- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
Summary
Experts are warning against sharing sensitive data with DeepSeek, a Chinese AI app that has become the most downloaded free app in the US and UK.
Concerns include potential Chinese government data access, alignment with Chinese state narratives (e.g., censoring topics like Tiananmen Square), and user data exploitation.
DeepSeek stores user data on servers in China, governed by laws requiring compliance with national intelligence efforts.
While praised for its affordability and innovation, critics urge caution, citing risks of surveillance, disinformation, and privacy breaches.
You must log in or register to comment.
The article mentions in passing that DeepSeek is open source, but completely fails to mention why that’s important and makes regional censorship irrelevant. The model is perfectly able to answer your questions about Tank Man or Xinjiang as long as you run it on a server outside China. https://github.com/deepseek-ai/DeepSeek-V3
The open model is still censored when you run it privately. People are retraining it not to be, but the official one is censored.
And even methods like ablation don’t seem to circumvent since of the censorship. Still, in time we will see less restricted DeepSeek-R1 derivatives.
Fair, but most people are going to rush to a website because they are either ignorant or lazy, so it makes sense to warn them about how that data will be used.
Absolutely, but they should still elaborate on why it being open source is so important, especially when the industry leader is not.
Who is the industry leader right now? Honest question; my job has a few commercial ones and in my personal use, I mostly use open source ones offline already via Ollama.
Easily OpenAI, if only because their API is so widely used.
Let me guess the expert… sam altman?
Lmao
Shut up
Only share sensitive information with US companies please. 😉
You can literally run it offline, unlike chatgpt.
You have 16 NVidia H100’s?
Is that needed to run it or to train it?
Running. Most of these models require an absurd amount of VRAM compared to your standard desktop. You can run models on desktop hardware but they’re not as good (see the “distill” versions).
This is just not true. https://github.com/deepseek-ai/DeepSeek-V3?tab=readme-ov-file#6-how-to-run-locally
DeepSeeek-V3 is not the same as DeepSeek-R1. V3 is a distilled version.
Yeah, no shit.
