Live data from Hacker News

Promising results from DeepSeek R1 for code

simonwillison.net

681–690 of 765 posts

Re: Promising results from DeepSeek R1 for code

#681

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…

It will be interesting to see which models update to the "Gulf of America" and which keep the "Gulf of Mexico" in their training data/self-censorship stages.

Re: Promising results from DeepSeek R1 for code

#682
post #677

Earlier quoted context omitted.

Thanks for the concern of my happiness, but can I express my concern for your eye sight - where did you read that I am unhappy about results? My, as you have named it - "unhappiness", is about not knowing rules and not being pointed out, that I am overstepping those rules. If you are going with the approach, that silence is also an answer, then yes they can be considered as results, just as receiving complete garbage…

maybe the developers are just tired of this childish game and decided to block interactions like this in place of creating news headlines? Garbage in, garbage out. DeepSeek is more efficient but even more efficient is to not waste computing.

[flagged]

Re: Promising results from DeepSeek R1 for code

#683
post #682

Earlier quoted context omitted.

maybe the developers are just tired of this childish game and decided to block interactions like this in place of creating news headlines? Garbage in, garbage out. DeepSeek is more efficient but even more efficient is to not waste computing.

[flagged]

[flagged]

Re: Promising results from DeepSeek R1 for code

#684

Earlier quoted context omitted.

I fully believe this as well. And I have 15 years of SWE experience at top tech. Its over for this field

[flagged]

Frankly, I can't agree on any of this. Majority of the state of AI is way faar from where it can be really useable. We are nowhere near AI, that is emulating our intellect, besides - the byproduct of AI is much bigger than any pesky LLMs - understanding how our brain works and eventually making human megamind, that can persist through hormonal changes that humans go and what makes our life so unstable and full of changes.

Robotics - is nowhere near the promise as well - we are nowhere near biological entities(not made from metal) with syntetic brains, not to mention biological robotic arms that humans can use as prostetics while they are regrowing natural limbs. So much to learn.

As for the Jesus. That is not really a deep subject. We know what Jesus was as a human - his real life and his violent and human nature(as a military representative of cult, that was lead by John the Baptist) has nothing to do with how it is portrayed by religion. History of how Christianity started and including about Jesus was one of the easiest problems that I have encountered and wished to know and I fullfilled just recently.

Re: Promising results from DeepSeek R1 for code

#685

This is exactly what Michael Crichton warned of in Westworld. Computers writing their own programs and designing their own hardware. Soon humans won't be able to understand what makes them tick, and when they run amok, we're helpless. I don't think it'll really come to that, but if it does, you can't say you haven't been warned.

I'll start to "worry" when the AI creates something to replace React. If what's coming of LLMs is react/next.js code, I'm not worried at all.

Re: Promising results from DeepSeek R1 for code

#687

Earlier quoted context omitted.

Is DeepSeek really that big of a deal that everyone else should worry?

Deepseek is a big deal but we should be happy not worried that our tools are improving.

This added momentum to two things: reducing AI costs and increasing quality.

I don't know when the threshold of "replace the bottom X% of developers because AI is so good" happens for businesses based on those things, but it's definitely getting closer instead of stalling out like the bubble predictors claimed. It's not a bubble if the industry is making progress like this.

Re: Promising results from DeepSeek R1 for code

#688

Earlier quoted context omitted.

Glad to hear you’re finding aider useful! There’s a faq entry about how these stats are computed [0]. Basically using git blame, since aider is tightly integrated with git. The faq links to the script that computes the stats. It’s not designed to be used on any repo, but you (or aider) could adapt it. You’re not the first to ask for these stats about your own repo, so I may generalize it at some point. [0] https://ai…

Thank you so much for linking me to that! I think an `aider stats`-type command would be really cool (it would be cool to calculate stats based activity since the first aider commit or all-time commits of the repo).

Slightly longer than `aider stats` but here you go:

  uv run --with=semver,PyYAML,tqdm https://raw.githubusercontent.com/Aider-AI/aider/refs/heads/main/scripts/blame.py

Re: Promising results from DeepSeek R1 for code

#689
post #682

Earlier quoted context omitted.

[flagged]

[flagged]

>>>Now talking serious.

Honestly, you should change your statements for other people, as by default I assumed, that you were serious... I'm being sarcastic here - have to add this as people do not hear sarcastic tone in text and assume it to be serious.

Also, what makes you think I did not try it for code? It did not generate code, that I found acceptable to me and required a lot more work. But at least it gave me honest answer there, that it could offer links to better papers. I don't see that much difference with ChatGPT, as they might probably allow more queries to paying customers, but on the other side - did I mentioned that I read TOS? I would never use AI tools to create my own code for commercial use, that is not open source. Because why in the right mindd should I do that?

I eventually got bored with this tool, just ike with ChatGPT(also, I can write better code anyway, so not real use to me now). Code is not as important as data which is the basis of programming. And I am still interested in understanding logic of other programmers, when I see a code(and behaviour of their creation), that I have to ask wth were they thinking. And test it more.

I am a human, that can programm and I will ask political questions first, because morality and politics are affecting my efficiency as a programmer. If I can't think freely, I won't work on that. So, unless you are a CCP shill and not concerned that your code and logics is recorded and eventually can be stolen, you can use whatever.

Re: Promising results from DeepSeek R1 for code

#690

My number 1 criticism of long term LLM claims is that we already hit the limit. If you see the difference between a 7B model and a 70B model, its only slightly impressive. a 70B and a 400B model is almost unnoticeable. Does going from 400B to 2T do anything? Every layer like using python to calculate a result, or using chain of thought, destroys the purity. It works great for Strawberries, but not great for developin…

When you use words like purity, you're making an ideological value judgment. You're not talking about computer science or results.
Post reply on HN