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

781–790 of 819 posts

Re: Claude Sonnet 4.5

#781

Post by gwern »

hackernewds wrote: post #763
coolfox wrote: post #646

Earlier quoted context omitted.

> A lack of determinism comes from many places, but primarily: 1) The models change 2) The models are not deterministic... models themselves are deterministic, this is a huge pet peeve of mine, so excuse the tangent, but the appearance of nondeterminism comes from a few sources, but imho can be largely attributed to the probabilistic methods used to get appropriate context and enable timely responses. here's an examp…

Here is a definite scientific nail down and solve for non-determinism in LLM outputs (Mira Murati's new outfit but really credit the author) https://bff531bb.connectionism.pages.dev/blog/defeating-nond...

Requires a login?

Re: Claude Sonnet 4.5

#782

Post by martin-t »

dwaltrip wrote: post #643
martin-t wrote: post #583

Earlier quoted context omitted.

In a way, yes, you'd be paying the people not just to write down the rules but to discover them first. And there's the accuracy/correctness/interpretability tradeoff. But also, have there been any attempts on the scale of the Manhattan project attempting to create a GOFAI? Because one idea I ran into is that we might be able to use genAI to create a GOFAI soon. And it would be as hard as using genAI for any kind of l…

You can’t “write down the rules” for intelligence. Not for any reasonable definition of “writing”. The medium of writing is not rich enough to express what is needed. This is why GOFAI failed.

Do you believe intelligence can be achieved using ANNs? If so, ANNs can be serialized, therefore writing is rich enough.

It might not be an easy to work with format though. If you believe the broad LLM architecture is capable of reaching true intelligence, then writing is still enough because all LLMs are is the written training data and the written training algorithm. It's just that is was impossible to pay people to write enough training data and provide enough compute to process it before.

Re: Claude Sonnet 4.5

#783

Post by kirkoplamen »

Forgeties79 wrote: post #350
mh- wrote: post #304

Earlier quoted context omitted.

The reality of that specific ask is it would not be difficult to build, but I believe it would be extremely difficult to build and offer at a price that users would pay for. So you're unlikely to find a commercial offering that does that using a (V)LM.

Yeah I imagine so. Hell I would pay like $100 for them to just do it once. If they really could do it with like 99% accuracy I would pay upwards of $300 tbh. Still, that’s probably not good enough lol

Hey bro, I'd like to take this project using Claude for $300 :) Do you mind contacting me? stxcth9aoj at mozmail.com

Re: Claude Sonnet 4.5

#784

Post by tripzilch »

simonw wrote: post #779
tripzilch wrote: post #758

Earlier quoted context omitted.

I get it, you would trust yourself if you said that, but it doesn't really matter whether you say that or not, what counts for your ongoing credibility if you will preface every future blog post with, whether you got special access, a special deal, sponsorship, or the fact that you didn't get any of those things. You're a reviewer. This is how reviewers stay credible. If you don't disclose your relationship with the…

I recently started doing that: https://simonwillison.net/about/#disclosures and https://simonwillison.net/tags/disclosures/

awesome, thanks a lot that's important but ... sorry I just checked those, and I do think it's better to do it on a per-article basis, because a lot of your audience (I'm guessing) comes from external links, not browsing your website

this is (or should be) a pretty standard thing to do on youtube review channels (that I would trust), and it's not a bad thing to remind people of, on every occasion, plus it can function as a type of "canary" in cases of particularly restrictive NDAs

Re: Claude Sonnet 4.5

#785

Post by mwiesenthal »

coolfox wrote: post #646
manofmanysmiles wrote: post #521

I haven't shouted into the void for a while. Today is as good a day as any other to do so. I feel extremely disempowered that these coding sessions are effectively black box, and non-reproducible. It feels like I am coding with nothing but hopes and dreams, and the connection between my will and the patterns of energy is so tenuous I almost don't feel like touching a computer again. A lack of determinism comes from m…

> A lack of determinism comes from many places, but primarily: 1) The models change 2) The models are not deterministic... models themselves are deterministic, this is a huge pet peeve of mine, so excuse the tangent, but the appearance of nondeterminism comes from a few sources, but imho can be largely attributed to the probabilistic methods used to get appropriate context and enable timely responses. here's an examp…

It's also a pet peeve of mine, enough that I actually wrote a blog about it

https://hi-mil.es/blog/human-slop-vs-ai-slop

Re: Claude Sonnet 4.5

#786

Post by dang »

catigula wrote: post #733
dang wrote: post #508

Earlier quoted context omitted.

