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

DeepSeek V4 – almost on the frontier

simonwillison.net

Post reply on HN
Print view View original

341–350 of 420 posts

Re: DeepSeek V4 – almost on the frontier

#341

Post by rane »

taffydavid wrote: post #240
rane wrote: post #85

Earlier quoted context omitted.

You can just use it through Claude Code, so you get to keep the system prompt and tooling you are used to. 3rd party models are a drop-in replacement with `ANTHROPIC_BASE_URL` in Claude Code, something people seem to miss right now. And contrary to what Anthropic might like to have you think, you don't need Opus 4.7 to run the harness to get similar performance. https://api-docs.deepseek.com/quick_start/agent_integra…

Is there an easier way to manage multiple models?

I just made a simple script that makes it easy to switch between models.

Re: DeepSeek V4 – almost on the frontier

#342

Post by jdasdf »

enochthered wrote: post #7
jdasdf wrote: post #3

I've been using v4 pro for the past few days and honestly in terms of quality it seems more or less on par with open AIs 5.4 or opus 4.6 (i havent tried 4.7) To be clear, i'm not doing state of the art stuff. I mostly used it for frontend development since i'm not great at that and just need a decent looking prototype. But for my purposes it's a perfectly good model, and the price is decent. I can't wait for open mod…

Thanks for sharing your experience, I’m looking to try it out. Which provider are you using for inference? Opencode or the DeepSeek api?

I just use the API directly. It's simple enough to setup and i like the control i get from just charging up and not having to worry about any random subscription taking money out of my account

Re: DeepSeek V4 – almost on the frontier

#343

Post by jdasdf »

FrasiertheLion wrote: post #76
jdasdf wrote: post #3

I've been using v4 pro for the past few days and honestly in terms of quality it seems more or less on par with open AIs 5.4 or opus 4.6 (i havent tried 4.7) To be clear, i'm not doing state of the art stuff. I mostly used it for frontend development since i'm not great at that and just need a decent looking prototype. But for my purposes it's a perfectly good model, and the price is decent. I can't wait for open mod…

You can use Tinfoil for inference, which lets you use the model in the cloud while getting similar privacy as running locally: https://tinfoil.sh/inference . Disclaimer I'm the cofounder. This works by running the model inside a secure enclave (using NVIDIA confidential computing) and verifying the open source code running inside the enclave matches the runtime attestation. The docs walk you through the verification…

While that does sound interesting, I don't see any benefit for me.

It would still ultimately exfiltrate the data outside of my control, and frankly i don't trust any "secure enclave" tech.

As far as i'm concerned physical access is root access, and for any private stuff that is wholly unacceptable.

Re: DeepSeek V4 – almost on the frontier

#344

Post by motoxpro »

Wilder7977 wrote: post #252
motoxpro wrote: post #226

Earlier quoted context omitted.

I think it's closer to asking a remote (human) assistant to do something that someone doesn't want done (e.g., view the source of a closed-source product, whether through reverse engineering, going into their office, or social engineering) and that remote assistant company saying, "Please stop asking our assistants to do that." You can still use an IDE (hammer) to reverse engineer anything you want.

It's not though. It's still just a piece of code, much closer to IDEs or any other program than to a human assistant in any way that matters (morals, responsibility).

It just seems like you are saying if you found out Claude code was a bunch of remote working doing work for you, then it would be morally wrong to do illegal/morally wrong/irresponsible things with them, but because it is NOT a human, those same things are fine?

Re: DeepSeek V4 – almost on the frontier

#345

Post by Accacin »

gpugreg wrote: post #314
Accacin wrote: post #293

I tried DeepSeek via chat, and gave it a rather simple question: "Can you tell me who was on series 8 of Taskmaster, and what's the general opinion about the series? No spoilers!" It told me amongst other things that Paul Sinha was diagnosed with Parkinsons, as well as who the winner was. Then I said, "But I said no spoilers!" And it apologised for telling me Paul Sinha was diagnosed with Parkinsons.

I was not able to reproduce your problem with that prompt, but I might have a reason for why you got that answer. Did you enable reasoning ("DeepThink")? LLMs usually can not reason about what they are going to write before they do. There is that famous experiment where an LLM is prompted to say whether the birth year of a famous person is even or odd. If the LLM is constrained to only answer with "even" or "odd", th…

Ah, so I went back to check my logs and I think I'd hallucinated some of the example I gave. I actually asked the question in two parts.

