vatsachak

8 ms·

151.
by vatsachak 2mo ago
These days I have a good intuition as to where an agent can actually do what I want as opposed to creating spaghetti. The best analogy I have come up with is that LLMs are like water and will take the shape of the bowl you build. The upside
152.
by vatsachak 2mo ago
This might need controversial but there were too many devs to begin with. I'm young and know people who are getting jobs, it's just that they actually enjoy programming.
153.
by vatsachak 2mo ago
Some of us use nix as a package manager. And others...
154.
by vatsachak 2mo ago
Yeah and sometimes even worse, they do things correctly but in most spaghetti way. I've had to rewrite my whole codebase. It's just the thrill of getting things done quick. Not getting things done right.
155.
by vatsachak 2mo ago
How does a git hook need more labor than an LLM?
156.
by vatsachak 2mo ago
Yeah. LLMs have their place and they are definitely super human at short length tasks, but I feel like a large part of the "AI boom" is trying to get the computer to do something in a worse way then it already could.
157.
by vatsachak 2mo ago
Do you use deepseek models to improve deepseek training and inference?
158.
by vatsachak 2mo ago
True that
159.
by vatsachak 2mo ago
LLMs are both the best compression and prediction algorithm for English text.
160.
by vatsachak 2mo ago
I feel like both abstractions and LLMs are time savers, so why not combine both. Also, if I don't create abstractions and guide the LLM I've found that the codebase starts getting out of my reach. And until LLMs can maintain code
161.
by vatsachak 2mo ago
https://bellard.org/ts_zip/
162.
by vatsachak 2mo ago
In a world where we see 0+ strings more often the first compressor would be better
163.
by vatsachak 2mo ago
Even with an ideal AGI you would need abstractions because in order to talk in specifics you would need an abstraction.
164.
by vatsachak 2mo ago
But you end up writing your own abstractions in the process?
165.
by vatsachak 2mo ago
Abstractions aren't a burden though, they are more like a support. A good abstraction let's you write 10 programs in the time it would have taken to write one. Also, we now have "production" languages with generics and m
166.
by vatsachak 2mo ago
I mean yes the python interpreter is not great, but I'm talking about bugs created by accidentally passing a negative number where a positive one should go, using a tensor of the wrong size etc. I don't have any ml experience with
167.
by vatsachak 2mo ago
The machine learning community decided to choose Python as the main language, which costs probably tens of millions of dollars in bugs every year due to it not being type checked (astral sh is fixing this though)
168.
by vatsachak 2mo ago
This is such a poor idea. LLM Agents don't have taste (abstraction capabilities). They are kind of like water, they take the shape of the container they're in. They are so much powerful when you build your own framework they can w
169.
by vatsachak 2mo ago
Fair I work in very mathematical code
170.
by vatsachak 2mo ago
SIMD is actually underrated still. Programmers should always be thinking about it. It's a free 4x in a lot of cases
171.
by vatsachak 2mo ago
You can tell that a human wrote it because X is a noun and Y is an adjective
172.
by vatsachak 2mo ago
That's a fair use case for them though. LLMs are superhuman at short term performance engineering/debugging/testing. But even that comes with the caveat that these commits come from talented coders using LLMs as a grad studen
173.
by vatsachak 2mo ago
Yeah, why hasn't AMD added every single CUDA feature lol
174.
by vatsachak 2mo ago
I would submit it but the pr was too long to fit on GitHub
175.
by vatsachak 2mo ago
Did you try using the website? It's broken lol
176.
by vatsachak 2mo ago
LLM code is obvious to spot. If an LLM designed a screwdriver set it would make a screwdriver for each head instead of making a replaceable head. LLMs are superhuman at short term coding such as debugging and writing tests though and you&#x
177.
by vatsachak 2mo ago
Nah this take is wrong. I used Claude code with my custom skill and I wrote a more performant scheduler than the default Linux one in Rust. It's not just productivity, it's life changing.
178.
by vatsachak 2mo ago
> now taking us months where it'd have taken weeks... So 0<whatever<1
179.
by vatsachak 3mo ago
Why do LLM coded websites seem so obvious? Like an earnest project in this direction would probably look like a 90s html page
180.
by vatsachak 3mo ago
Is there any advantage of this over an agent using a Google search? I feel like the main benefit of such an endeavor would be to create a centralized database of mathematical theorems. But why aren't we using Lean theorems as the nodes

More ›