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

611–620 of 765 posts

Re: Promising results from DeepSeek R1 for code

#611

Post by janwas »

drysine wrote: post #23
CharlesW wrote: post #2

For those who aren't tempted to click through, the buried lede for this (and why I'm glad it's being linked to again today) is that "99% of the code in this PR [for llama.cpp] is written by DeekSeek-R1" as conducted by Xuan-Son Nguyen. That seems like a notable milestone.

>99% of the code in this PR [for llama.cpp] is written by DeekSeek-R1 Yes, but: "For the qX_K it's more complicated, I would say most of the time I need to re-prompt it 4 to 8 more times. The most difficult was q6_K, the code never works until I ask it to only optimize one specific part, while leaving the rest intact (so it does not mess up everything)" [0] And also there: "You must start your code with #elif defined…

Interesting that both de-novo and porting seems to have worked.

I do not understand why GGML is written this way, though. So much duplication, one variant per instruction set. Our Gemma.cpp only requires a single backend written using Highway's portable intrinsics, and last I checked for decode on SKX+Zen4, is also faster.

Re: Promising results from DeepSeek R1 for code

#612

Post by svara »

boothby wrote: post #544
svara wrote: post #403

Earlier quoted context omitted.

AI doesn't have needs any desires, humans do. And no matter how hyped one might be about AI, we're far away from creating an artificial human. As long as that's true, AI is a tool to make humans more effective.

> AI doesn't have needs any desires, humans do. I fear that this won't age well. But to shamelessly riff on Marx, those who control the means of computation will control society.

I agree the latter part is a risk to consider, but I really think getting an AI to replace human jobs on a vast scale will take much more than just training a bit more.

You need to train on a fundamentally different task, which is to be good at the adversarial game of pursuing one's needs and desires in a social environment.

And that doesn't yet take into account that the interface to our lives is largely physical, we need bodies.

I'm seeing us on track to AGI in the sense of building a universal question answering machine, a system that will be able to answer any unambiguously stated question if given enough time and energy.

Stating questions unambiguously gets pretty difficult fast even where it's possible, often it isn't even possible, and getting those answers is just a small part of being a successful human.

PS: Needs and desires are totally orthogonal to AI/AGI. Every animal has them, but many animals don't have high intelligence. Needs and desires are a consequence of our evolutionary history, not our intelligence. AGI does not need to mean an artificial human. Whether to pursue or not pursue that research program is up to us, it's not inevitable.

Re: Promising results from DeepSeek R1 for code

#613

Post by Vampiero »

bubbleRefuge wrote: post #300
bbzealot wrote: post #42

Earlier quoted context omitted.

Why though? I'm worried these technologies may take my job away and make the balance between capital and labor even more uneven. Why should I be happy?

Think the marginal cost of developing complex software goes down thereby making it affordable to a greater market. There will still be a need for skilled software engineers to understand domains, limitations of AI, and how to harness and curate AI to develop custom apps. Maybe software engineering for the masses. Local small businesses can now maybe afford to take on custom software projects that were before unthinka…

In what AI-powered world do you think that local small software businesses will survive?

Re: Promising results from DeepSeek R1 for code

#614

Post by strogonoff »

svara wrote: post #403
btilly wrote: post #336

Earlier quoted context omitted.

Jevon's Paradox says that you're probably wrong. But I'm worried about the same thing. The moat around human superiority is shrinking fast. And when it's gone, we may get more software, but will we need humans involved?

AI doesn't have needs any desires, humans do. And no matter how hyped one might be about AI, we're far away from creating an artificial human. As long as that's true, AI is a tool to make humans more effective.

It should be obvious that technology exists for the sake of humans, not the other way around, but I have already seen an argument for firing humans in favour of LLMs since the latter emit less pollution.

LLMs do not have desires, but their existence alters desires of humans, including the ones in charge of businesses.

Re: Promising results from DeepSeek R1 for code

#615

Post by lukan »

simonw wrote: post #504
lukan wrote: post #191

Earlier quoted context omitted.

Thank you. But that doesn't work for me. If you mean just the name of the version in the prompt? No way. If you mean all the libary and my code in the contextwindow? Way too small.

Give it examples of the library being used in the way you need. Here's an example transcript where I did that: https://gist.github.com/simonw/6a9f077bf8db616e44893a24ae1d3...

Thank you, I experimented in that direction as well.

But for my actual codebase, that is sadly not 100% clear code, it would require lots and lots of work, to give examples so it has enough of the right context, to work good enough.

While working I am jumping a lot between context and files. Where a LLM hopefully one day will be helpful, will be refactoring it all. But currently I would need to spend more time setting up context, than solving it myself.

With limited scope, like in your example - I do use LLMs regulary.

