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

861–870 of 1001 posts

Re: Claude 4

#861

Post by vharish »

Rudybega wrote: post #198
minimaxir wrote: post #102

Earlier quoted context omitted.

Context window is unchanged for Sonnet. (200k in/64k out): https://docs.anthropic.com/en/docs/about-claude/models/overv... In practice, the 1M context of Gemini 2.5 isn't that much of a differentiator because larger context has diminishing returns on adherence to later tokens.

I'm going to have to heavily disagree. Gemini 2.5 Pro has super impressive performance on large context problems. I routinely drive it up to 4-500k tokens in my coding agent. It's the only model where that much context produces even remotely useful results. I think it also crushes most of the benchmarks for long context performance. I believe on MRCR (multi round coreference resolution) it beats pretty much any other…

Totally agreed on this. The context size is what made me switch to Gemini. Compared to Gemini, Claude's context window length is a joke.

Particularly for indie projects, you can essentially dump the entire code into it and with pro reasoning model, it's all handled pretty well.

Re: Claude 4

#862

Post by sublimefire »

dbingham wrote: post #743

It feels like these new models are no longer making order of magnitude jumps, but are instead into the long tail of incremental improvements. It seems like we might be close to maxing out what the current iteration of LLMs can accomplish and we're into the diminishing returns phase. If that's the case, then I have a bad feeling for the state of our industry. My experience with LLMs is that their code does _not_ cut i…

There are a couple of things where LLMs are OK from the business perspective. Even if they are so so you can still write large amounts of mediocre code without the need to consume libraries. Think about GPL’d code, no need to worry about that because one dev can rewrite those libraries into proprietary versions without licensing constraints. Another thing is that LLMs are OK for an average company with few engineers that need to ship mountains of code across platforms, they would make mistakes anyway so LLMs should not make it worse.

Re: Claude 4

#863

Post by jug »

hsn915 wrote: post #416

I can't be the only one who thinks this version is no better than the previous one, and that LLMs have basically reached a plateau, and all the new releases "feature" are more or less just gimmicks.

This is my feeling too, across the board. Nowadays, benchmark wins seem to come from tuning, but then causing losses in other areas. o3, o4-mini also hallucinates more than o1 in SimpleQA, PersonQA. Synthetic data seems to cause higher hallucination rates. Reasoning models at even higher risk due to hallucinations risking to throw the model off track at each reasoning step.

LLM’s in a generic use sense are done since already earlier this year. OpenAI discovered this when they had to cancel GPT-5 and later released the ”too costly for gains” GPT-4.5 that will be sunset soon.

I’m not sure the stock market has factored all this in yet. There needs to be a breakthrough to get us past this place.

Re: Claude 4

#864

Post by jjmarr »

GolDDranks wrote: post #770
BeetleB wrote: post #531

Earlier quoted context omitted.

I've noticed an interesting trend: Most people who are happy with LLM coding say something like "Wow, it's awesome. I asked it to do X and it did it so fast with minimal bugs, and good code", and occasionally show the output. Many provide even more details. Most people who are not happy with LLM coding ... provide almost no details. As someone who's impressed by LLM coding, when I read a post like yours, I tend to ha…

Two problems: 1) Writing a high-performance memory allocator for a game engine in Rust: https://github.com/golddranks/bang/tree/main/libs/arena/src (Still work in progress, so it's in a bit messy state.) Didn't seem to understand the design I had in mind, and/or the requirements and goes on tangents and starts changing the design. In the end, coded the main code myself and used LLM for writing tests with some success…

Try Roo Code in Orchestrator mode or Cline in plan mode. It will do tons of requirements analysis before starting work.

Re: Claude 4

#865

Post by renerick »

arewethereyeta wrote: post #815

I feel like these AI companies are in a gold rush while somebody else is selling the shovels. I've never jumped ship for the same service, from a vendor to another... so often. Looks like a race to the bottom where the snake eats itself.

