Promising results from DeepSeek R1 for code • Hacker News Board

Re: Promising results from DeepSeek R1 for code

#251

Post by herval » Tue, Jan 28, 2025, 6:13 PM UTC

plainOldText wrote: post #245

herval wrote: post #214

Earlier quoted context omitted.

I don't think trusting that someone wrote the code was ever a good assurance of anything, and I don't see how that changes with AI. There will always be certain _individuals_ who are more reliable than others, not because they handcraft code, but because they follow through with it (make sure it works, fix bugs after release, keep an eye to make sure it worked, etc). Yes, AI will enable exponentially more people to w…

Until now, the march of technology has taken place through a realm which was somewhat limited or slowed down only by our advancements in the physical and cognitive realities. This has given us ample time to catch up, to adjust. The cognitive reality of AI, and more specifically of AI+Humans in the context of a social and globally connected world, is on a higher level of sophistication and can unfold much faster, whic…

Has it really? What evidence do we have that it's such an insanely exponential advancement?

Re: Promising results from DeepSeek R1 for code

#252

Post by kikimora » Tue, Jan 28, 2025, 6:13 PM UTC

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.…

I don’t understand. When I asked DeepSeek how to find AWS IoT Thing creation time it suggested me to use “version” field and treat it as a Unix timestamp. This is obvious nonsense. How can this tool generate anything useful other than summaries of pre-existing text? My knowledge of theory behind LLMs also suggests this is all they can do reasonably well.

When I see claims like this I suspect that either people around me somehow 10x better at promoting or they use different models.

Re: Promising results from DeepSeek R1 for code

#253

Post by simonw » Tue, Jan 28, 2025, 6:16 PM UTC

kikimora wrote: post #252

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.…

I don’t understand. When I asked DeepSeek how to find AWS IoT Thing creation time it suggested me to use “version” field and treat it as a Unix timestamp. This is obvious nonsense. How can this tool generate anything useful other than summaries of pre-existing text? My knowledge of theory behind LLMs also suggests this is all they can do reasonably well. When I see claims like this I suspect that either people around…

You're making the mistake of treating an LLM like a search engine, and expecting it to be able to answer questions directly from its training data.

Sometimes this works! But it's not guaranteed - this isn't their core strength, especially once you get into really deep knowledge of complex APIs.

They are MUCH more useful when you use them for transformation tasks: feed in examples of the APIs you need to work with, then have them write new code based on that.

Working effectively with LLMs for writing code is an extremely deep topic. Most people who think they aren't useful for code have been mislead into believing that the LLMs will just work - and that they don't first need to learn a whole bunch of unintuitive stuff in order to take advantage of the technology.

Re: Promising results from DeepSeek R1 for code

#254

Post by reitzensteinm » Tue, Jan 28, 2025, 6:16 PM UTC

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…

R1 is available on both together.ai and fireworks.ai, it should be a drop in replacement using the OpenAI API.

Re: Promising results from DeepSeek R1 for code

#255

Post by r00fus » Tue, Jan 28, 2025, 6:17 PM UTC

rane wrote: post #184

lukan wrote: post #109

Earlier quoted context omitted.

"I hope we can put to rest the argument that LLMs are only marginally useful in coding" I more often heard the argument, they are not useful for them. I agree. If a LLM would be trained on my codebase and the exact libaries and APIs I use - I would use them daily I guess. But currently they still make too many misstake and mess up different APIs for example, so not useful to me, except for small experiments. But if I…

The idea is that you give the libraries and APIs as context with your prompt.

There's a fairly low ceiling for max context tokens no matter the size of the model. Your hobby/small codebase may work, but for large codebases, you will need to do RAG and currently it's not perfect at absorbing the codebase and being able to answer questions on it.

Re: Promising results from DeepSeek R1 for code

#256

Post by cyanydeez » Tue, Jan 28, 2025, 6:17 PM UTC

entropicdrifter wrote: post #224

cyanydeez wrote: post #202

Earlier quoted context omitted.

that's basically the AI rubicon everywhere. From flying plans to programming: Soon there'll be no real fallback. When AI fails, you can't just put the controls in front of a person and expect them to have reasonable expertise to respond. Really, what seems on the horizon is a cliff of techno risks that have nothing to do with "AI will take over the world" and more "AI will be so integral to functional humanity that a…

>So it's more a conceptual belief: Will AI actually make driving cares safer or will the fatalities of AI just be so randomly stochastic that it's more acceptable. I would argue that we already accept relatively random car fatalities at a huge scale and simply engage in post-hoc rationalization of the why and how of individual accidents that affect us personally. If we can drastically reduce the rate of accidents, th…

Well, we don't accept it in the sense of "we can't blame someone", which is what I'm saying. Soon it'll be like a forest fire or a conspiracy theory.

Currently, car crashes are blamed on the individuals involved.

Re: Promising results from DeepSeek R1 for code

#257

Post by danielbln » Tue, Jan 28, 2025, 6:20 PM UTC

motorest wrote: post #220

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…

> 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 testing times for so many of them. I don't agree. LLMs work as template engines on steroids. The role of a developer now includes more code reviewing than code typing. You need the exact same core curriculum to be able to parse code, regardless if you're t…

That's a broad statement. If the IDE checks types and feeds errors back to the LLM, then that loop is very well able to fix an issue or implement a feature all on its own (see aider, cline etc )

Re: Promising results from DeepSeek R1 for code

#258

Post by karmasimida » Tue, Jan 28, 2025, 6:21 PM UTC

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.…

I am mixed on this point.

I 100% agree with you our trade is changed forever.

On the other hand, I am writing like 1000+ LOC daily, without much compromise on quality and my mental health, and thought of writing some code that is necessary but feels like a chore is not longer the case. The boost in output is incredible.

Re: Promising results from DeepSeek R1 for code

#259

Post by submeta » Tue, Jan 28, 2025, 6:22 PM UTC

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.…

The dev jobs won‘t go away, but they will change. Devs will be more and more like requirements engineers who need to understand the problem to then write prompts with the peoper context so that the llm can produce valuable and working code. And the next level will be to prompt llms to generate prompts for llms to produce code and solutions.

But already I hire less and less developers for smaller tasks. The things that I‘d assign to a dev in Ukraine to explore an idea, do a data transformation, make a UI for the internal company tool. I can do these things quicker with llm than trying to find a dev and explain the task.

Re: Promising results from DeepSeek R1 for code

#260

Post by sitkack » Tue, Jan 28, 2025, 6:23 PM UTC

LeFantome wrote: post #173

sitkack wrote: post #78

Earlier quoted context omitted.

We have already entered a new paradigm of software development, where small teams build software for themselves to solve their own problems rather than making software to sell to people. I think selling software will get harder in the future unless it comes with special affordances.

I think some of the CEOs have it right on this one. What is going to get harder is selling “applications” that are really just user friendly ways of getting data in and out of databases. Honestly, most enterprise software is just this. AI agents will do the same job. What will still matter is software that constrains what kind of data ends up in the database and ensures that data means what it is supposed to. That so…

Many applications can and should be replaced by a prompt and a database. This is the nature of increased expressive and computational power. So many whip manufacturers are about to go out of business, especially those offering whips-as-a-service.