Elon Musk is now claiming that the update to Grok, his dad joke-spewing AI chatbot, is going to drop next month — but as with all of the multi-hyphenate's big promises, we'll believe it when we see it.
This latest claim came with little prompting after Musk reply-guy'd an account called "@BasedJeffBezos," which posted a video from a podcast interview about AI models "training on each other's data is like a human centipede effect."
"Sadly quite true," the owner of X-formerly-Twitter responded. "It takes a lot of work to purge [large language models] from the Internet training data. Grok 2, which comes out in August, will be a giant improvement in this regard."
To be fair, Musk isn't wrong to point out that large language models (LLMs) often hoover up each others' training data like so many hungry ghosts, a fact well-known to anyone who's read more than two articles about the way LLMs work and the conundrums facing their future.
Where things get sticky, of course, is his claim that the second iteration of Grok, which was launched last fall, is somehow going to be a "giant improvement" on other chatbots when heretofore, we've heard literally nothing about the training data for either version one or two.
Since its beta launch last November, the purportedly "anti-woke" chatbot has proven time and again to be gimmicky and goofy, spewing vulgarities like a middle schooler who just mustered up the courage to utter curse words aloud and with all the lack of inspiration or uniqueness such a vibe denotes. Not long after its launch, Grok was even caught plagiarizing OpenAI's ChatGPT, which is the exact thing Musk decried as "sadly quite true" in his apropos-to-nothing announcement of its successor.
In the ensuing eight-ish months, Grok has been lobotomized, partially updated, and ultimately mostly forgotten even as its founder raised $6 billion to fund his AI efforts — but never once has there been any indication that Musk and his team at xAI had achieved some breakthrough that could constitute the sort of "giant improvement" on current LLM data training practices.
What's more, after @BasedJeffBezos quote-tweeted its fortuitous Musk response and claimed, without any evidence, that Grok 2 will achieve "based" artificial general intelligence (AGI), the X owner replied yet again to say that a third version will be up and running by years end thanks to the expensive AI chips he's purchased for the purpose.
"Grok 3 end of year after training on 100k [Nvidia H100 chips] should be really something special," Musk posted.
And yeah, it probably will be — but not necessarily in the way the bombastic billionaire is suggesting.
More on Grok: AI Researcher Elon Musk Poached From OpenAI Returns to OpenAI
Share This Article