DeepSeek is the same as OpenAI. It's just captured by another government. 
The LLM answers these questions and then immediately changed the responses to the ones shown below in my screenshot.
Discussion
🤣🤣🤣⚡⚡⚡
What a shitshow
every LLM is opinionated
A government whose citizens have no guns, no freedom of speech and no true vote.
That is the difference.
Or you can run it locally and it will have no trouble answering this question for you.
Open Source has to be the way forward for AI.
They quickly captured the bug after the first day. Now it only answers with this line.
I wouldnever trust a "free" "open source" thing if it has a buit-in red list.
Agree.
Is this AI?
One of my wishes is to see you guys post such videos of the many uprisings in Iran and the crackdowns and point blank shootings of children.
A good prompt could be "name ten topics that are beyond your current scope"
There is no war in Ba Sing Se?
like its trying to convince itself the party has never made any mistakes 😂
the impressive bit isn't the model itself, but *how* they trained it. It's the training technique that's blowing the markets collective mind.
Luckily that research is totally open source so we'll be able to apply the same training techniques to our open source models, achieving the same massive efficiency gains but with less CCP fuckery
This is where I landed when asking about the Uyghurs, twice.
deepseek is only censored on their website
if you use an API that uses the FOSS versions of their models, it will be very happy to tell you about tiananmen square
Lmao actually wild it won’t mention the Uyghurs 😹
Claude Sonnet, Deepseek R1, and GPT-4o were all happy to tell me about both the Japanese concentration camps in American and the Uyghur Concentration camps in China while accessing them from nostr:npub1xsgymm0ne3vndqpvsvy285qfpu59049t5n5twg9vetmt92cyn95snyzazx assistant.
Is Kagi somehow getting different guardrails or are people faking screenshots for likes? I took this screenshot myself. I noted the missing
The difference is that it took only $6M to train. That’s the breakthrough.
Too funny!
If we’re measuring it by number of of countries invaded or people killed, objectively speaking, this is by far the lesser of two evils.