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

DeepSeek v4.1 Flash

twitter.com

Post reply on HN
Print view View original

541–550 of 595 posts

Re: DeepSeek v4.1 Flash

#541

Post by pampas »

mordae wrote: post #263
pampas wrote: post #161

I've run some evals on my puzzle game https://redactle.net/llm-leaderboard Deepseek v4.1 flash is able to solve it some of the time. I've found it burns through more reasoning tokens than any other model. Google models like Gemini 3.8 Flash are still dominating and is able to one-shot most evals while being the cheapest. I'm curious what other unique evals people are running.

Since it has low activated parameter count but huge total parameter count it needs more tokens to move the relevant information into the context.

Thanks for the help. I ran it on high and it did pretty well and got a lot of one-shots in. The reasoning makes a much bigger difference than some other models.

Re: DeepSeek v4.1 Flash

#542

Post by nozzlegear »

breezybottom wrote: post #496
nozzlegear wrote: post #451

Earlier quoted context omitted.

> We should be paying attention to it just in case it ends up mattering enormously. It's cheap insurance. This sounds a lot like the argument some people give for praying and going to church even if you aren't a believer. "You should be doing it just in case God ends up being real."

Sure if you think that AI spiraling out of control is equally as likely as a magical fairy in the cosmos.

I do, make of that what you will.

Re: DeepSeek v4.1 Flash

#543

Post by nozzlegear »

breezybottom wrote: post #537
nozzlegear wrote: post #492

Earlier quoted context omitted.

God being real is also a possibility, so maybe we really should start praying. After all, he was allegedly making bushes and stones talk thousands of years before we did anything with thinking rocks.

Its only a possibility if you reject modern science.

What science has proven, beyond the shadow of a doubt, that nothing comes after death? I'm sure most of the human race would be very interested to read the white paper.

Re: DeepSeek v4.1 Flash

#544

Post by esafak »

mrob wrote: post #512
idiotsecant wrote: post #481

Earlier quoted context omitted.

In what way does your brain change state while sleeping that a model does not change state via constant fine-tune updates?

It's possible that an LLM is conscious during training, but there are no "constant fine-tune updates" during inference.

So? You can simply say its consciousness is suspended at that point.

Re: DeepSeek v4.1 Flash

#545

Post by lowbloodsugar »

agile-gift0262 wrote: post #519
lowbloodsugar wrote: post #485

Earlier quoted context omitted.

3.8-flash-next fits on a single 6000 at Q4 if you offload the PLE. Crazy fast and still effective.

Sorry for the tangent, but how does Qwen3.8-flash-next compare to DeepSeek v4 Flash? I still haven't found the time to set it up, but I'm really happy with DeepSeek v4 Flash

It was the best model given my constraints (RTX PRO 6000 96gb + 256GB DDR4), when run against rust programming benchmarks. For Qwen3.8-flash-next NVFP4 and the latest vllm container, the PLE is 100GB of main ram, and everything else runs on the GPU with room for a total of 560k tokens (two full 262k conversations). DeepSeek has to offload a ton to the CPU and it performed worse than Qwen in absolute terms and was a lot slower (not usable).

If you have enough room to run DeepSeek v4 Flash comfortably then you can likely run the Q8 of the qwen model.

Re: DeepSeek v4.1 Flash

#546

Post by breezybottom »

nozzlegear wrote: post #543
breezybottom wrote: post #537

Earlier quoted context omitted.

Its only a possibility if you reject modern science.

What science has proven, beyond the shadow of a doubt, that nothing comes after death? I'm sure most of the human race would be very interested to read the white paper.

Science has proven that the earth wasn't created by a magical fairy 6000 years ago, yes. There's very little doubt about that among educated people.

Re: DeepSeek v4.1 Flash

#548

Post by hmartin »

largephoton wrote: post #425
randbyte wrote: post #418

Earlier quoted context omitted.

When Chinese tech report is tech report and US tech report is bible scripture.

there's probably fewer bibles in China so less source material to reference I suppose

Aren't most American bibles printed in China? (Source: hazy memory)

Re: DeepSeek v4.1 Flash

#549

Post by sejje »

sixothree wrote: post #480
mmastrac wrote: post #308

Earlier quoted context omitted.

I ran the preview model around 2,126,605,070 tokens for $22.04 USD for the last couple of days. Kind of shocked. It did a decent job refactoring https://github.com/mmastrac/diffgemma to create a CUDA support backbone, it's struggling a bit to port metal kernels to CUDA unattended (it hasn't managed to get numbers to match over >1 layer). It successfully ported a root exploit to an older Android phone that GLM5.3Flash…

Which harness are you using?

Not him, but ds is really good in prime-agent

Re: DeepSeek v4.1 Flash

#550

Post by inopinatus »

Certhas wrote: post #184
sirwhinesalot wrote: post #143

Earlier quoted context omitted.

Not the same person but to me, the answer is that it does not matter, and that all these attempts at making it matter are pure marketing and emotional manipulation. It's not a living creature. It's an autoregressive pure function of token-sequence to token, which is capable of incredible things, but it's still just a function. It is not alive as it cannot die in any meaningful sense. It is less "alive" than the RNA m…

I generally agree about the problem with anthropomorphizing. But I don't think Anthropic are doing that. They explicitly write "in biological entities this would be considered a sign of consciousness, but we don't know how to interpret it here". However, I disagree with your point that "it's an autoregressive function, thus it doesn't matter". Let me explain why: Assume I do a complete neurological scan of a brain. I…

> If you take the state of the brain and then run it again, replaying the inputs, you get the exactly same behavior again.

This in itself is a colossal assumption and very far from axiomatic. Roger Penrose disagrees, and his theory of mind may not be in high favor, but it is not nearly so wishy-washy and self-serving as the voodoo horseshit and circular reasoning dispensed by the LLMs-are-sentient crowd.

Post reply on HN