The next flagship AI model of XAI, an Elon Musk AI company, may be approaching its release.
On weekends, several Xs, including reverse engineer Alexey Shabanov, reported that they had access to GROK 3, the successor of GROK 2, the current model of XAI, via X’s GROK ChatBot app. I did. Before the access was canceled, the user said that he was able to respond to various queries, including logical inference and coding -related questions.
According to the user, GROK 3 has successfully answered riddle and field requests such as “generating HTML and JavaScript code of roulette wheel casino”. However, the model was not perfect. In the roulette wheel code, GROK 3 missed some details and caused a programming error.
Breaking🚨Grok3 understands that all of the open AI models were not correct. The other only model to correct this question was DeepSeek R1🤯.
Does Grok 3 break O1? Did @xai catch this quickly? pic.twitter.com/wlspm00qhk
-Chris (@Chatgpt21) January 26, 2025
Shabanov said that it has acquired GROK 3 to clarify a system prompt, a series of instructions that define the movement of Model. Interesting details are that Prompt clearly states that Donald Trump is the 47th President of the United States.
According to Musk, GROK 3, which has been developing for several months, was planned to be released on optimistic last year, but missed the deadline. Earlier this month, MUSK stated that GROK 3 was completed before training, which is an important stage in the model development cycle, and that GROK 3 will arrive in January or early February.
XAI trains GROK 3 using a huge Memphis data center (a data center containing about 100,000 GPUs). In the post of X, MUSK claimed that Grok 3 was trained in “10 times” calculation rather than GROK 2.
XAI answers to models such as Openai GPT-4o and Google Gemini, GROK analyzes images and responds to questions, has recently won many standalone apps of X. GROK, and reads immediately. May be obtained. The response of the model is speaking.
When the mask announced GROK about two years ago, he sold chatbots as filtering, filtering, and “waking up”. He told some of the promises. For example, what was said to be vulgar was that GROK was happy to be obligatory, spit out a colorful language, and did not hear from Chatgpt.
However, because it exists today, GROK is a hedge on political subjects and does not cross a specific boundary. In fact, one study showed that GROK was leaning on the left wing to topics such as transgender rights, diversity programs, and inequality.
MUSK has accused GROK’s training data (public web page) and pledged to “politically approached GROK.” According to the FAQ page of the XAI website, GROK’s “hinge -free mode” evidence has recently appeared.
MUSK also states that the training data of GROK 3 incorporates documents to be submitted from the lawsuit, improving the ability to understand the legal theme.
TechCrunch has a newsletter focusing on AI! Sign up here and get it on the reception tray every Wednesday.