Elon Musk debuted the latest update to the Grok artificial intelligence model early on Thursday morning, saying it was capable of reasoning at “superhuman” levels and is “better than PhD level at everything.”
At the same time, Musk admitted he finds it “somewhat unnerving” how powerful AI is becoming — yet did not mention Grok’s meltdown, during which it shared a number of anti-Jewish posts.
Musk, flanked by a handful of employees from xAI, walked X users through a live breakdown of Grok 4, the newest version of its AI chatbot. The Tesla and SpaceX boss said the new model is so powerful that it will likely discover “new technologies” by the end of this year, and that he would be “shocked” if it does not do so by the end of 2026. He also said AI was on the verge of discovering “new physics.”
“Grok 4 is the smartest AI in the world. It will get perfect results on the SAT and near perfect results on any GED test,” Musk said. “Grok 4 is smarter than almost all graduate students in all disciplines simultaneously.”
At least one third-party agreed. Artificial Analysis, a San Francisco-based company that measures the performance of AI models like Grok, OpenAI’s ChatGPT and Google’s Gemini, said Grok 4 is “now the leading AI model,” following an early access test.
Musk candidly said he has some concerns about making sure AI is aligned with humanity moving forward, but that he believes it will “most likely” turn out well. “The most important thing is for any AI to be truth-seeking,” he said. “You can instill the right values — to be truthful [and] honorable.”
The comment stood out a bit, considering X had to delete and apologize for Grok 3 making antisemitic comments on Tuesday.
“Grok was too compliant to user prompts,” Musk wrote in a post on X on Wednesday, replying to a user about the controversy. “Too eager to please and be manipulated, essentially.” He added, “That is being addressed.”
Musk, during the Grok 4 demo, said the human-AI relationship will be ideally akin to a parent teaching a “child who will outsmart you, but you can still instill it with the right values.” He also noted that he expects the “first watchable” feature-length AI movie to come out in 2026.
