Chinese AI Startup DeepSeek Overtakes ChatGPT on Apple App Store | Hacker News Reader

Chinese AI Startup DeepSeek Overtakes ChatGPT on Apple App Store

reuters.com

23 pointsnabla914 comments

The title is: "DeepSeek hit by outages as users flock to Chinese AI startup"

Reuters changed the original title.

You can see the original in the url "/chinese-ai-startup-deepseek-overtakes-chatgpt-apple-app-store-2025-01-27/"

News sites try multiple headlines and settle for the one that gets most clicks.

Everyone is amazed by the benchmarks, but I've yet to encounter journalists that cover how much R1 and V3 are propaganda machines.

They deny the tiananmen massacre, the abuses on uyghurs, and defend the CCP. LLMs are replacing Google and becoming the source of truth for an entire generation. This is extremly dangerous and medias have the duty to cover that.

> They deny the tiananmen massacre, the abuses on uyghurs, and defend the CCP. LLMs are replacing Google and becoming the source of truth for an entire generation. This is extremly dangerous and medias have the duty to cover that.

doesn't matter for business use-cases (except for search engines or knowledge base usecases).

Would you like a CCP sleeper agent to work in your company?

If I allow NSA why not CCP?

I tried with Kagis LLM assistant.

Without web search:

Chat V3 discusses the events: violent government crackdown, human rights violations, censoring.

R1 evades answering, stating the topic is sensitive in China (simlar how it reacts to American sensitive topics)

With web search enabled, R1 discusses the found sources (which are not censored, and the model does not add any censoring as far as I can tell, certainly not enough to respect the Chinese governments stance.)

Yes they censorship seems to be done at the RLHF level and is easy to evade when you know what you're looking for. Refusals happen around 80% of the time for me.

Interestingly, even the 8B version, distilled from Llama 3 repeats CCP propaganda just fine.

I've never asked an AI about "the tiananmen massacre, the abuses on Uyghurs" and no AI will convice me about communism. And what isn't propaganda today? It's not that Google Gemini is any better with black nazis and whatever.

That's because you're an adult with established opinions, educated with methods of the previous century. Millions of teenagers right now are using LLMs as their tutor every day.

Also I'm interested if you can find Gemini actively lying on something (not image representation bias, which was immediately corrected btw)

If that's your only concern, the I'd say the model is a huge success.

Yes, it's a huge success for China. Many of my friends are using LLMs more than the internet now, the CCP can rewrite history and push their totalitarian propaganda easily.

The silver lining is that the R1 paper is the most detailed LLM sota paper since 2019, and western labs will reproduce it soon.

DeepSeek open sourced the model

HuggingFace is currently replicating DeepSeek, and others will follow. That will remove all CCP censorship in the model. DeepSeek has huge impact because it takes only $6 million to train from scratch.

Yes I'm quite happy they gave so many details in the paper, it's the first sota paper with actual science in it!

This thread is about americans installing the DeepSeek app though, which is quite worrisome.

Reply on news.ycombinator.com