Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Promising results from DeepSeek R1 for code

simonwillison.net

Post reply on HN
Print view View original

121–130 of 765 posts

Re: Promising results from DeepSeek R1 for code

#121

Post by bick_nyers »

jeswin wrote: post #86
simonw wrote: post #52

Earlier quoted context omitted.

"Jobs are going to be lost unless there's somehow a demand for more applications." That's why I'm not worried. There is already SO MUCH more demand for code than we're able to keep up with. Show me a company that doesn't have a backlog a mile long where most of the internal conversations are about how to prioritize what to build next. I think LLM assistance makes programmers significantly more productive, which makes…

> That's why I'm not worried. There is already SO MUCH more demand for code than we're able to keep up with. Show me a company that doesn't have a backlog a mile long where most of the internal conversations are about how to prioritize what to build next. I worry about junior developers. It will be a while before vocational programming courses retool to teach this new way of writing code, and these are going to be te…

I definitely agree with you in the interim regarding junior developers. However, I do think we will eventually have the AI coding equivalent of CICD built into perhaps our IDE. Basically, when an AI generated some code to implement something, you chain out more AI queries to test it, modify it, check it for security vulnerabilities etc.

Now, the first response some folks may have is, how can you trust that the AI is good at security? Well, in this example, it only needs to be better than the junior developers at security to provide them with benefits/learning opportunities. We need to remember that the junior developers of today can also just as easily write insecure code.

Re: Promising results from DeepSeek R1 for code

#122

Post by bick_nyers »

onetimeusename wrote: post #107
jeswin wrote: post #86

Earlier quoted context omitted.

> That's why I'm not worried. There is already SO MUCH more demand for code than we're able to keep up with. Show me a company that doesn't have a backlog a mile long where most of the internal conversations are about how to prioritize what to build next. I worry about junior developers. It will be a while before vocational programming courses retool to teach this new way of writing code, and these are going to be te…

Just as a side note, at my university about half the CS people are in the AI track. I would guess that number will keep increasing. There is also a separate major that kind of focuses on AI/psychology that is pretty popular but I am not sure how many people are in it. A good number of the students have some kind of "AI startup". Also, although it violates the honor code, I would be willing to bet many students use AI…

An actual hardcore technical AI "psychology" program would actually be really cool. Could be a good onboarding for prompt engineering (if it still exists in 5 years).

Re: Promising results from DeepSeek R1 for code

#123

Post by UncleEntity »

20k wrote: post #74
jeswin wrote: post #31

> 99% of the code in this PR [for llama.cpp] is written by DeekSeek-R1 I hope we can put to rest the argument that LLMs are only marginally useful in coding - which are often among the top comments on many threads. I suppose these arguments arise from (a) having used only GH copilot which is the worst tool, or (b) not having spent enough time with the tool/llm, or (c) apprehension. I've given up responding to these.…

Eh it performed a 1:1 conversion of ARM NEON to wasm SIMD, which with the greatest will in the world is pretty trivial work. Its something that ML is good at, because its the same problem area as "translate this from english to french", but more mechanistic This is a task that would likely have taken as long to write by hand as the AI took to do it, given how long the actual task took to execute. 98% of the work is f…

IDK, I was playing with Claude yesterday/this morning and before I hit the free tier context limit it managed to create a speech-to-phoneme VQ-VAE contraption with a sliding window for longer audio clips and some sort of "attention to capture relationships between neighboring windows" that I don't quite understand. That last part was due to a suggestion it provided where I was like "umm, ok..."

Seems pretty useful to me where I've read a bunch of papers on different variational autoencoder but never spent the time to learn the torch API or how to set up a project on the google.

In fact, it was so useful I was looking into paying for a subscription as I have a bunch of half-finished projects that could use some love.

Re: Promising results from DeepSeek R1 for code

#124

Post by realo »

simonw wrote: post #57
lenerdenator wrote: post #53

Earlier quoted context omitted.

My greatest problem is duplicating the secret sauce of GHCP: it has access to your project and can use it as context. Admittedly, I haven't looked too hard, but how could I do that with a model from, say, Ollama and run exclusively on my machine?

There are a bunch of tools that might be able to do that. I'd start by exploring https://aider.chat/

Neat. Can I use aider with a local model running in LMStudio (or ollama)?

After a very quick reading of their pages it does not seem so.

Hopefully I am wrong...

