Last week, Elon Musk flew to the UK to hype up the existential danger posed by synthetic intelligence. A few days later, he announced that his newest firm, xAI, had developed a strong AI—one with fewer guardrails than the competitors.

The AI mannequin, referred to as Grok (a reputation which means “to understand” in tech circles), “is designed to answer questions with a bit of wit and has a rebellious streak, so please don’t use it if you hate humor!” reads an announcement on the corporate’s web site. “It will also answer spicy questions that are rejected by most other AI systems.”

The announcement doesn’t clarify what a “spicy” or “rebellious” means, however most business AI fashions will refuse to generate sexually specific, violent, or unlawful content material, and they’re designed to keep away from expressing biases picked up from their coaching information. Without such guardrails, the fear is that an AI mannequin may assist terrorists develop a bomb or may end in merchandise that discriminate towards customers based mostly on traits corresponding to race, gender, or age.

xAI doesn’t listing any contact info on its web site, and emails despatched to frequent addresses bounced again. An e mail despatched to the press handle for X obtained an automatic response studying, “Busy now, please check back later.”

The xAI announcement says that Grok is constructed on high of a language mannequin referred to as Grok-1 that has 33 billion parameters. The firm says it developed Grok in two months, a comparatively quick period of time by business requirements, and in addition claims {that a} basic benefit is its “real-time knowledge of the world via the X platform,” or the platform previously often known as Twitter, which Musk acquired for $44 billion in 2022.

Stella Biderman, an AI researcher with EleutherAI, an open supply AI endeavor, says the claims made within the xAI announcement appear believable. Biderman means that Grok will use what’s often known as “retrieval augmented generation” so as to add up-to-date info from X to its output. Other cutting-edge language fashions do that utilizing search engine outcomes and different info.

Large language fashions have confirmed stunningly succesful over the previous 12 months or so, as highlighted most famously by OpenAI’s groundbreaking chatbot, ChatGPT.

These fashions feed on large quantities of textual content taken from books and the net, after which generate textual content in response to a immediate. They are usually additionally given additional coaching by people to make them much less more likely to produce offensive, impolite, or harmful outputs, and to make them extra more likely to reply questions in ways in which appear coherent and plausibly right, though they’re nonetheless liable to producing errors and biases.