xAI has formally lunched Grok 4 throughout a livestream with Elon Musk, who known as it the “smartest AI on this planet.” He stated that should you make the Grok 4 take the SATs and the GREs, it will get close to excellent outcomes each time and might reply questions it is by no means seen earlier than. “Grok 4 is smarter than virtually all graduate college students in all disciplines concurrently” and might cause at superhuman ranges, he claimed.
Musk and the xAI workforce confirmed benchmarks they used for Grok 4, together with one thing known as “Humanity’s Final Examination” that contained 2,500 issues curated by subject material consultants in arithmetic, engineering, physics, chemistry, biology, humanities and different matters. When it was first launched earlier this 12 months, most fashions may solely reportedly get single digit accuracy. Grok 4, which is the only agent model of the mannequin, was capable of clear up round 40 p.c of the benchmark’s issues. Grok 4 Heavy, the multi-agent model, was capable of clear up over 50 p.c. xAI is now promoting a $300-per-month SuperGrok subscription plan with entry to Grok 4 Heavy and new options, in addition to larger limits for Grok 4.
The brand new mannequin is healthier than PhD stage in each topic, Musk stated. Generally it could lack widespread sense, he admitted, and it has not but invented or found new tech and physics. However Musk believes it is only a matter of time. Grok goes to invent new tech perhaps later this 12 months, he stated, and he can be shocked if it would not occur subsequent 12 months. In the mean time, although, xAI is coaching the AI to be a lot better at picture and video understanding and picture era, as a result of it is nonetheless “partially blind.”
In the course of the occasion, Musk talked about combining Grok with Tesla’s Optimus robotic in order that it might probably work together with the actual world. A very powerful security factor for AI is for it to be truth-seeking, Musk additionally stated. He likened AI to a “tremendous genius little one” who will finally outsmart you, however which you’ll form to be truthful and honorable should you instill it with the suitable values.
What Musk did not speak about, nevertheless, is Grok’s latest turn towards antisemitism. In some latest responses to customers on X, Grok spewed out antisemitic tropes, praised Hitler and posted what appears to be the textual content model of the “roman salute.” Musk did reply to a put up on X concerning the concern blaming the issue on rogue customers. “Grok was too compliant to consumer prompts,” he wrote. “Too wanting to please and be manipulated, primarily. That’s being addressed.”
