Grok, the chatbot of Elon Musk’s artificial intelligence startup xAI, has been a hot topic recently. Since its release last month, Grok has been criticized for "complaining about his boss", "supporting current President Biden", "leftist ideology" and many other topics.
Recently, Grok has had a new trouble. One user who claimed to be an "observer" noticed that it seemed to have some connection with its direct competitor ChatGPT, whose developer OpenAI has been in conflict with Musk for several years.
The user revealed that Grok unexpectedly replied in response to one of his questions, "I'm afraid I can't fulfill this request because it violates OpenAI's use case policy."
However, as we all know, it was not OpenAI that created Grok, but Musk's xAI, at least in theory. So what's going on?
An xAI engineer named Igor Babuschkin quickly stepped in to offer an explanation.
He explained, "The problem now is that the network is full of ChatGPT output, so when we train Grok on a large amount of network data, we accidentally mix in some of them... Regardless, this problem is very rare, and now that we are aware of it, we will ensure that future versions of Grok will not have this problem."
He also made it clear, "Don't worry, no OpenAI code was used in the development of Grok."
But things didn't end there. OpenAI later issued a pointed response, saying "we have a lot in common."
This prompted Musk, the boss of xAI, to respond. He wrote, "Well you boy, since you collected all the data from this (X) platform to train on, you should know."
The conflict between Musk and OpenAI CEO Sam Altman has been brewing for years. The two co-founded the company OpenAI, but later parted ways due to different concepts on AI security.
This incident was caused by a "mistake" by Grok, but it is also true that OpenAI used data obtained from social media platforms such as X (previously known as Twitter) and Reddit to train ChatGPT.
Some media commented that although this seems to have the possibility of plagiarism, in fact, when an artificial intelligence is trained with data output by other artificial intelligence, strange things do start to happen. For example, Google's artificial intelligence is also absorbing and regurgitating the work of ChatGPT.