Skip to content

DeepSeek Is So 1984

Perhaps the most important thing to remember that everything that comes out of China has a hidden, secondary and nefarious purpose.

Firstly, I work a lot with AI – and I mean a lot. In my opinion OpenAI is by far the best. Gemini is OK and as for DeepSeek…let’s just say I wouldn’t trust it to babysit my Mac.

Secondly, whatever you do, don’t download the DeepSeek app. If you really want to, go the website and, if you want to be really careful, use a throwaway email when you sign up.

So what makes DeepSeek so bad?

Well, if you want to find out all about Chinese censorship, DeepSeek’s got your back. Try asking it about Tibet, Dalai Lama, Tiananmen Square or even ‘Tank Man’ and see what happens. It actually starts giving you the answer and then deletes it and replaces the text with a generic “Sorry, I’m not sure how to approach this type of question yet.” It’s almost as if there’s a Chinese worker at a computer screen anxiously watching for when to press the delete button (which isn’t that farfetched.)

Then there’s the fact that DeepSeek used output from OpenAI’s models to train its own models: a method know as “distillation”.

As for ‘But OpenAI done stole data too!’ please allow me to explain.

OpenAI hasn’t been accused of stealing data. They’ve been accused of using copyrighted data without permission. This involves things like what’s meant by ‘fair use’ and whether additional licensing is required, etc.

Imagine one of those cooking shows where you have to use only raw ingredients: it’s the difference between being caught using packaged sauce in your potato bake and straight up stealing someone else’s potato bake and claiming you baked it.

What about claims that DeepSeek is faster, better and meaner? Yeah, sure: proponents say DeepSeek is better at reasoning, as in math and logic. But try asking DeepSeek this: 

Three gods A, B, and C are called, in no particular order, True, False and Random. True always speaks truly, False always speaks falsely but whether Random speaks truly or falsely is a completely random matter. Your task is to determine the identities of A, B and C by asking three yes-no question; each question must be put to exactly one god. The gods understand English, but will answer all questions in their own language, in which the words for yes and no are da and ja, in some order. You do not know which word means which.

Don’t worry, I’ll wait.

In any case, this kind of reasoning is only a small fraction of what AI can do. If all your AI model can do is add one plus one better, then you really haven’t got much.

What about claims that DeepSeek uses breakthrough tech: its code written by gods who descended from heaven with every line a work of pure genius?

Standard AI models tokenise one word at time while DeepSeek tokenises groups of words, making it faster. While this is plausible, it begs the question: why haven’t the major AI players switched to it? (DeepSeek claims it uses a custom multi-token prediction, which I won’t get into here.)

But perhaps the most important thing to remember that everything that comes out of China has a hidden, secondary and nefarious purpose.

DeepSeek knows it’s BS and it also knows it’s going to be caught out on its BS. Its purpose is actually to disrupt in a negative way and cause economic and financial harm by making bogus claims and causing panic, which is what it has done.

Latest