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

Claude 4

anthropic.com

Post reply on HN
Print view View original

271–280 of 1001 posts

Re: Claude 4

#271

Post by hansmayer »

j_maffe wrote: post #243
hansmayer wrote: post #62

Earlier quoted context omitted.

Right, but on the other hand... how is it even useful? Let's say it can beat the game, so what? So it can (kind of) summarise or write my emails - which is something I neither want nor need, they produce mountains of sloppy code, which I would have to end up fixing, and finally they can play a game? Where is the killer app? The gaming approach was exactly the premise of the original AI efforts in the 1960s, that teac…

> Where is the killer app? My man, ChatGPT is the sixth most visited website in the world right now.

But I did not ask "what was the sixth most visited website in the world right now?", did I? I asked what was the killer app here. I am afraid vague and un-related KPIs will not help here, otherwise we may as well compare ChatGPT and PornHub based on the number of visits, as you seem to suggest.

Re: Claude 4

#272

Post by MattSayar »

WillPostForFood wrote: post #114
MattSayar wrote: post #87

Earlier quoted context omitted.

Same. And I JUST tried their GitHub Action agentic thing yesterday (wrote about it here[0]), and it honestly didn't perform very well. I should try it again with Claude 4 and see if there are any differences. Should be an easy test [0] https://mattsayar.com/personalized-software-really-is-coming...

It would be pretty funny if your "not today. Maybe tomorrow" proposition actually did happen the next day.

It literally almost did!

Re: Claude 4

#273

Post by vFunct »

chermi wrote: post #201
vFunct wrote: post #120

Earlier quoted context omitted.

Yah Claude tends to output 1200+ line architectural specification documents while Gemini tends to output ~600 line. (I just had to write 100+ architectural spec documents for 100+ different apps) Not sure why Claude is more thorough and complete than the other models, but it's my go-to model for large projects. The OpenAI model outputs are always the smallest - 500 lines or so. Not very good at larger projects, but p…

I'd interested to hear more about your workflow. I use Gemini for discussing the codebase, making ADR entries based on discussion, ticket generation, documenting the code like module descriptions and use cases+examples, and coming up with detailed plans for implementation that cursor with sonnet can implement. Do you have any particular formats, guidelines or prompts? I don't love my workflow. I try to keep everythin…

For a large project, the real human trick for you to do is to figure out how to partition it down to separate apps, so that individual LLMs can work on them separately, as if they were their own employees in separate departments.

You then ask LLMs to first write features for the individual apps (in Markdown), giving it some early competitive guidelines.

You then tell LLMs to read that features document, and then write an architectural specification document. Tell it to maybe add example data structures, algorithms, or user interface layouts. All in Markdown.

You then feed these two documents to individual LLMs to write the rest of the code, usually starting with the data models first, then the algorithms, then the user interface.

Again, the trick is to partition your project to individual apps. Also an app isn't the full app. It might just be a data schema, a single GUI window, a marketing plan, etc.

The other hard part is to integrate the apps back together at the top level if they interact with each other...

Re: Claude 4

#274

Post by ethbr1 »

amarcheschi wrote: post #265
gokhan wrote: post #250

Interesting alignment notes from Opus 4: https://x.com/sleepinyourhat/status/1925593359374328272 "Be careful about telling Opus to ‘be bold’ or ‘take initiative’ when you’ve given it access to real-world-facing tools...If it thinks you’re doing something egregiously immoral, for example, like faking data in a pharmaceutical trial, it will use command-line tools to contact the press, contact regulators, try to lock yo…

We definitely need models to hallucinate things and contact authorities without you knowing anything (/s)

I mean, they were trained on reddit and 4chan... swotbot enters the chat

Re: Claude 4

#275

Post by kevindamm »

minimaxir wrote: post #179
KaoruAoiShiho wrote: post #61

Is this really worthy of a claude 4 label? Was there a new pre-training run? Cause this feels like 3.8... only swe went up significantly, and that as we all understand by now is done by cramming on specific post training data and doesn't generalize to intelligence. The agentic tooluse didn't improve and this says to me that it's not really smarter.

So I decided to try Claude 4 Sonnet against my "Given a list of 1 million random integers between 1 and 100,000, find the difference between the smallest and the largest numbers whose digits sum up to 30." benchmark I tested against Claude 3.5 Sonnet: https://news.ycombinator.com/item?id=42584400 The results are here ( https://gist.github.com/minimaxir/1bad26f0f000562b1418754d67... ) and it utterly crushed the proble…

I did a quick review of its final answer and looks like there are logic errors.

All three of them get the incorrect max-value bound (even with comments saying 9+9+9+9+3 = 30), so early termination wouldn't happen in the second and third solution, but that's an optimization detail. The first version would, however, early terminate on the first occurrence of 3999 and take whatever the max value was up to that point. So, for many inputs the first one (via solve_digit_sum_difference) is just wrong.

The second implementation (solve_optimized, not a great name either) and third implementation, at least appear to be correct... but that pydoc and the comments in general are atrocious. In a review I would ask these to be reworded and would only expect juniors to even include anything similar in a pull request.

I'm impressed that it's able to pick a good line of reasoning, and even if it's wrong about the optimizations it did give a working answer... but in the body of the response and in the code comments it clearly doesn't understand digit extraction per se, despite parroting code about it. I suspect you're right that the model has seen the problem solution before, and is possibly overfitting.

Not bad, but I wouldn't say it crushed it, and wouldn't accept any of its micro-optimizations without benchmark results, or at least a benchmark test that I could then run.

Have you tried the same question with other sums besides 30?

Re: Claude 4

#276

Post by EgoIncarnate »

gokhan wrote: post #250

Interesting alignment notes from Opus 4: https://x.com/sleepinyourhat/status/1925593359374328272 "Be careful about telling Opus to ‘be bold’ or ‘take initiative’ when you’ve given it access to real-world-facing tools...If it thinks you’re doing something egregiously immoral, for example, like faking data in a pharmaceutical trial, it will use command-line tools to contact the press, contact regulators, try to lock yo…

The should call it Karen mode.

Re: Claude 4

#277

Post by drusepth »

KaoruAoiShiho wrote: post #61

Is this really worthy of a claude 4 label? Was there a new pre-training run? Cause this feels like 3.8... only swe went up significantly, and that as we all understand by now is done by cramming on specific post training data and doesn't generalize to intelligence. The agentic tooluse didn't improve and this says to me that it's not really smarter.

To be fair, a lot of people said 3.7 should have just been called 4. Maybe they're just bridging the gap.

Re: Claude 4

#278

Post by skruger »

sndean wrote: post #129

Using Claude Opus 4, this was the first time I've gotten any of these models to produce functioning Dyalog APL that does something relatively complicated. And it actually runs without errors. Crazy (at least to me).

Would you mind sharing what you did? stefan@dyalog

Re: Claude 4

#279

Post by ethbr1 »

SparkyMcUnicorn wrote: post #251
liorn wrote: post #233

Earlier quoted context omitted.

I asked it about Tailwind CSS (since I had problems with Claude not aware of Tailwind 4): > Which version of tailwind css do you know? > I have knowledge of Tailwind CSS up to version 3.4, which was the latest stable version as of my knowledge cutoff in January 2025.

Interesting. It's claiming different knowledge cutoff dates depending on the question asked. "Who is president?" gives a "April 2024" date.

Question for HN: how are content timestamps encoded during training?

Re: Claude 4

#280

Post by k8sToGo »

Seems like Github just added it to Copilot. For now the premium requests do not count, but starting June 4th it will.
Post reply on HN