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

Claude Opus 4.6

anthropic.com

Post reply on HN
Print view View original

691–700 of 1001 posts

Re: Claude Opus 4.6

#691

Post by sdf2erf »

atonse wrote: post #687

Wow, I have been using Open 4.6 and for the last 15 minutes, and it's already made two extremely stupid mistakes... like misunderstanding basic instructions and editing the file in a very silly, basic way. Pretty bad. Never seen this with any model before. The one bone I'll throw it was that I was asking it to edit its own MCP configs. So maybe it got thoroughly confused? I dunno what's going on, I'm going to give it…

To me its obvious.

Theres a trade off going on - in order to handle more nuance/subtleties, the models are more likely to be wrong in their outputs and need more steering. This is why personally my use of them has reduced dramatically for what I do.

Re: Claude Opus 4.6

#692

Post by freakynit »

rstuart4133 wrote: post #452
surajkumar5050 wrote: post #396

I think two things are getting conflated in this discussion. First: marginal inference cost vs total business profitability. It’s very plausible (and increasingly likely) that OpenAI/Anthropic are profitable on a per-token marginal basis, especially given how cheap equivalent open-weight inference has become. Third-party providers are effectively price-discovering the floor for inference. Second: model lifecycle econ…

> It’s very plausible (and increasingly likely) that OpenAI/Anthropic are profitable on a per-token marginal basis There any many places that will not use models running on hardware provided by OpenAI / Anthropic. That is the case true of my (the Australian) government at all levels. They will only use models running in Australia. Consequently AWS (and I presume others) will run models supplied by the AI companies fo…

Genuine question: Given Anthropic's current scale and valuation, why not invest in owning data centers in major markets rather than relying on cloud providers?

Is the bottleneck primarily capex, long lead times on power and GPUs, or the strategic risk of locking into fixed infrastructure in such a fast-moving space?

Re: Claude Opus 4.6

#694

Post by obirunda »

NiloCK wrote: post #677
obirunda wrote: post #645

Earlier quoted context omitted.

This underestimates how much of the Internet is actually compressed into and is an integral part of the model's weights. Gemini 2.5 can recite the first Harry Potter book verbatim for over 75% of the book.

I'm getting astrology when I search for this. Any links on this?

https://arxiv.org/abs/2601.02671?hl=en-US

Re: Claude Opus 4.6

#695

Post by niobe »

Aztar wrote: post #527
niobe wrote: post #462

Is there a good technical breakdown of all these benchmarks that get used to market the latest greatest LLMs somewhere? Preferably impartial.

I just ask claude and ask for sources for each one.

Reminds me of how if you make a complaint against a lawyer or a judge it's evaluated by lawyers and judges.

Re: Claude Opus 4.6

#696

Post by 1970-01-01 »

tototrains wrote: post #684
raincole wrote: post #678

Earlier quoted context omitted.

GPT 5.2 Chat: > There isn’t enough information to determine the inner diameter *for sure*. All we’re told is: - The coaster has a diameter of 9. - The glass sits on the coaster. - The glass wall thickness is 1. Nothing explicitly states that the *outer diameter of the glass equals the coaster’s diameter*. The glass could be smaller than the coaster, which is very common. *However*, if we make the usual implicit assum…

Opus 4.6 Extended thinking: ``` Finding the Inner Diameter The coaster sits beneath the glass on the table, meaning the glass rests on top of it. Assuming the coaster matches the outer diameter of the glass base: Outer diameter of glass = diameter of coaster = 9 Wall thickness = 1 (on each side, so subtract twice) dinner=9−2(1)=7d_{\text{inner}} = 9 - 2(1) = 7dinner =9−2(1)=7 The inner diameter of the glass is 7. ```…

Assumptions need to be stated or you're solving only a discreet part of the problem! Try this, see if you get another deadpan assumption.

