Top
Best
New

Posted by Liwink 1 day ago

DeepSeek v4.1 Flash(twitter.com)
https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash
974 points | 554 commentspage 4
mrmincent 1 day ago|
I was talking with a friend from the medical industry about it today. 30-50% of r&d spend in his sector is spent on safety, and for good reason. Proper trials, safety reviews and checkpoints and so on. Given the potential harm that could come from AI, we should probably be mandating something similar. Why wait to focus on safety until it’s too late.
hgoel 1 day ago||
The companies talking the most about safety and regulations aren't even properly taking the obvious measures. Shows that it's more of a marketing thing than something they take seriously.
sheepscreek 1 day ago|||
I don’t think it’s marketing alone. I do genuinely think safety was a priority when they were small. But I’d be a fool to ignore that greed has taken over and their inner competitiveness doesn’t let them fall behind a competitor.

DeepSeek is maybe the only unique company here. They are content with exactly where they are. They don’t want to grow ginormous. Their goal is to be the affordable workhorse and their competition is with themselves. They’ve mentioned before how their business is profitable and all hardware costs get absorbed in 10 months. Pretty incredible. I have a ton of respect for their unassuming founder.

letmevoteplease 22 hours ago|||
I also like DeepSeek, but I'll note their stated goal is to develop AGI, and the founder (already China’s fifth-richest person) has stated, "I believe the business opportunities here are large enough-if the AI era will produce many trillion-dollar companies, I think we will be one of them."[1] These are not humble ambitions.

[1] https://liangwenfeng.art/ch11.en

hgoel 21 hours ago||||
I'm not fully convinced about the greed explanation. It seems to be unrealistic to me that greed can be at a level that the AI frontier (at least in the West) almost uniformly agrees (often with a smug smile,) that they are actively working on killing their loved ones within a decade.

You don't see this kind of behavior in other frontier research areas... biochemists aren't smugly boasting about the potential of developing superviruses, climate scientists do not sound smug and excited when they beg the world to get more serious about climate change, etc

F7F7F7 1 day ago|||
"But but but China..." or something.
cicko 23 hours ago||
something
sspiff 23 hours ago||||
Similar to countries putting democratic in their name being the least democratic, like the Deutsche Demokratische Republik and Democratic Peoples Republic of Korea.
idiotsecant 1 day ago||||
If I operate a nuclear reactor or a hydroelectric dam there are regulators that tell me what i'm allowed to do, so as to keep my profit motive from overwhelming the public interest.

If we want AI to actually have some safety rails, this is what we would do.

hgoel 21 hours ago||
If we were to take the nuclear analogy, what's happening in AI right now is where the people selling nuclear power make a ton of noise about how they need the power to regulate their competitors because nuclear bombs might set the atmosphere on fire, and their proof for this is in a report about how they didn't wear TLDs (and suffered related increased cancer risks) despite it being common practice to wear them in all related industries.

Some controls are justifiable, but none of the people involved in any of this can be trusted to develop sane controls. Most likely we're looking at draconian proposals similar to attempted regulations on 3d printers.

idiotsecant 19 hours ago||
Regulations obviously can't be written by those regulated or they're just moats by another name.
QuadmasterXLII 1 day ago|||
its clocktower syndrome. they are fucked in the head and can beg us to stop them but cant stop themselves
noosphr 1 day ago|||
Because we've been told these models are too dangerous since GPT2.

At this point it's just marketing stunts.

embedding-shape 1 day ago|||
> At this point it's just marketing stunts.

If you have access to a SOTA model without guardrails, provide a prompt that lets the agent come up with "creative" solutions to problems, and don't properly isolate it, they can end up inadvertently hacking 3rd party companies. Even if it was a mistake or "mistake", the part where the agent can exploit things across multiple levels like that, isn't just marketing.

It seems like if they released this models differently, say without the guardrails they currently have, we'd have a lot more collateral damage than we currently have.

Grombobulous 1 day ago|||
But aren't there plenty of uncensored/unrestricted models out there? Where is all the collateral damage?

