DeepSeek limits new accounts amid cyberattack
- Reference: 1737998028
- News link: https://www.theregister.co.uk/2025/01/27/deepseek_suspends_new_registrations_amid/
- Source link:
"Due to large-scale malicious attacks on DeepSeek's services, we are temporarily limiting registrations to ensure continued service," the biz said in [1]a note on its status page. "Existing users can log in as usual. Thanks for your understanding and support."
The incident appears to have begun around 2133 CST on Monday, January 27, or around 0733 PST and was ongoing at the time this article was filed. Sign-ups and logins via Google's single-sign-on appeared to be working at 1030 PST.
[2]
DeekSeek did not immediately respond to a request for comment.
[3]
[4]
The outfit's AI app for iOS, DeepSeek – AI Assistant, is presently the top free download in Apple's US App store, just above OpenAI's ChatGPT.
The Chinese AI firm [5]released DeepSeek R1 as an open source LLM last week, following its parent V3 model, claiming reasoning capabilities that rival OpenAI's GPT-o1 in a number of benchmarks.
[6]
Both can be used [7]locally or via [8]the web or DeepSeek's apps for free, though as stated above, signups for the web chatbot interface at least have been limited.
[9]China's DeepSeek just dropped a free challenger to OpenAI's o1 – here's how to use it on your PC
[10]What happens when we can't just build bigger AI datacenters anymore?
[11]AI agents? Yes, let's automate all sorts of things that don't actually need it
Having allegedly built V3 using just [12]$5.58 million [PDF], significantly less than Western AI firms, investors in companies like Nvidia have begun to wonder whether they need to revise their financial assumptions. The result has been [13]a selloff in AI stocks.
The viability of open source models has long been [14]a concern among commercial AI firms with proprietary models. AI skeptics such as Gary Marcus have previously [15]questioned the valuation of firms like OpenAI when corporations like Meta have been giving away open source models, namely Llama, at no charge.
We will update this story as it develops. ®
Updated to add at 1825 UTC
As noted above, you may be able to login at least with a Google account by now, or you may be able to get through with a regular account. Also, DeepSeek has released a family of MIT-licensed open-source models called [16]Janus Pro that it claims can beat OpenAI's DALL-E 3 image generator in benchmarks.
Get our [17]Tech Resources
[1] https://status.deepseek.com/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/cybercrime&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z5gQDReb0I4Tip_FruAKHwAAABc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/cybercrime&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z5gQDReb0I4Tip_FruAKHwAAABc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/cybercrime&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z5gQDReb0I4Tip_FruAKHwAAABc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.reddit.com/r/LocalLLaMA/comments/1hmn55p/deepseekv3_officially_released/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/cybercrime&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z5gQDReb0I4Tip_FruAKHwAAABc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2025/01/26/deepseek_r1_ai_cot/
[8] https://chat.deepseek.com/
[9] https://www.theregister.com/2025/01/26/deepseek_r1_ai_cot/
[10] https://www.theregister.com/2025/01/24/build_bigger_ai_datacenters/
[11] https://www.theregister.com/2025/01/27/ai_agents_automate_argument/
[12] https://github.com/deepseek-ai/DeepSeek-V3/blob/main/DeepSeek_V3.pdf
[13] https://www.theregister.com/2025/01/27/tech_stocks_tank_as_us/
[14] https://www.theregister.com/2023/06/22/small_custom_ai_models/
[15] https://www.theregister.com/2024/09/26/openai_execs_leave/
[16] https://huggingface.co/organizations/deepseek-ai/activity/models
[17] https://whitepapers.theregister.com/
The result has been a selloff in AI stocks.
And expect a US ban for 'national security reasons' in 3...2....1
Re: The result has been a selloff in AI stocks.
Currently, ruspublican media is pushing the "we told you! You have to give us the money for an AI state or china will do it first!" scare tactic.
Every rag, front page.
Re: The result has been a selloff in AI stocks.