Please don't cross into personal attack, regardless of how wrong someone is or you feel they are. Also, please stop posting flamebait to HN generally. It's not what this site is for, and destroys what it is for. https://news.ycombinator.com/newsguidelines.html

What is "flamebait"? Unpopular beliefs I genuinely hold? I looked it up and the definition seems to include being disingenuous, which I am not. Claiming I am is annoying and feels like it is a personal attack.

It doesn't include being disingenuous. We can't know what your intent is (https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que....

Re: Claude Sonnet 4.5

#787

Post by catigula »

dang wrote: post #786
catigula wrote: post #733

Earlier quoted context omitted.

What is "flamebait"? Unpopular beliefs I genuinely hold? I looked it up and the definition seems to include being disingenuous, which I am not. Claiming I am is annoying and feels like it is a personal attack.

It doesn't include being disingenuous. We can't know what your intent is ( https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que... .

Thanks

Re: Claude Sonnet 4.5

#788

Post by Huppie »

bobbylarrybobby wrote: post #778
iagooar wrote: post #258

Anecdotal evidence. I have a fairly large web application with ~200k LoC. Gave the same prompt to Sonnet 4.5 (Claude Code) and GPT-5-Codex (Codex CLI). "implement a fuzzy search for conversations and reports either when selecting "Go to Conversation" or "Go to Report" and typing the title or when the user types in the title in the main input field, and none of the standard elements match, a search starts with a 2s de…

It's also my experience that Claude loves to reimplement the wheel instead of reading code to look for an existing implementation of what it wants to do.

I've been working with Opus 4 on ultrathink quite a bit recently and did some quick tests with Sonnet 4.5, I'm fairly impressed, especially with its speed but I did feel it was a lot less strict with my rules, existing patterns, etc. compared to Opus 4.

Maybe it's better with a better CLAUDE.md structure? I don't use those a lot, just telling Opus to think got 'good enough' results I guess. Not sure.

I hope there's an Opus 4.5 coming out soon too. In the meantime I'll see if I can get to do better with some extra prompting or I'll go back to Opus of if I don't need the speedier responses.

Re: Claude Sonnet 4.5

#789

Post by manofmanysmiles »

jstummbillig wrote: post #666
manofmanysmiles wrote: post #521

I haven't shouted into the void for a while. Today is as good a day as any other to do so. I feel extremely disempowered that these coding sessions are effectively black box, and non-reproducible. It feels like I am coding with nothing but hopes and dreams, and the connection between my will and the patterns of energy is so tenuous I almost don't feel like touching a computer again. A lack of determinism comes from m…

And now imagine you'd have to rely on humans to build your software instead

This is the question though isn't it?

With sufficient structure and supervision, will a "team" of agents out-perform a team of humans?

Military, automotive and other industries have developed rigorous standards consisting of among other things detailed processes for developing software.

Can there be an AI waterfall? With sufficiently unambiguous, testable requirements, and a nice scaffolding of process, is it possible to achieve the dream of managers, and eliminate software engineers? My intuition is evenly split.

Re: Claude Sonnet 4.5

#790

Post by enjo »

nerdix wrote: post #570
enjo wrote: post #216

Earlier quoted context omitted.

This is more or less my experience with Go right now. For a bunch of reasons I want to avoid the standard React, Typescript, and Node stack but the sheer velocity that might enable from the LLM side might make it worth it.

Wait... Are you saying that your experience with Go has been bad? I would think Go would be as good as any other language (if not better). The language itself is simple, the Go team is very methodical about adding new features so it changes fairly slowly, it has excellent built in CLI based tooling that doesn't require third party packages or applications, and there are plenty of large open source Go codebases to tra…

Yep my experience has been pretty bad. As in Claude with Opus can rarely produce even compiling code in my particular project (a year old, mid-complexity one). This is with adhering to best practices including a robust Claude.md and detailed PRD's.
Post reply on HN