Re: Promising results from DeepSeek R1 for code

#125

Post by amarcheschi »

ngxson wrote: post #115

Hi I'm Xuan-Son, Small correct, I'm not just asking it to convert ARM NEON to SIMD, but for the function handling q6_K_q8_K, I asked it to reinvent a new approach (without giving it any prior examples). The reason I did that was because it failed writing this function 4 times so far. And a bit of context here, I was doing this during my Sunday and the time budget is 2 days to finish. I wanted to optimize wllama (wasm…

Aistudio.google.com offers free long context chats (1/2mln tokens), just select the appropriate model, 1206 or 2.0 flash thinking

Re: Promising results from DeepSeek R1 for code

#126

Post by simonw »

realo wrote: post #124
simonw wrote: post #57

Earlier quoted context omitted.

There are a bunch of tools that might be able to do that. I'd start by exploring https://aider.chat/

Neat. Can I use aider with a local model running in LMStudio (or ollama)? After a very quick reading of their pages it does not seem so. Hopefully I am wrong...

Yes you can: https://aider.chat/docs/llms/ollama.html

Re: Promising results from DeepSeek R1 for code

#127

Post by withinboredom »

simonw wrote: post #99
UncleOxidant wrote: post #95

Earlier quoted context omitted.

> That's why I'm not worried. There is already SO MUCH more demand for code than we're able to keep up with. Show me a company that doesn't have a backlog a mile long where most of the internal conversations are about how to prioritize what to build next. And yet many companies aren't hiring developers right now - folks in the C suite are thinking AI is going to be eliminating their need to hire engineers. Also "dema…

I continue to suspect that the hiring problems are mainly due to massive over-hiring during Covid, followed by layoffs that flooded the market with skilled developers looking for work. I'd love to see numbers around the "execs don't think they need engineers because of AI" factor. I've heard a few anecdotal examples of that but it's hard to tell if it's a real trend or just something that catches headlines.

I think execs don’t see the problems we have with AI because you don’t need to be an expert to be an exec. I run into the edges of AI every day. There are things it is good at and things not so good at, and it varies from model to model and context to context (you can have two conversations with the same model, about the same thing, and get vastly different outputs; eg a test that uses different assertion patterns/libraries that are different from the rest of the project). As an “expert” or “highly skilled” person, I recognize these issues when I see them, but to a layman, it just looks like code.

Re: Promising results from DeepSeek R1 for code

#128

Post by Scipio_Afri »

simonw wrote: post #52
jeswin wrote: post #31

> 99% of the code in this PR [for llama.cpp] is written by DeekSeek-R1 I hope we can put to rest the argument that LLMs are only marginally useful in coding - which are often among the top comments on many threads. I suppose these arguments arise from (a) having used only GH copilot which is the worst tool, or (b) not having spent enough time with the tool/llm, or (c) apprehension. I've given up responding to these.…

"Jobs are going to be lost unless there's somehow a demand for more applications." That's why I'm not worried. There is already SO MUCH more demand for code than we're able to keep up with. Show me a company that doesn't have a backlog a mile long where most of the internal conversations are about how to prioritize what to build next. I think LLM assistance makes programmers significantly more productive, which makes…

100% agree with this take. People are spouting economic fallacies, and it’s in part cause CEOs don't want the stock prices to fall too fast. Eventually people will widely realize this and by then the economic payoffs are still immense.

Re: Promising results from DeepSeek R1 for code

#129

Post by simonw »

ngxson wrote: post #115

Hi I'm Xuan-Son, Small correct, I'm not just asking it to convert ARM NEON to SIMD, but for the function handling q6_K_q8_K, I asked it to reinvent a new approach (without giving it any prior examples). The reason I did that was because it failed writing this function 4 times so far. And a bit of context here, I was doing this during my Sunday and the time budget is 2 days to finish. I wanted to optimize wllama (wasm…

Thanks very much for sharing your results so far.

Re: Promising results from DeepSeek R1 for code

#130

Post by thrance »

jeswin wrote: post #31

> 99% of the code in this PR [for llama.cpp] is written by DeekSeek-R1 I hope we can put to rest the argument that LLMs are only marginally useful in coding - which are often among the top comments on many threads. I suppose these arguments arise from (a) having used only GH copilot which is the worst tool, or (b) not having spent enough time with the tool/llm, or (c) apprehension. I've given up responding to these.…

Did you even look at the generated code? DeepSeek simply rewrote part of the inference code making use of SIMD instructions on wasm. It literally boils down to inserting `if defined __wasm_simd128__` at some places then rewritting the loops to do floating point operations two by two instead of one after the other (which is where the 2X claim comes from). This is very standard and mostly boilerplate.

Useful, sure, in that it saved some time in this particular case. But most of the AI-generated code I interact with is a hot unmaintainable mess of very verbose code, which I'd argue actually hurts the project in the long term.

Post reply on HN