I know, they can't be reasoned with.. but regardless. The lesson should be: To combat random Chinese start-up eclipsing us with $6mil, give every AI graduate $60,000 rather than giving OpenAI $6 billion.
It's not actually that surprising in a young field. We know the key to training neural nets is a) the training regime and b) the design of the network. OpenAI found a reasonable combination of both and then immediately threw unimaginably stupid amounts of money at both to boost them a little bit, which is what you don't do when some more innovation is likely around the corner (and you don't have a fool-proof way to get proportionate returns very rapidly). Innovation 1 - Hubris 0, not for the first time.
Nvidia & the cloud businesses will be relatively okay, because the new innovation will always be increment-able with more processing power.
Re: The result has been a selloff in AI stocks.
The lesson should be: To combat random Chinese start-up eclipsing us with $6mil, give every AI graduate $60,000 . . .
, , , on the condition that they find honest work, like selling used cars or enrolling in law school.
Re: The result has been a selloff in AI stocks.
Ruspublican? If Russia is controlling the Republican party, why are they upset that China's got a great AI model? You do realize that China and Russia are pretty good allies, right? BRICS, much??
I think you need to get your conspiracy theories straight. TDS is truly a terrible affliction!
"National security reasons"? Perhaps Trump-Muskian reasons.
I'm pretty sure those confabs at the Merde-Largo strip mall with the big tech bros included discussions of how the US chief honcho could benefit from the AI tech craze. Now those damn Chinese have thrown rice in the gears.
Re: The result has been a selloff in AI stocks.
Chortle. OpenAI valued at $150B. It’s going to make HP’s writing down of Autonomy by $8B look like chump change,
Given the discussion about bit barns and their power consumption, it's good to see at least one AI thing being able to run well on lower powered hardware.
Such timing...
Smoke Bomb!
It'll be those alleged Chinese state sponsored hacking groups the Reg is sooo fond of shoving in our faces, morning, noon and night.
Hurricane Pepper or something.
Oh wait.
Can we be completely sure the LLMs haven't started fighting each other?
After all, we know ChatGPT has been granted DDoS capabilities: https://www.theregister.com/2025/01/19/openais_chatgpt_crawler_vulnerability/
Well, that didn't take long.
"Also, DeepSeek has released a family of MIT-licensed open-source models called Janus Pro that it claims can beat OpenAI's DALL-E 3 image generator in benchmarks."
Stop. Please have mercy. OpenAI is already dead.
It's come back from the brink before. There's no harm in making perfectly sure.
What a surprise
I wonder who could possibly be behind for such an attack. I doubt it'd be some script-kiddy in a darkened room.
LOL
I can hardly wait to see people asking it questions about best for of governments and all it says is Xi should be Emperor of the world. All mutants (not having black hair) must be eliminated to save the human race.
Give it a week.
Damn It!
New cheap kick ass service. I wanted to sign up to have a play for a project and it turns out the yanks have taken it offline because they have hurty feels.
Uhh, guys...
from the article:
The incident appears to have begun around 2133 CST on Monday, January 27, or around 0733 PST and was ongoing at the time this article was filed. Sign-ups and logins via Google's single-sign-on appeared to be working at 1030 PST.
Living in CST, I can assure you that 2133 CST is not 0733 PST. In fact I'm posting this at 3:36p CST (or 1536 CST). Which of course means that the DDoS hasn't happened yet, nor will it for another roughly 6 hours.
Hey, can you post the winning PowerBall numbers for tonight? Thanks!
And don't give me trouble about not sending this to the corrections mailbox, because there doesn't appear to be on posted in this article.
Putting AI in its place
By first using human brain power rather than money to work out how to do it. I'm shocked!
It's just let me sign up
Earlier, it wouldn't let me sign up using Google, and wouldn't send a code to another email account. Once in, it correctly answered a couple of the classic questions that stumped the previous generation of LLMs.