Chinese AI Startup DeepSeek Stuns Tech Industry, Causes Stock Panic

[Free]

I accidentally bumped into the NOT ALLOWED filters of DeepSeek's chatbot.

I asked it to tell me a short story about Free Xue. (It's a thing I do when performing AI prompt evaluation...) When I made a request of ChatGPT for this, it generated a 300 word tale about a digital artist in a future metropolis called NeoCity who refuses to compromise her work, and challenges the status quo. It was trite Mary Sue slop, but cute enough.

DeepSeek, however, started outputting several paragraphs on the current legal situation of a real individual, I believe this person. DeepSeek's chatbot appears to have a retroactive banning feature, as the "story" I saw written out was - before I could copy and paste it - wiped from the chatbot and replaced with "Sorry, that's beyond my current scope. Let’s talk about something else."

If I suddenly vanish from the internet, consider this a potential reason...

 

[Noodles]

Perplexity understands spelling and counting like a proper first grader.

The word "Raspberry" contains 3 instances of the letter "R". Breaking it down:

• First letter: R
• Seventh letter: R
• Eighth letter: R

The full spelling is R-A-S-P-B-E-R-R-Y, with the "R" appearing at the beginning and twice consecutively near the end.

 

[Bartholomew Gallacher]

Deepseek's SaaS just had a major data leak.

Wiz Research Uncovers Exposed DeepSeek Database Leaking Sensitive Information, Including Chat History | Wiz Blog

A publicly accessible database belonging to DeepSeek allowed full control over database operations, including the ability to access internal data. The exposure includes over a million lines of log streams with highly sensitive information.

www.wiz.io

 

[Caete]

 

[Caete]

I accidentally bumped into the NOT ALLOWED filters of DeepSeek's chatbot.

I asked it to tell me a short story about Free Xue. (It's a thing I do when performing AI prompt evaluation...) When I made a request of ChatGPT for this, it generated a 300 word tale about a digital artist in a future metropolis called NeoCity who refuses to compromise her work, and challenges the status quo. It was trite Mary Sue slop, but cute enough.

DeepSeek, however, started outputting several paragraphs on the current legal situation of a real individual, I believe this person. DeepSeek's chatbot appears to have a retroactive banning feature, as the "story" I saw written out was - before I could copy and paste it - wiped from the chatbot and replaced with "Sorry, that's beyond my current scope. Let’s talk about something else."

If I suddenly vanish from the internet, consider this a potential reason...

Yeah, the "sorry that's beyond my current scope" is what it shows anytime it self censors. Seems like they are having issues with the whole thing due to the influx of users just like every other start up has since forever... Maybe they can use their scrape for cash to buy more slots on a SAE data farm.

 

[Argent Stonecutter]

tl;dw?

 

[Casey Pelous]

tl;dw?

Maybe ask DeepSeek for a summary.

 

[Kamilah Hauptmann]

Maybe ask DeepSeek for a summary.

When asking Copilot what part of a particular request was forbidden it wouldn't even tell me that. I would have to start experimenting by changing one part at a time until I found the problem area.

 

[Bartholomew Gallacher]

Tom's Hardware is running a story, that the $6 mio. training cost for Deepseek are misleading, because the company behind it invested $1.6bn into server buildouts and has 50.000 Hopper GPUs at its disposal.

So while the its right, getting there took a lot of more work and cost. This does not change the fact though that Deepseek did this all with domestic hires only, and they've improved the method on a major level with Multi-Head Latent Attention and other innovations.

So no good news for Nvidia, because it still could translate into wide spread usage will cause less GPU demand.

https://www.tomshardware.com/tech-industry/artificial-intelligence/deepseek-might-not-be-as-disruptive-as-claimed-firm-reportedly-has-50-000-nvidia-gpus-and-spent-usd1-6-billion-on-buildouts

 

[Free]

DeepSeek goes beyond “open weights” AI with plans for source code release

Chinese AI firm says daily releases will reveal "code that moved our tiny moonshot forward."

arstechnica.com

Last month, DeepSeek turned the AI world on its head with the release of a new, competitive simulated reasoning model that was free to download and use under an MIT license. Now, the company is preparing to make the underlying code behind that model more accessible, promising to release five open source repos starting next week.

In a social media post late Thursday, DeepSeek said the daily releases it is planning for its "Open Source Week" would provide visibility into "these humble building blocks in our online service [that] have been documented, deployed and battle-tested in production. As part of the open-source community, we believe that every line shared becomes collective momentum that accelerates the journey."

While DeepSeek has been very non-specific about just what kind of code it will be sharing, an accompanying GitHub page for "DeepSeek Open Infra" promises the coming releases will cover "code that moved our tiny moonshot forward" and share "our small-but-sincere progress with full transparency." The page also refers back to a 2024 paper detailing DeepSeek's training architecture and software stack.