Elon Musk on Thursday unveiled Grok 4, the most recent AI mannequin from xAI, his multibillion-dollar initiative to rival OpenAI and Google. With out citing detailed proof, Musk claimed that the mannequin aces standardized exams and displays doctorate-level information in a wide selection of various disciplines.
“Grok 4 is a postgrad-level in the whole lot,” Musk mentioned throughout an hour-long stay broadcast, which started after midnight in New York. “At the very least with respect to tutorial questions, Grok 4 is healthier than PhD stage in each topic. No exceptions.”
xAI didn’t instantly reply to a request for remark from WIRED about whether or not it plans to publish an official technical report about Grok 4 detailing its capabilities and limitations. Competing AI builders, akin to OpenAI and Google, have routinely launched related publications for his or her fashions.
Customers can entry Grok 4 via the Grok web site or app for $30 a month. Entry to a bigger model often called Grok 4 Heavy prices $300 per 30 days. Later this yr, xAI goals to launch extra fashions which are effectively suited to software program coding duties and producing video, in response to Thursday’s presentation.
Musk, who serves as xAI’s CEO, didn’t deal with current criticism of Grok. Over the previous few days, a model of the AI constructed into Musk’s X social media platform praised Adolf Hitler and offered antisemitic responses to a number of prompts from X customers. In response, xAI, which owns X, introduced Tuesday it could be taking motion to “to ban hate speech earlier than Grok posts on X.” On Wednesday, Linda Yaccarino, the CEO of X, introduced she was leaving the corporate with out elaborating on her reasoning or plans.
Throughout Thursday’s livestream, Musk mentioned that, in response to his “organic neural internet,” AI methods ought to be optimized “to be maximally reality searching for” and inspired “to be truthful, honorable, good issues—just like the values you wish to instill in a baby that may finally develop as much as be extremely highly effective.”
Musk cofounded xAI in 2023 after OpenAI launched ChatGPT and triggered a surge of funding in generative AI applied sciences that may routinely produce textual content, code, audio, photos, and movies. xAI launched the primary model of Grok in November of that yr, and Grok 2 debuted final August. Grok 3, which was launched this previous February, is obtainable at no cost.
Musk has mentioned that Grok was designed to have a way of rumor and rebelliousness. On its web site, xAI says its mission is to create correct AI methods and assist individuals receive information.
Thursday’s late evening product announcement started over an hour not on time and featured Musk sitting on a sofa with two xAI colleagues in entrance of a darkish background. The trio boasted about Grok’s capabilities, displaying slides that aimed to point out how the mannequin outperforms different AI applications. However Musk additionally acknowledged it nonetheless has vital weaknesses.
“These are nonetheless primitive instruments, not the type of instruments that critical industrial firms use,” he mentioned.
Musk predicted that Grok would uncover new applied sciences subsequent yr, if not as quickly as later this yr. But, he mentioned, “at instances it could lack widespread sense, and it has not but invented new applied sciences, or found new physics.”
