines (k/v cache ~ turing tape) so their architecture is strongly predisposed to be able to find any algorithm , possibly including reasoning
502.
by dnautics 8mo ago
> still applicable belief that slavery was really ended by industrialization making abolition economically... not crazy especially as slavery was supplanted by debt, which is in a way fractional slavery (minus the chattel part ofc)
503.
by dnautics 8mo ago
you could probably autoclave it with a standard dental/tattooing autoclave (~500 USD and requires a gas stove)
504.
by dnautics 8mo ago
It's more or less what comes out of the box with plan mode, plus a few extra bits?
505.
by dnautics 8mo ago
this is literally reinventing claude's planning mode, but with more steps. I think Boris doesn't realize that planning mode is actually stored in a file. https://x.com/boristane/status/20216286521366732
506.
by dnautics 8mo ago
hate to say it, because who likes monopolies, but it's easier to send people docusign because then they don't go Wtf?
507.
by dnautics 8mo ago
library vs framework (you call a library, a framework calls you) is pretty typical and arguably very useful distinction. calling a framework necessarily magic is the weird thing.
508.
by dnautics 8mo ago
int main() is magic (and it's a framework).
509.
by dnautics 8mo ago
the advantage of frameworks is that there are about 20-ish security/critical usage considerations, of which you will remember about 5. if you don't use a framework, you are so much more likelihood of getting screwed. you should
510.
by dnautics 8mo ago
the common factor was sort of left as an exercise to the reader to think about moats in the age of AI... but basically anything that has touchpoints to the legal and financial systems im not gonna touch with a 20 ft vibecoded pole.

More ›