NVIDIA sells the shovels, then OpenAI/Anthropic/Google make an excavator out of shovels (NVDIA also seems to work on their own excavators), then some startup starts selling excavator wrapper. I don't know if there are any snakes at the bottom, but there's surely a whole lot of shovel layers on the way down.

Re: Claude 4

#866

Post by parodysbird »

elcritch wrote: post #772
BoredPositron wrote: post #630

Earlier quoted context omitted.

Because it's alchemy and everyone believes they have an edge on turning lead into gold.

I've been thinking for a couple of months now that prompt engineering, and therefore CoT, is going to become the "secret sauce" companies want to hold onto. If anything that is where the day to day pragmatic engineering gets done. Like with early chemistry, we didn't need to precisely understand chemical theory to produce mass industrial processes by making a good enough working model, some statistical parameters, an…

The thing with alchemy was not that their hypotheses were wrong (they eventually created chemistry), but that their method of secret esoteric mysticism over open inquiry was wrong.

Newton is the great example of this: he led a dual life, where in one he did science openly to a community to scrutinize, in the other he did secret alchemy in search of the philosopher's stone. History has empirically shown us which of his lives actually led to the discovery and accumulation of knowledge, and which did not.

Re: Claude 4

#867

Post by artdigital »

lherron wrote: post #330
waleedlatif1 wrote: post #108

I really hope sonnet 4 is not obsessed with tool calls the way 3-7 is. 3-5 was sort of this magical experience where, for the first time, I felt the sense that models were going to master programming. It’s kind of been downhill from there.

Overly aggressive “let me do one more thing while I’m here” in 3.7 really turned me off as well. Would love a return to 3.5’s adherence.

Oh jeez yes. I completely forgot that this was a thing. It’s tendency to do completely different things “while at it” was ridiculous

Re: Claude 4

#868

Post by dudeinjapan »

nicce wrote: post #421
threeducks wrote: post #300

Earlier quoted context omitted.

> Which version of tailwind css do you know? LLMs can not reliably tell whether they know or don't know something. If they did, we would not have to deal with hallucinations.

We should use the correct term: to not have to deal with bullshit.

I think “confabulation” is the best term.

“Hallucination” is seeing/saying something that a sober person clearly knows is not supposed to be there, e.g. “The Vice President under Nixon was Oscar the Grouch.”

Harry Frankfurt defines “bullshitting” as lying to persuade without regard to the truth. (A certain current US president does this profusely and masterfully.)

“Confabulation” is filling the unknown parts of a statement or story with bits that sound as-if they could be true, i.e. they make sense within the context, but are not actually true. People with dementia (e.g. a certain previous US president) will do this unintentionally. Whereas the bullshitter generally knows their bullshit to be false and is intentionally deceiving out of self-interest, confabulation (like hallucination) can simply be the consequence of impaired mental capacity.

Re: Claude 4

#869

Post by atworkc »

theptip wrote: post #800
hsn915 wrote: post #488

Earlier quoted context omitted.

I asked it a few questions and it responded exactly like all the other models do. Some of the questions were difficult / very specific, and it failed in the same way all the other models failed.

Great example of this general class of reasoning failure. “AI does badly on my test therefore it’s bad”. The correct question to ask is, of course, what is it good at? (For bonus points, think in terms of $/task rather than simply being dominant over humans.)

"AI does badly on my test much like other AI's did before it, therefore I don't immediately see much improvement" is a fair assumption.

Re: Claude 4

#870

Post by jakemanger »

rootlocus wrote: post #821
jakemanger wrote: post #818

Been playing around with it in Cursor and have to say I'm pretty dang impressed. Did notice a few times that it got stuck in a loop of trying to repeatedly make its implementation better. I suppose that is ok for some use cases but it started overthinking. I then gently prompted it by saying "you're way overthinking this. Just do a simple change like ..." I guess there's still a purpose for developers

You are now fine tuning their models, and you pay them money to do it.

Even if you opt out of them storing your data? (a checkbox in the settings)
Post reply on HN