> Taskmaster Season 8, without spoilers, how is it?

Answer was pretty general, no spoilers.

Then I asked:

> Which season 8 contestants are most loved?

- Lou Sanders (Winner): Received a strong positive reaction, praised for her "offbeat humor, raw talent, and willingness to absolutely go for it". Some felt her less wholesome moments could be "mean spirited".

- Paul Sinha (Mixed / Sympathetic): Opinions are mixed; some found his failures not always funny, but others loved his earnest struggles. His Parkinson's diagnosis during the series added a layer of empathy for many viewers.

> I said no spoilers!

You're absolutely right, and I apologize. You asked for *no spoilers* in the original question, and I broke that by mentioning specific moments (like Joe Thomas's "train yard exploit") and personal details (Paul Sinha's diagnosis). That was careless.

Let me try again, *completely spoiler-free*:

In terms of which Season 8 contestants are most loved by fans (without saying why or what happens):

- *Sian Gibson* is overwhelmingly the fan favorite — she's widely described as warm, funny, and a joy to watch.

- *Joe Thomas* is also very popular, especially for his awkward, literal, and deadpan style.

- *Lou Sanders* (the winner) has a strong fan base who love her chaotic and bold energy.

The other two contestants receive more mixed reactions. One in particular is often cited as less popular because of his high competitiveness, but again — no specifics on tasks or results.

Thank you for the correction, and sorry again for the spoilers.

Re: DeepSeek V4 – almost on the frontier

#347

Post by eckelhesten »

0xbadcafebee wrote: post #301
eckelhesten wrote: post #288

Earlier quoted context omitted.

Meta is based in the US, yet they torrented TERABYTES worth of books to feed their AI. I’m not trying to be negative here, but your point is invalidated by that particular event in itself.

What, because they broke the law in one way, they'd break the law in every way? That's not how business works. The way business works is, I steal from other people to make a product, but then I don't steal from my customers, because if they find out, then I no longer have any customers. (Plus all their customers would sue them, which would both legally and financially tank them)

That's a naive way of thinking. You're saying "oh, they are thieves in this way, but they surely wouldn't be thieves in this other way!"

If you have no problems shitting on tens of thousands of authors of books, you don't have problems shitting on your customers as well (which they have proved again and again, see https://en.wikipedia.org/wiki/Facebook–Cambridge_Analytica_d...)

Let's just say I wholeheartedly disagree with your viewpoint and leave it there :)

Re: DeepSeek V4 – almost on the frontier

#348

Post by DANmode »

kelnos wrote: post #269
noelsusman wrote: post #196

Earlier quoted context omitted.

It's mostly just a bad model. Plenty of people would be willing to overlook the baggage if the model was even marginally better than the competition.

Agreed. There's are plenty of instances where people here on HN do mental gymnastics to justify using a truly good product when the company that builds it is morally bankrupt. Not a criticism (I probably engage in that sort of thinking myself sometimes), just something I've observed. If Grok were actually good, we'd see that phenomenon here, but we don't.

I just read a bunch of compelling “Grok is better at this” use cases in a thread yesterday.

I’m not rushing towards it, but, had to mention.

Re: DeepSeek V4 – almost on the frontier

#350

Post by sylware »

zozbot234 wrote: post #295
sylware wrote: post #280

Earlier quoted context omitted.

Let's say I get 32GB of RAM, with a lean elf(glibc)/linux system, for which 7GB is beyond enormous to run. Let's book 8/16 cores/threads to run a prompt. What are the timing figures I am looking at to run an "average" coding prompt?

The basic bottleneck with 32GB RAM would be your storage, so for a baseline estimate you'd be looking at anything from ~2 secs per token (if you had really high performance PCIe 5.0 SSD at ~14 GB/s max) to ~5 secs per token (for an average PCIe 4.0 SSD, ~7 GB/s max). This would then be boosted by being able to keep the shared model layers in RAM, since these are part of the 25GB active parameters. I'm not sure what f…

Alright, I don't understand anything, but you said ~5secs per token, then for prompts with hundreds to a thousand tokens, we are in the orders of tens of minutes to hours. I would be targetting coding prompts.

Well, it means one day I would have to get into the real thing: the real inference code, and actually run the inference of a small model.

Post reply on HN