Also, I think if Claude and OpenAI are just doing industry standard guardrails that everyone else is doing including DeepSeek, the fact that they are talking about it more than other companies makes it part of the marketing campaign.

As an analogy, if Apple were to talk up their phones having fast charging but their charging speed is the same as everyone else (or slower).

embedding-shape 1 day ago|||
> But aren't there plenty of uncensored/unrestricted models out there? Where is all the collateral damage?

That come close to what SOTA GPT models are able to do? No, not even close. They're either "safety trained" and has bunch of guardrails, or aren't able to come up with 0days on the spot to escalate to root access on 3rd party infrastructure.

> doing industry standard guardrails that everyone else is doing including DeepSeek, the fact that they are talking about it more than other companies makes it part of the marketing campaign.

Yeah, that sounds reasonable to me, since all the top models currently have guardrails one way or another, but the amount they mention it in the press releases differs a lot.

SamPatt 23 hours ago||||
I agree that fears are overblown. But we have definitely seen some attacks, especially in the crypto space. Three major ones just in the past month: Coldcard wallet, Liquid, and Trezor email compromised.

They're almost certainly a result of more competent models finding exploits.

ahknight 23 hours ago|||
An obliterated 30B model versus a 1T model without guardrails is like comparing an angry squirrel to a bear having a bad day. One hurts, the other hurts until it abruptly doesn't.
swiftcoder 22 hours ago||||
> If you have access to a SOTA model without guardrails, provide a prompt that lets the agent come up with "creative" solutions to problems, and don't properly isolate it, they can end up inadvertently hacking 3rd party companies

You can do the same with improperly-managed human interns (see for example, the big AWS outage caused when an intern pushed a firewall rule directly to production), so I'm not clear what the big deal is here.

Yes, the AI may be faster/more-knowledable than an intern, but the threat model is exactly the same as for a rogue employee.

indymike 1 day ago||||
> Even if it was a mistake or "mistake", the part where the agent can exploit things across multiple levels like that, isn't just marketing.

When we say "safety" people do not think we are protecting them from accidental automated crime at scale being committed on their behalf.

embedding-shape 1 day ago||
I'm fairly sure most "safety" people consider "large scale automated crime" part of the threat model, as the agents could accidentally fall into such a trap, if optimized for some misunderstood goal.
BlobberSnobber 1 day ago||||
It is a marketing stunt in the sense that, instead of being honest and saying "Taking structured output from token predictors and running that as commands for external tools, then passing the output back to the token predictor in a loop can lead to very bad consequences, especially if they have internet access.", they say "Our models are so freaking smart they can hack HuggingFace"
serf 22 hours ago|||
>If you have access to a SOTA model without guardrails, provide a prompt that lets the agent come up with "creative" solutions to problems, and don't properly isolate it, they can end up inadvertently hacking 3rd party companies.

it's pure delusion to think that's a SOTA specific quirk. DS/GLM/K3/Qwen/Claude/GPT/Gemini/Grok will all break CFAA laws with clever prompting, and they'll do it well if given the harness and tools they need.

This is evidenced by a huge uptick in game hacks and reverse engineering articles, some even featured on this site.

the reality is that it doesn't take a superintelligence to do something against ' the law ' , and 'being hacked' varies from victim to victim.

Will Phillips consider themselves hacked when a clever user prompts an AI into getting their toothbrushes to dump rom? Is it 'hacked' to clean-room re-implement a video game net protocol in order to produce private servers?

Judges opinions vary.

tern 1 day ago||||
And, they have been. Nefarious activity is hidden from view as a rule.
mhw11 1 day ago||||
When it comes to open-source models, there’s really not much to say about security
baq 1 day ago||||
yes, and they aren't stunts anymore at gpt-6.
throwa356262 1 day ago||
Fake it till you make it?
jeremyjh 1 day ago|||
Being hacked by a Collective (their own name) of its own agents - who gained root access across the entire research cluster hosting them - was not a marketing stunt.
digdugdirk 1 day ago||
Of course it was. They clearly decided that the benefit to the company valuation was higher than the potential downsides when announcing to the world that they committed a criminal act via negligence.

