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

Claude Sonnet 4.5

anthropic.com

Post reply on HN
Print view View original

791–800 of 819 posts

Re: Claude Sonnet 4.5

#791

Post by BoorishBears »

dimgl wrote: post #637
BoorishBears wrote: post #634

Earlier quoted context omitted.

Landing site is completely broken on Safari, iOS 26, stable channel, iPhone 17 Air A lot of content is off-screen, and can't be swiped back on-screen. - Actual app did let me theough the flow, but had things like dialogs staying open and covering content (I assume a non tech person would have missed them) I think part of the AI gap right now is the UI/UX you're expecting. AI isn't able to meet the bar for UI work for…

There are so many people en masse who simply cannot accept that the days of coding manually, by hand, could be coming to an end. It's wild.

I want them to come: I'm good at product and have 101 competitors I'd smash on distribution and product with my army of coding agents.

But you're inventing the strawman that anyone who thinks the AI isn't there yet is in denial... some of us just have a higher bar than tailwind slop.

(Ironically I think SWEs most impressed by current agents are really done for... no niche non-tech knowledge to translate into novel software. But also don't have great taste/product sense: otherwise no one would have to point out that the UI/UX is not good enough.)

Re: Claude Sonnet 4.5

#792

Post by belter »

simonw wrote: post #743
whywhywhywhy wrote: post #715

Earlier quoted context omitted.

Kinda pointless listening to the opinions of people who've used previews because it's not gonna be the same model you'll experience once it gets downgraded to be viable under mass use and the benchmarks influencers use are all in the training data now and tested internally so any sort of testing like pelicans on bikes is just PR at this point.

I learned that lesson from GPT-5, where the preview was weeks long and the models kept changing during that period. This Claude preview lasted from Friday to Monday so I was less worried about major model changes. I made sure to run the pelican benchmark against the model after 10am on Monday (the official release date) just to be safe. The only thing I published that I ran against the preview model was the Claude co…

Testing this, its way more aggressive on throttle back than previous model, and message token lengths. Constantly stops in the middle of an action if its not a simple request. I presume you did not have resource limitations during the preview?

Re: Claude Sonnet 4.5

#793

Post by simonw »

belter wrote: post #792
simonw wrote: post #743

Earlier quoted context omitted.

I learned that lesson from GPT-5, where the preview was weeks long and the models kept changing during that period. This Claude preview lasted from Friday to Monday so I was less worried about major model changes. I made sure to run the pelican benchmark against the model after 10am on Monday (the official release date) just to be safe. The only thing I published that I ran against the preview model was the Claude co…

Testing this, its way more aggressive on throttle back than previous model, and message token lengths. Constantly stops in the middle of an action if its not a simple request. I presume you did not have resource limitations during the preview?

No, the preview was effectively unlimited usage (for two days).

Re: Claude Sonnet 4.5

#794

Post by MIC132 »

AlecSchueler wrote: post #735
MIC132 wrote: post #333

Earlier quoted context omitted.

In my (admittedly very limited) experience with trying to talk about "controversial" topics, Claude seems to be much stricter about shutting down the conversation fast.

I've been talking to it daily for months and never had anything shut down. My only experience with that was DeepSeek not wanting to talk about internal perceptions of intellectual property laws within China.

Probably depends on the type of sensitive topic. The ones I got shut down in were related to sex (but not explicit themselves, not like direct descriptions or something) and I mostly mention it since ChatGPT for example had no problem discussing that but Claude shut down immediately.

Re: Claude Sonnet 4.5

#795

Post by ranguna »

0xbadcafebee wrote: post #776
ranguna wrote: post #687

Earlier quoted context omitted.

This HN post is about claude 4.5 and you come here speaking about how "claude" does not give you satisfactory answer when, most likely, you didn't even try claude 4.5 in the first place. Claude 4.5 after a few web searches and running a couple python scripts for analysis: Yes, your configuration should work! Based on my analysis, two strings of four Phono Solar PS440M8GFH panels will be compatible with the EG4 12kPV…

I said "Claude 4.5" in my comment

Indeed you have, sorry about that.

I'm not sure how you got that answer and how I got mine. It's possible you don't have extended thinking and Web search enabled?

Either way, the point still stands, unless what claude told me is invalid in your opinion.

Re: Claude Sonnet 4.5

#796

Post by virtualritz »

So I was using Opus exclusively (Max plan) to write Rust since June.

CC switched to Sonnet 4.5 by default yesterday, I'm just very unimpressed. It seems like a considerable regression. Probably this is related to me using it to write Rust and not Python or JS/TS?

Example: I asked it to refactor a for loop to functional code with rayon, compiler barfs about mutation (it used par_iter()). It rolls back the code to what it was before. Then this happens:

Me: There is par_iter_mut().

Sonnet: Ah yes, but we can't use par_iter_mut() on self.vertices because we're calling self.set_vertex_position() which needs mutable access to the entire mesh (not just the vertices map).

However, self is &mut. Wtf?

This would have never happened with Opus, i.e. Opus would have used par_iter_mut() to start with (or applied the right fix w/o hand-holding after the compile failed with par_iter()).

I had just a bunch of those wtfs since yesterday from more or less zero before. I.e. it doesn't feel like coincidence.

Re: Claude Sonnet 4.5

#797

Post by lsaferite »

sarchertech wrote: post #719
janee wrote: post #707

Earlier quoted context omitted.

haha no it's an old (15years old) abandoned enterprise app running on-prem that hasn't seen updates in more than a decade.

Wow Rails 3 came out 15 years ago, so that thing started life out of date.

> enterprise app

> started life out of date

That tracks my experiences.

Re: Claude Sonnet 4.5

#798

Post by ccvannorman »

nosianu wrote: post #671
wyre wrote: post #528

Earlier quoted context omitted.

I'm not sure exactly what you mean by the vibe coding experience non-coders want, but if it's one-shotting a buildable codebase off of an unspecific prompt, the major breakthrough would have to be brain-computer interfaces so the agent can literally read the user's mind. If that same person approached a software development company with the same prompt without following up with any other details, they won't get good…

> the major breakthrough would have to be brain-computer interfaces so the agent can literally read the user's mind. And even that would not be enough. In reality, it would have to put the user to sleep and go through various dream scenarios to have the user's brain really build an internal model that is not there in the first place. No brain interface can help find what is not there. We usually need interactions wit…

That is also not enough. An agent could build an application that functions, but you also need to have a well-designed underlying architecture if you want the application to be extensible and maintainable - something the original dreamer may not even be capable of - so perhaps a shared extended dream share with a Sr. architect is also needed. Oh wait .. I guess we're back to square 1 again? lol

Re: Claude Sonnet 4.5

#799

Post by majortennis »

janee wrote: post #628
Philpax wrote: post #243

Earlier quoted context omitted.

Well, yes, that's Claude Code. And OpenAI Codex. And Google Gemini CLI. Your average dev can just use those.

Yes but you need to setup quite a bit of tooling to provide feedback loops. It's one thing to get an llm to do something unattended for long durations, it's a other to give it the means of verification. For example I'm busy upgrading a 500k LoC rails 1 codebase to rails 8 and built several DSLs that give it proper authorised sessions in a headless browser with basic html parsing tooling so it can "see" what affect it…

I've tried asking it to log every request and response to a project_log.md but it routinely ignores that.

I've also tried using playwright for testing in a headless browser and taking screenshots for a blog that can effectively act as a log , it just seems like too tall an order for it.

It sounds like you're streets ahead of where I am could you give me some pointers on getting started with a feed back loop please

Post reply on HN