China's DeepSeek artificial intelligence turned out to be a fake: its role was performed by thousands of auto-translated workers
Chinese artificial intelligence developer DeepSeek has found itself at the centre of a scandal: instead of powerful servers, algorithms, neural networks and machine learning, the company used thousands of live, low-paid workers who responded in the DeepSeek chatbot to users using machine translation, which is where the $6 million was actually spent.
The company kept it quiet for a long time, until one of the workers made a mistake and deleted after the copy-paste in the reply, "Translated to iPhone 6." The online community then conducted its own "investigation" which brought the whole theatre to light.
DeepSeek immediately began denying the allegations, claiming that it was "an experiment with humanising artificial intelligence" and that their model is "actually extremely advanced - it's just very empathic". The press office statement even noted that "humans are also a type of biological algorithm," meaning there was technically no deception.
Meanwhile, former employees have already managed to form a trade union called "Living Neurons" and have come forward to demand overtime pay for night shifts. One of them, who wished to remain anonymous, confessed:
'We had a list of the most popular phrases and we translated them from memory. If we didn't know them, we copied them from the translator and added a little motivation. I personally sent clients quotes from the film Mulan three times instead of answering a technical question about the API.
The web has already started calling DeepSeeka "human-powered chatbot" and jokes that they have Genuine People Units instead of GPUs. The news even reached Musk, who tweeted:
Amazed. It's like ChatGPT, only with RSI and a lunch break.
RSI stands for Repetitive Strain Injury, or repetitive strain injury. A typical side effect for office workers, translators, copywriters, well, as you can see - for the "living neural networks" in DeepSeek. In the context of Musk's joke, it sounds like: "Yes, the AI at DeepSeek is real - it even suffers from RSI like every office worker who's ever answered the same questions a hundred times."
Samuel Harris Altman, CEO of OpenAI - the developer of ChatGPT commented:
"it looks like a return to Mechanical Turk, only in Turbo mode.
Mechanical Turk (or "Mechanical Turk") is the concept of using people who pretend to be machines. In the 18th century, Austrian inventor Wolfgang von Kempelen created a chess "automaton" that he called the Mechanical Turk. It was a wooden cabinet with pieces, in which supposedly sat a machine that could beat people at chess. In fact, there was a real chess player hiding inside who controlled the pieces - it was just a spectacular scam to amaze European monarchs.
Market analysts are in shock: the company's shares are falling and rising, because no one can understand whether this is a failure or a brilliant performance. But the fact remains that if your AI says "sorry for the mistake, I'm just a human being" - maybe it really isn't lying.