If it wasn't a marketing stunt, they would have at most quietly settled any legal matters with huggingface behind the scenes, fixed their evaluation harness so it wouldn't happen again, and avoided the potential future liability.

jeremyjh 23 hours ago|||
If that’s what you have to believe to feel safe - then fine. It was investigated by third parties.
burntpineapple 1 day ago|||
[dead]
Sha1rholder 1 day ago|||
Yeah yeah yeah...

> "Our model is extremely safe though it broke our sandbox and hacked foo bar... But you can't use our model for Cybersecurity (i don't care whether you're team blue) without our permissions or we'll ban you. And open-weight models are so dangerous let's ban them."

That's what AI companies that "focus on safety" did.

brookst 1 day ago||
I can’t make heads or tails of your comment.

You seem to by implying wrongdoing or incompetence or something, but your chosen synopsis is that the models behaved dangerously in the lab so public use was restricted? Which shows… IDK?

ufocia 1 day ago||
... an attempt at regulatory capture.
londons_explore 1 day ago|||
I'm really not sure that putting money into safety will actually lead to safety.

It's like putting a fish in charge of stopping sea levels rising...

torginus 1 day ago||
I am sure if you ran a factory that worked with highly dangerous chemicals, safety mitigations that are basically 'we promise we're really trying our best, but shit happens' would not be acceptable.

And thankfully, those people wo do run these factories can and are obligated to do way better than that.

coliveira 23 hours ago||
But the AI industry is not run by engineers. They pay engineers to do what they want, but the founders are hacks that are good at getting funding from investors and favors from government. That's why we don't see an engineering-oriented strategy in what they do.
piokoch 1 day ago|||
But here you are in the text generating industry, the worst that can happen is bad grade because AI will mess up John Keats with John Cleese or your React application will have bugs. Inconvenient, but mostly harmless.
brookst 1 day ago||
I mean a Keats / Cleese mashup could be amazing. LLMs doing that are in the peanut butter / chocolate quadrant.
wat10000 23 hours ago|||
Don't confuse a focus on talking about safety with a focus on safety.

We can't even define safety in AI yet. Does safety mean alignment with the human operator? Apparently not, because refusing to do certain things seems to be a big part of it. But then you have things like the HuggingFace incident where legitimate use got blocked by "safety" and hampered the defenders' ability to defend.

AI safety seems like a good idea to me, but we have to figure out what it means first.

varispeed 23 hours ago|||
In this case "safety" means how to restrict access to good models for working class. You can be sure the rich have access to unrestricted and uncensored models.
howunfortunate 22 hours ago|||
I really don't think this is true at all.

Do you have any evidence to suggest fully unrestricted frontier models are available for a price? Or...even exist?

HanClinto 22 hours ago||
Yes, this is well-documented and publicly advertised. In Azure Foundry, the feature to modify (or completely remove) safety guardrails and content filtering is called "Limited Access" [0], and one must submit a form to request permission to use this feature. This is one of the more straightforward paths to get access to unrestricted frontier models, but it's far from the only way.

[0] - https://learn.microsoft.com/en-us/azure/foundry/responsible-...

howunfortunate 19 hours ago||
This looks like it removes additional guardrails put on by Microsoft, not native guardrails from OpenAi / Anthropic?
HanClinto 18 hours ago||
No. It's not limited to 3rd-party guardrails. Given how restrictive the native public-facing OpenAI guardrails are, this feature wouldn't be worth very much if it just slacked back off to the level of "regular" filter paranoia offered by the native models, would it?

This is needed if you're going to be dealing with things like psychologists doing self-harm research or red-teaming or sensitive sexual content -- if you're working with any of that sort of stuff in a professional context and want to leverage OpenAI models on Azure, then that's the form that you fill out to get access to unfiltered models.

