DeepSeek thought for 19 seconds earlier than answering the query, “Are you smarter than Gemini?” Then, it delivered a whopper: DeepSeek thought it was ChatGPT.
This seemingly innocuous mistake might be proof – a smoking gun per se – that, sure, DeepSeek was educated on OpenAI fashions, as has been claimed by OpenAI, and that when pushed, it’s going to dive again into that coaching to talk its reality.
Nonetheless, when requested level clean by one other TechRadar editor, “Are you ChatGPT?” it stated it was not and that it’s “DeepSeek-V3, an AI assistant created completely by the Chinese language Firm DeepSeek.”
Okay, certain, however in your relatively prolonged response to me, you, DeepSeek, made a number of references to your self as ChatGPT. I’ve included some screenshots beneath as proof:
As you possibly can see, after attempting to discern if I used to be speaking about Gemini AI or another Gemini, DeepSeek replies, “If it is concerning the AI, then the query is evaluating me (which is ChatGPT) to Gemini.” Later, it refers to “Myself (ChatGPT).”
Why would DeepSeek try this beneath any circumstances? Is it a type of AI hallucinations we like to speak about? Maybe, however in my interplay, DeepSeek appeared fairly clear about its identification.
I bought to this line of inquiry, by the best way, as a result of I requested Gemini on my Samsung Galaxy S25 Extremely if it is smarter than DeepSeek. The response was shockingly diplomatic, and after I requested for a easy sure or no reply, it informed me, “It isn’t potential to offer a easy sure or no reply. ‘Sensible’ is simply too advanced an idea to use in that strategy to language fashions. They’ve totally different strengths and weaknesses.”
I can not say I disagree. In reality, DeepSeek’s reply was fairly related, besides it was not essentially speaking about itself.
This does not add up
I believe I have been clear about my DeepSeek skepticism. Everybody says it is probably the most highly effective and cheaply educated AI ever (everybody besides Alibaba), however I do not know if that is true. To be truthful, there is a large quantity of element on GitHub about DeekSeek’s open-source LLMs. They at the very least seem to indicate that DeepSeek did the work.
However I don’t suppose they reveal how these fashions had been educated. Both manner, I wouldn’t have proof that DeepSeek educated its fashions on OpenAI or anybody else’s massive language fashions – or at the very least I did not till at the moment.
Who’re you?
DeepSeek is more and more a thriller wrapped inside a conundrum. There may be some consensus on the truth that DeepSeek arrived extra absolutely fashioned and in much less time than most different fashions, together with Google Gemini, OpenAI’s ChatGPT, and Claude AI.
Only a few within the tech neighborhood belief DeepSeek’s apps on smartphones as a result of there isn’t a strategy to know if China is all that immediate knowledge. Alternatively, the fashions DeepSeek has constructed are spectacular, and a few, together with Microsoft, are already planning to embrace them in their very own AI choices.
Within the case of Microsoft, there’s some irony right here. Copilot was constructed primarily based on cutting-edge ChatGPT fashions, however in latest months, there have been some questions on if the deep monetary partnership between Microsoft and OpenAI will final into the Agentic and later Synthetic Common Intelligence period.
So what if Microsoft begins utilizing DeepSeek, which is probably simply one other offshoot of its present if not future, good friend OpenAI?
The entire thing feels like a complicated mess – and within the meantime, DeepSeek seemingly has an identification disaster.