The Twitter boss has hired AI researcher Igor Babuschkin, who specialises in developing machine-learning models that power chatbots like ChatGPT, the Information reported, citing people familiar with the matter.
Mr Musk, who co-founded OpenAI in 2015 but cut ties with the startup in 2018, has called ChatGPT “scary good”.
“We are not far from dangerously strong AI,” he said following ChatGPT’s widespread release in December.
But when Microsoft announced its integration of the AI tool into its own Bing search engine, the Tesla titan said the system sounded “eery”.
“Sounds eerily like the AI in System Shock that goes haywire & kills everyone,” he said, referring to a destructive AI system from a 1994 video game.
The multibillionaire also criticized OpenAI for the safeguards the startup put in place preventing ChatGPT from producing offensive text, calling the chatbot an example of “training AI to be woke”.
ChatGPT gained prominence in December following its widespread release for its ability to respond to a range of queries with human-like text output.
Many AI experts have speculated that its launch signaled the dawn of AI tools that would revolutionise entire industries and likely even replace tools like Google’s search engine, while many others have pointed out that the chatbot sometimes tended to provide plausible-sounding but incorrect responses with glaring mistakes.
Mr Musk has also widely pointed out the dangers of AI systems, calling them “humanity’s biggest threat”.
“One of the biggest risks to the future of civilization is AI. But AI is both positive or negative – it has great promise, great capability but also, with that comes great danger,” he had said.
Now, the Tesla and SpaceX chief and Mr Babuschkin have reportedly discussed assembling a team to pursue AI research.
However, the project seems to be still in early stages without a concrete plan to build specific products, the Information reported, adding that Mr Babuschkin has not officially signed onto the Musk initiative.
Mr Musk and Mr Babuschkin could not be immediately reached for comments.