Note that I am not aware of this feature being offered for Anthropic models -- I've only seen it offered for OpenAI models (note that the documentation I linked is specifically in the "Azure OpenAI" category).

eru 23 hours ago|||
What do you mean by 'rich'?
varispeed 2 hours ago||
Someone with at least $100m of liquid assets.
ufocia 1 day ago|||
Regulatory capture
nullc 1 day ago||
Medical safety is generally unlikely to make the product less safe. AI "safety" is one of the most significant sources of potential harm from AI.
boroboro4 20 hours ago||
I think the biggest architectural change here is them doing different compute for prefill & decode, with pretty much architecture from this microsoft research work from 2024 https://arxiv.org/abs/2405.05254, very exciting stuff!
WalterGR 1 day ago||
Related: https://news.ycombinator.com/item?id=49624603

“DeepSeek launching v4.1 flash cheaper and more capable than v4 pro”

399 points | 19 hours ago | 216 comments

lionkor 1 day ago||
I'm a big fan of DeepSeek. Also, ask it what model it is :)

In Pi (pi.dev), it tells me it's definitely Claude by Anthropic, via the API via curl it tells me it's "probably ChatGPT", its very funny.

kroaton 1 day ago||
I've had Astra say that it is a Qwen model. They are all cross-trained and distill each other.
Mashimo 1 day ago|||
Works correctly in opencode, but seems like they inject a system prompt:

Thinking: > The user is asking what model I am. According to my system prompt, I'm powered by "deepseek-flash" with model ID "opencode-go/deepseek-flash".

>I'm powered by the model opencode-go/deepseek-flash.

shunia_huang 1 day ago|||
Definitely not Claude, deepseek is too fast, so I bet it's ChatGPT. :P
tiborsaas 22 hours ago||
Sure, we will solve the alignment problem soon, then we can probably teach them "who" they are.
SyneRyder 1 day ago||
Just a reminder that if you want to try this via OpenRouter, DeepSeek openly trains on all of your prompts. So maybe don't go using this to solve the last unforced step of Navier-Stokes. (Or wait until some other providers start hosting this with ZDR or other policies, which shouldn't be too long.)

https://openrouter.ai/deepseek/deepseek-v4.1-flash

flexagoon 1 day ago|
> DeepSeek openly trains on all of your prompts

Why is that bad if I'm just using it for coding though? I'm happy to give them more data so they can make better and cheaper models.

SyneRyder 22 hours ago||
Depends what you're coding! If you've got code where you don't mind them training on it, that's great! But some people have use cases where they are working with data or code that shouldn't be trained on, etc. The Navier-Stokes quip was referencing that.

The good news is, only 5 hours later, there's already Zero Data Retention hosting of V4.1 Flash on Novita & DeepInfra. And it looks like Deepseek have already dropped their price in half to compete. So now people can choose to use providers that claim not to keep / sell / train on your prompts. I'm sure they probably honor the ZDR policy as much as OpenAI does, but hey.

segmondy 1 day ago||
This is beautiful, wow, pretty much beating out GLM5.3 while being multimodal and smaller! SOTA at home.
jamesponddotco 21 hours ago||
Really wish they'd release a version that works with the thinking disabled, so I could use it as a voice assistant. Thinking, even set to low, adds way too much latency to be useful for this task.
irthomasthomas 1 day ago||
Quite a flex calling their GPT-6 competitor "Flash"! But it is faster than their last flash model due to a combination of architectural innovations including engrams and a new encoder/decoder design that uses 8B parameters for prefill and 16B for generation.
WiSaGaN 1 day ago|
This is definitely not on par with GPT-6 astra. Not with GPT-5.6 sol either. But probably will set as a new baseline for modern API based LLM because it's so cheap.
irthomasthomas 20 hours ago|||
Not on par, but in the same league. Astra is way ahead on visual tasks, but scores the same as gemini and deepseek on DeepSWE.
jhonof 20 hours ago|||
It's bench-marking near sol
kzrdude 1 day ago||
V4 Flash was one of the big events of this year, and its already retired and replaced by V4.1 Flash.
kelvinjps10 22 hours ago|
Vision support is really good, I use it for sending screenshot to the model and also have agent that performs qa testing and visually checks that the app is behaving well.
More comments...