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

Claude Opus 4.7

anthropic.com

Post reply on HN
Print view View original

681–690 of 1001 posts

Re: Claude Opus 4.7

#681

Post by andai »

vessenes wrote: post #588
babelfish wrote: post #343

Earlier quoted context omitted.

Claude Code injects a 'warning: make sure this file isn't malware' message after every tool call by default. It seems like 4.7 is over-attending to this warning. @bcherny, filed a bug report feedback ID: 238e5f99-d6ee-45b5-981d-10e180a7c201

Interesting. The model card mentions 4.7 is much more attentive to these instructions and suggests you will need to review and soften or remove or focus them at times.

It's been known for years that prompts which boost performance with one model, can harm performance with a different model. The same goes for harnesses. It looks like they'll need to customize Claude Code's prompts depending on which model is running, for optimal results.

For example if you read the prompts, it's pretty clear that a lot of them are leftovers from the early days when the models had way less common sense than they do now. I think you could probably remove 2/3rds of those over-explained rules now and it would be fine. (In fact you might even expect to see improvement to performance due to decreased prompt noise.)

Re: Claude Opus 4.7

#682

Post by axpy906 »

MarkMarine wrote: post #516
puppystench wrote: post #364

Earlier quoted context omitted.

Does this mean Claude no longer outputs the full raw reasoning, only summaries? At one point, exposing the LLM's full CoT was considered a core safety tenet.

Anthropic was chirping about Chinese model companies distilling Claude with the thinking traces, and then the thinking traces started to disappear. Looks like the output product and our understanding has been negatively affected but that pales in comparison with protecting the IP of the model I guess.

That’s kind of funny since a Chinese model started the thinking chains being visible in Claude and OA in the first place.

Re: Claude Opus 4.7

#683

Post by Arubis »

So far most of what I'm noticing is different is a _lot_ more flat refusals to do something that Opus 4.6 + prior CC versions would have explored to see if they were possible.

Re: Claude Opus 4.7

#684

Post by andai »

MintPaw wrote: post #448
bayesnet wrote: post #276

This is a CC harness thing than a model thing but the "new" thinking messages ('hmm...', 'this one needs a moment...') are extraordinarily irritating. They're both entirely uninformative and strictly worse than a spinner. On my workflows CC often spends up to an hour thinking (which is fine if the result is good) and seeing these messages does not build confidence.

Sounds really minor, but was actually a big contributor to me canceling and switching. The VS Code extension has a morphing spinner thing that rapidly switches between these little catch phrases. It drives me crazy, and I end up covering it up with my right click menu so I can read the actual thinking tokens without that attention vampire distracting me. And of course they recently turned off all third party harness…

I'm not sure if this is official, but from what I gathered, they just bill 3rd party stuff as extra usage now:

https://news.ycombinator.com/item?id=47633568

(They were against ToS before (might still be?), and people were having their Anthropic accounts banned. Actually charging people money for the tokens they're using seems like a much more sensible move.)

Re: Claude Opus 4.7

#686

Post by kadushka »

andai wrote: post #674
davesque wrote: post #662

> We stated that we would keep Claude Mythos Preview’s release limited and test new cyber safeguards on less capable models first. Opus 4.7 is the first such model: its cyber capabilities are not as advanced as those of Mythos Preview (indeed, during its training we experimented with efforts to differentially reduce these capabilities). We are releasing Opus 4.7 with safeguards that automatically detect and block req…

The fundamental tension is that the models are getting weirdly good at hacking while still sort of sucking at a bunch of economically valuable tasks. So they've hit the point where the models are simultaneously too smart (dangerous hacking abilities) and too stupid (can't actually replace most employees). So at this point they need to make the models bigger, but they're already too big. So the only thing left to do i…

models are getting weirdly good at hacking while still sort of sucking at a bunch of economically valuable tasks

like most human hackers

Re: Claude Opus 4.7

#687

Post by antirez »

trueno wrote: post #455

noticing sharp uptick in "i switched to codex" replies lately. a "codex for everything" post flocking the front page on the day of the opus 4.7 release me and coworker just gave codex a 3 day pilot and it was not even close to the accuracy and ability to complete & problem solve through what we've been using claude for. are we being spammed? great. annoying. i clicked into this to read the differences and initial exp…

Are you sure you selected GPT 5.4-xhigh as model, in Codex? Because this makes a huge difference, and with this setting in my experience Codex outperforms Opus for almost every coding/reasoning task. Opus is still better often times when there is to call a lot of tools, interact with servers to do operations and alike, but not always. But for low level coding, Codex with GPT 5.4-xhigh is really powerful.

Re: Claude Opus 4.7

#688

Post by pkilgore »

JamesSwift wrote: post #460
simonw wrote: post #203

I'm finding the "adaptive thinking" thing very confusing, especially having written code against the previous thinking budget / thinking effort / etc modes: https://platform.claude.com/docs/en/build-with-claude/adapti... Also notable: 4.7 now defaults to NOT including a human-readable reasoning token summary in the output, you have to add "display": "summarized" to get that: https://platform.claude.com/docs/en/build-…

Its especially concerning / frustrating because boris’s reply to my bug report on opus being dumber was “we think adaptive thinking isnt working” and then thats the last I heard of it: https://news.ycombinator.com/item?id=47668520 Now disabling adaptive thinking plus increasing effort seem to be what has gotten me back to baseline performance but “our internal evals look good“ is not good enough right now for what ma…

Seconded. After disabling adaptive thinking and using a default higher thinking, I finally got the quality I'm looking for out of Opus 4.6, and I'm pleased with what I see so far in Opus 4.7.

Whatever their internal evals say about adaptive thinking, they're measuring the wrong thing.

Re: Claude Opus 4.7

#689

Post by jacobsenscott »

earthnail wrote: post #664
davesque wrote: post #662

> We stated that we would keep Claude Mythos Preview’s release limited and test new cyber safeguards on less capable models first. Opus 4.7 is the first such model: its cyber capabilities are not as advanced as those of Mythos Preview (indeed, during its training we experimented with efforts to differentially reduce these capabilities). We are releasing Opus 4.7 with safeguards that automatically detect and block req…

I feel it’s fine as a short term solution, and probably a good thing. Gives the good guys some time to stay on top. Always remember: a defender must succeed every time , an attacker only once.

Given the list of very large companies in the "glasswing" project - it is likely every competent state actor and criminal organization already has access to Mythos in one way or another. Meanwhile the opensource volunteers responsible for the security of the entire internet don't have access.

Re: Claude Opus 4.7

#690

Post by sharms »

dgb23 wrote: post #337
simonw wrote: post #203

I'm finding the "adaptive thinking" thing very confusing, especially having written code against the previous thinking budget / thinking effort / etc modes: https://platform.claude.com/docs/en/build-with-claude/adapti... Also notable: 4.7 now defaults to NOT including a human-readable reasoning token summary in the output, you have to add "display": "summarized" to get that: https://platform.claude.com/docs/en/build-…

Don't look at "thinking" tokens. LLMs sometimes produce thinking tokens that are only vaguely related to the task if at all, then do the correct thing anyways.

This is because the "thinking" you see is a summary by a highly quantized model - not the actual model, to mask these tokens
Post reply on HN