     A solar system has 3 planets in concentric orbit. PlanetZ is the farthest with an orbit diameter of 9. PlanetY has an obit diameter one greater than PlanetX. What is the orbit diameter of PlanetX?

Re: Claude Opus 4.6

#697

Post by IAmGraydon »

ck_one wrote: post #562
golfer wrote: post #560

Earlier quoted context omitted.

There's lots of websites that list the spells. It's well documented. Could Claude simply be regurgitating knowledge from the web? Example: https://harrypotter.fandom.com/wiki/List_of_spells

It didn't use web search. But for sure it has some internal knowledge already. It's not a perfect needle in the hay stack problem but gemini flash was much worse when I tested it last time.

I'm not sure what your knowledge level of the inner workings of LLMs is, but a model doesn't need search or even an internet connection to "know" the information if it's in its training dataset. In your example, it's almost guaranteed that the LLM isn't searching books - it's just referencing one of the hundreds of lists of those spells in it's training data.

This is the LLM's magic trick that has everyone fooled into thinking they're intelligent - it can very convincingly cosplay an intelligent being by parroting an intelligent being's output. This is equivalent to making a recording of Elvis, playing it back, and believing that Elvis is actually alive inside of the playback device. And let's face it, if a time traveler brought a modern music playback device back hundreds of years and showed it to everyone, they WOULD think that. Why? Because they have not become accustomed to the technology and have no concept of how it could work. The same is true of LLMs - the technology was thrust on society so quickly that there was no time for people to adjust and understand its inner workings, so most people think it's actually doing something akin to intelligence. The truth is it's just as far from intelligence your music playback device is from having Elvis inside of it.

Re: Claude Opus 4.6

#698

Post by jazzyjackson »

viraptor wrote: post #573
ck_one wrote: post #562

Earlier quoted context omitted.

It didn't use web search. But for sure it has some internal knowledge already. It's not a perfect needle in the hay stack problem but gemini flash was much worse when I tested it last time.

If you want to really test this, search/replace the names with your own random ones and see if it lists those. Otherwise, LLMs have most of the books memorised anyway: https://arstechnica.com/features/2025/06/study-metas-llama-3...

Being that it has the books memorized (huh, just learned another US/UK spelling quirk), I would suppose feeding it the books with altered spells would get you a confused mishmash of data in the context and data in the weights.

Re: Claude Opus 4.6

#699

Post by anupamchugh » 

  Agent teams in this release is mcp-agent-mail [1] built into
  the runtime. Mailbox, task list, file locking — zero config,
  just works. I forked agent-mail [2], added heartbeat/presence
  tracking, had a PR upstream [3] when agent teams dropped. For
  coordinating Claude Code instances within a session, the
  built-in version wins on friction alone.

  Where it stops: agent teams is session-scoped. I run Claude
  Code during the day, hand off to Codex overnight, pick up in
  the morning. Different runtimes, async, persistent. Agent
  teams dies when you close the terminal — no cross-tool
  messaging, no file leases, no audit trail that outlives the
  session.

  What survives sherlocking is whatever crosses the runtime
  boundary. The built-in version will always win inside its own
  walls — less friction, zero setup. The cross-tool layer is
  where community tooling still has room. Until that gets
  absorbed too.

  [1] https://github.com/Dicklesworthstone/mcp_agent_mail
  [2] https://github.com/anupamchugh/mcp_agent_mail
  [3]
  https://github.com/Dicklesworthstone/mcp_agent_mail/pull/77

Re: Claude Opus 4.6

#700

Post by jazzyjackson »

petercooper wrote: post #531
xiomrze wrote: post #520

Earlier quoted context omitted.

Honest question, how do you know if it's pulling from context vs from memory? If I use Opus 4.6 with Extended Thinking (Web Search disabled, no books attached), it answers with 130 spells.

One possible trick could be to search and replace them all with nonsense alternatives then see if it extracts those.

A fine instruction following task but if harry potter is in the weights of the neural net, it's going to mix some of the real ones with the alternates.
Post reply on HN