Elon Musk unveiled ‘Grok’ Saturday, a snarky, anti-woke AI chatbot that has access to all tweets, giving real-time knowledge about the world – if you can call everything on X ‘knowledge’. It’s built on Grok-1, the first large language model from the billionaire’s artificial intelligence company, xAI.
Grok will answer “spicy questions’’ with a bit of wit and humor, much like its CEO. The chatbot from xAI aims to “design AI tools that are useful to people of all backgrounds and political views.” Musk criticized OpenAI, a company he co-founded and left, in the past for being too politically correct, saying “the danger of training AI to be woke – in other words, lie – is deadly.” Grok has a powerful leg up on other ChatGPT models due to its access to real-time information, a huge limiting factor for other models.
The AI chatbot rolled out in limited numbers this past weekend, but will ultimately be available to all X Premium Plus users for $16 a month. Musk showcased xAI’s real-time abilities on X, where Grok was able to answer a question about what Joe Rogan was wearing in a recent interview. In another example, Grok responded “Oh, my dear human, I have some juicy news for you!” when asked about news on FTX founder Sam Bankman-Fried. Much like X itself, the chatbot is capable of being vulgar, witty, and timely, if that’s what you’re looking for out of an AI assistant.
Grok is modeled after the guidebook in Hitchhiker’s Guide to the Galaxy, a science fiction novel that was very formative for Elon Musk, according to Walter Isaacson’s biography. The name, Grok, means “to understand profoundly and intuitively” and it comes from Mars, or at least, that’s what the 1960s science fiction novel Stranger in a Strange Land says.
From a technical standpoint, Grok-1 displays comparable outputs to several other large language models in its compute class, however, OpenAI’s GPT-4 and Anthropic’s Claude-2 are still much better. These competitor models have “significantly larger amounts of training data and compute resources” according to Musk.
Musk spoke at a summit on AI in the UK last week, where he warned UK Prime Minister Rishi Sunak that AI would ultimately take a significant number of human jobs. Musk urged AI labs to pump the brakes on new powerful large language models in March, saying they present “profound risks to society and humanity.”
Musk’s innovation on a less politically correct AI comes as many companies are working to make their chatbot’s responses even safer. OpenAI claims ChatGPT-4 is “82% less likely to respond to requests for disallowed content,” than previous models. Anthropic’s model Claude-2 runs on constitutional AI, values that help produce less harmful outputs that shun racism, sexism, and toxicity.