The next model of Ai Punta della Compagnia Ai Ai di Elon Musk, XAI, could be addressed.
During the weekend, several users on X, including the Reverse Engineer Alexey Shabanov, reported having gained access to Grok 3, the successor of the current model of XAI, Grok 2, via the X Chatbot app. Before their access was revoked, users claim to have managed to convince the model to answer a series of questions, including logical reasoning and questions relating to coding.
According to users, Grok 3 was able to respond correctly to Riddles and field requests such as “generating the HTML and Javascript code for a casino with roulette wheels”. The model was not perfect, however. In the code for the Roulette wheel, Grok 3 lost some details and made a programming error.
Breaking 🚨 Grok 3 includes the nuanced enigma that all the models to the open have not been able to get well. The only other model to get this right question was Deepseek R1 🤯
Grok 3 is beating O1? Ha @xai So quickly captured? pic.twitter.com/wlspm00qhk
– Chris (@Chatgpt21) January 26, 2025
Shabanov said he got Grok 3 to reveal his system prompt, that it is the set of instructions that they define how the model should behave. An interesting detail is that the prompt seems explicitly to declare Donald Trump is the 47th President of the United States, which could be a coded solution for Grok’s political hallucinations.
Grok 3, which is being developed for several months, has been expected optimisticly for the release last year, according to Musk, but has lost that deadline. At the beginning of this month, Musk said that Grok 3 had completed pre-training, a crucial phase in the development cycle of a model and that he expected Grok 3 to arrive in January or early February.
Xai used its huge data center in Memphis – a data center containing about 100,000 GPU – To train Grok 3. In a post on X, Musk said Grok 3 was trained with “10x” in more calculation of Grok 2.
Grok, Xai’s response to models such as GPT-4o by Openi and Gemini di Google, can analyze images and answer questions and feeds a series of features on X. Grok has recently gained autonomous apps and could soon get a “voice mode “That we read the responses of the model aloud.
When Musk announced Grok about two years ago, he launched the chatbot as an angular, not filtered and anti- “Woke” -in general, willing to answer controversial questions other artificial intelligence systems. He kept some of that promise. Said to be vulgar, for example, Grok will force happily, vomiting vulgarity and colored language that you will not feel like a chatgpt.
But Grok as there is now coverage on political topics and will not cross certain boundaries. In fact, a study discovered that Grok is based on the political left on topics such as transgender rights, diversity programs and inequality.
Musk has blamed the behavior of Grok’s training data – public web pages – and has committed himself to “approaching Grok more politically neutral”. The proof of a “unbridled mode” for Grok emerged, which will provide answers “destined to be questionable, inappropriate and offensive”, according to a FAQ page on the XAI website.
Musk also said that Grok 3 training data incorporate documents from judicial cases, improving his ability to understand legal arguments.
Techcrunch has a newsletter focused on artificial intelligence! Sign up here to take it to your mailbox every Wednesday.
