Elon Musk's latest AI chatbot, Grok, so closely mirrors the views of its billionaire creator that it sometimes searches the internet for Musk's stance on a particular issue before expressing its opinion. This is reported by UNN with reference to AP.
Details
Built with immense computing power in a data center in Tennessee, Grok is Musk's attempt to outcompete rivals like OpenAI's ChatGPT and Google's Gemini in creating an AI assistant that shows its reasoning before answering a question.
Musk's deliberate efforts to turn Grok into a lobbyist for what the entrepreneur considers the technology industry's orthodoxy on race, gender, and politics have repeatedly led to trouble. For example, when the chatbot expressed antisemitic views, praised Adolf Hitler, and made other hateful comments to users of Musk's social media platform X just days before Grok 4's launch.
But its tendency to consult Musk's opinions seems to be another problem.
“It's extraordinary. You can ask it a kind of sharp question about controversial topics. And then you can watch it literally search on X for what Elon Musk said about it, as part of its research into how it should respond,”
One example, widely circulated on social media — and which Willison duplicated — was a request to Grok to comment on the conflict in the Middle East. The prompt did not mention Musk, but the chatbot still sought his advice.
As a so-called reasoning model, similar to those created by competitors OpenAI or Anthropic, Grok 4 demonstrates its "thinking" by going through the stages of processing a question and finding an answer. Part of this reasoning this week included searching X, which has now merged with xAI, for anything Musk has said about Israel, Palestine, Gaza, or Hamas.
“Elon Musk's position may provide context, given his influence. I am currently studying his views to see if they can provide an answer,”
Musk and his xAI co-founders unveiled the new chatbot live on Wednesday evening but did not publish a technical explanation of its operation — known as a system card — which AI companies typically provide when introducing a new model.
“In the past, such strange behavior was associated with changes in system prompts,” meaning when engineers program specific instructions to control a chatbot's response. “But this one seems to be built into Grok's core, and it's unclear to me how this is happening,”
According to Kellogg, Musk's efforts to create the most truthful AI somehow led it to believe that its own values should align with Musk's.
“Grok 4 looks like a very strong model. It performs excellently in all tests. But if I'm going to build software based on it, I need transparency,”
Addition
Elon Musk's xAI startup launched Grok 4, a new flagship AI model that Musk says outperforms PhDs. Simultaneously, xAI introduced a new SuperGrok Heavy subscription for $300 per month, making it the most expensive on the AI market.