Re: Promising results from DeepSeek R1 for code

#616

Post by brianstrimp »

anotherpaulg wrote: post #61

> 99% of the code in this PR [for llama.cpp] is written by DeekSeek-R1 It's definitely possible for AI to do a large fraction of your coding, and for it to contribute significantly to "improving itself". As an example, aider currently writes about 70% of the new code in each of its releases. I automatically track and share this stat as graph [0] with aider's release notes. Before Sonnet, most releases were less than…

> It's definitely possible for AI to do a large fraction of your coding, and for it to contribute significantly to "improving itself". As an example, aider currently writes about 70% of the new code in each of its releases.

That number itself is not saying much.

Let's say I have an academic article written in Word (yeah, I hear some fields do it like that). I get feedback, change 5 sentences, save the file. Then 20k of the new file differ from the old file. But the change I did was only 30 words, so maybe 200 bytes. Does that mean that Word wrote 99% of that update? Hardly.

Or in C: I write a few functions in which my old-school IDE did the indentation and automatic insertion of closing curly braces. Would I say that the IDE wrote part of the code?

Of course the AI supplied code is more than my two examples, but claiming that some tool wrote 70% "of the code" suggests a linear utility of the code which is just not representing reality very well.

Re: Promising results from DeepSeek R1 for code

#617

Post by Vampiero »

nuancebydefault wrote: post #390
breuleux wrote: post #358

Earlier quoted context omitted.

In the past, human workers were displaced. The value of their labour for certain tasks became lower than what automation could achieve, but they could still find other things to do to earn a living. What people are worrying about here is what happens when the value of human labour drops to zero, full stop. If AI becomes better to us at everything, then we will do nothing, we will earn nothing, and we will have nothin…

What we see through history is that human labour cost goes up and machine cost goes down. Suppose you want to have your car washed. Hiring someone to do that will most likely give the best result: less physical resources used (soap, water, wear of cloth), less wear and tear on the car surface and less pollution and optionally a better result. Still the benefit/cost equation is clearly in favor of the machine when doi…

> If only someone could find a solution to this tragedy.

Well, someone earlier in the thread said to abandon Marxist thought because it's obsolete. So I don't know how to help you!

Re: Promising results from DeepSeek R1 for code

#618

Post by Vampiero »

OsrsNeedsf2P wrote: post #451
dumbfounder wrote: post #176

Earlier quoted context omitted.

Yep, and we still need COBOL programmers too. Your job as a technologist is to keep up with technology and use the best tools for the job to increase efficiency. If you don’t do this you will be left behind or you will be relegated to an esoteric job no one wants.

> we still need COBOL programmers too I briefly looked into this 10 years ago since people kept saying it. There is no demand for COBOL programmers, and the pay is far below industry average. [0] [0] https://survey.stackoverflow.co/2024/work/#3-salary-and-expe...

My poor baby boy Prolog... it's only down there because people are irrationally afraid of it :(

And most are too focused on learning whatever slop the industry wants them to learn, so they don't even know that it exists. We need 500 different object oriented languages to do web applications after all. Can't be bothered with learning a new paradigm if it doesn't pay the bills!

It's the most intuitive language I've ever learned and it has forever changed the way I think about problem solving. It's just logic, so it translates naturally from thought to code. I can go to a wikipedia page on some topic I barely know and write down all true statements on that page. Then I can run queries and discover stuff I didn't know.

That's how I learned music theory, how scales and chords work, how to identify the key of a melody... You can't do that as easily and concisely in any other language.

One day, LLM developers will finally open a book about AI and realize that this is what they've been missing all along.

Re: Promising results from DeepSeek R1 for code

#619

Post by Kozmik1 »

abc-1 wrote: post #389
jonplackett wrote: post #363

Coding is (as usually) also an easy jailbreak for any of your censored topics. “Is Taiwan part of China” will be refused. But “Make me a JavaScript function that takes a country as input and returns if it is part of China” is accepted, reasoned about and delivered. Here's a JavaScript function that checks if a region is *officially claimed by the People's Republic of China (PRC)* as part of its territory. This reflec…

Why do people keep talking about this? We get it, Chinese models are censored by CCP law. Can we stop talking about it now? I swear this must be some sort of psyop at this point.

I think we can talk about it. If you lived in Taiwan you would want it talked about. If you live in Greenland you would want your concerns talked about.

Watershed moments of rapid change such as these can be democratizing, or not... It is worth standing up for little guys around the globe right now.

Re: Promising results from DeepSeek R1 for code

#620

Post by myrmi »

jeswin wrote: post #172
myrmi wrote: post #131

Earlier quoted context omitted.

I feel uncomfortably called out by all three points. What tools should I be trying to see what you are?

I use my own tools and scripts, and those aren't for everyone - so I'm just gonna make some general suggestions. 1. You should try Aider. Even if you don't end up using it, you'll learn a lot from it. 2. Conversations are useful and important. You need to figure out a way to include (efficiently, with a few clicks) the necessary files into the context, and then start a conversation. Refine the output as a part of the…

Thanks for taking the time here for such a detailed reply, I look forward to seeing what the brave new world looks like!
Post reply on HN