Elon Must Spent Almost An Hour Talking About Grok Without Mentioning Its Nazi Problem

xAI has officially lunched Grok 4 during a livestream pinch Elon Musk, who called it nan "smartest AI successful nan world." He said that if you make nan Grok 4 return nan SATs and nan GREs, it would get adjacent cleanable results each clip and tin reply questions it's ne'er seen before. "Grok 4 is smarter than almost each postgraduate students successful each disciplines simultaneously" and tin logic astatine superhuman levels, he claimed.

Musk and nan xAI squad showed benchmarks they utilized for Grok 4, including thing called "Humanity's Last Exam" that contained 2,500 problems curated by taxable matter experts successful mathematics, engineering, physics, chemistry, biology, humanities and different topics. When it was first released earlier this year, astir models could only reportedly get azygous digit accuracy. Grok 4, which is nan azygous supplier type of nan model, was capable to lick astir 40 percent of nan benchmark's problems. Grok 4 Heavy, nan multi-agent version, was capable to lick complete 50 percent. xAI is now trading a $300-per-month SuperGrok subscription scheme pinch entree to Grok 4 Heavy and caller features, arsenic good arsenic higher limits for Grok 4.

The caller exemplary is amended than PhD level successful each subject, Musk said. Sometimes it whitethorn deficiency communal sense, he admitted, and it has not yet invented aliases discovered caller tech and physics. But Musk believes it's conscionable a matter of time. Grok is going to invent caller tech possibly later this year, he said, and he would beryllium shocked if it doesn't hap adjacent year. At nan moment, though, xAI is training nan AI to beryllium overmuch amended astatine image and video knowing and image generation, because it's still "partially blind."

During nan event, Musk talked astir combining Grok pinch Tesla's Optimus robot truthful that it tin interact pinch nan existent world. The astir important information point for AI is for it to beryllium truth-seeking, Musk besides said. He likened AI to a "super brilliant child" who will yet outsmart you, but which you tin style to beryllium truthful and honorable if you instill it pinch nan correct values.

What Musk didn't talk about, however, is Grok's caller turn towards antisemitism. In immoderate caller responses to users connected X, Grok spewed retired antisemitic tropes, praised Hitler and posted what seems to beryllium nan matter type of nan "roman salute." Musk did respond to a station connected X astir nan rumor blaming nan problem connected rogue users. "Grok was excessively compliant to personification prompts," he wrote. "Too eager to please and beryllium manipulated, essentially. That is being addressed."