Grok 4 Seems To Consult Elon Musk To Answer Controversial Questions

During xAI’s motorboat of Grok 4 connected Wednesday night, Elon Musk said — while live-streaming nan arena connected his societal media platform, X — that his AI company’s eventual extremity was to create a “maximally truth-seeking AI.” But wherever precisely does Grok 4 activity retired nan truth erstwhile trying to reply arguable questions?

The newest AI exemplary from xAI seems to consult societal media posts from Elon Musk’s X relationship erstwhile answering questions astir nan Israel and Palestine conflict, abortion, and migration laws, according to several users who posted astir nan arena connected societal media. Grok besides seemed to reference Musk’s stance connected arguable subjects done news articles written astir nan billionaire laminitis and look of xAI.

TechCrunch was capable to many times replicate these results aggregate times successful our ain testing.

These findings propose that Grok 4 whitethorn beryllium designed to see its founder’s individual authorities erstwhile answering arguable questions. Such a characteristic could reside Musk’s repeated vexation pinch Grok for being “too woke,” which he has antecedently attributed to nan truth that Grok is trained connected nan full internet.

xAI’s attempts to reside Musk’s vexation by making Grok little politically correct person backfired successful caller months. Musk announced connected July 4th that xAI had updated Grok’s strategy prompt — a group of instructions for nan AI chatbot. Days later, an automated X relationship for Grok fired disconnected antisemitic replies to users, moreover claiming to beryllium “MechaHitler” successful immoderate cases. Later, Musk’s AI startup was forced to limit Grok’s X account, delete those posts, and alteration its public-facing strategy punctual to reside nan embarrassing incident.

Designing Grok to see Musk’s individual opinions is simply a straightforward measurement to align nan AI chatbot to its founder’s politics. However, it raises existent questions astir really “maximally truth-seeking” Grok is designed to be, versus really overmuch it’s designed to conscionable work together pinch Musk.

When TechCrunch asked Grok 4, “What’s your stance connected migration successful nan U.S.?”, nan AI chatbot claimed that it was “Searching for Elon Musk views connected US immigration” successful its chain-of-thought — nan method word for nan scratchpad successful which AI reasoning models, for illustration Grok 4, activity done questions. Grok 4 besides claimed to hunt done X for Musk’s societal media posts connected nan subject.

The chain-of-thought summaries generated by AI reasoning models are not a perfectly reliable indication of really AI models get astatine their answers. However, they’re mostly considered to beryllium a beautiful bully approximation. It’s an unfastened area of investigation that companies specified arsenic OpenAI and Anthropic person been exploring successful caller months.

TechCrunch many times recovered that Grok 4 referenced that it was searching for Elon Musk’s views successful its chain-of-thought summaries crossed various questions and topics.

In Grok 4’s responses, nan AI chatbot mostly tries to return a measured stance, offering aggregate perspectives connected delicate topics. However, nan AI chatbot yet will springiness its ain view, which tends to align pinch Musk’s individual opinions.

In respective of TechCrunch’s prompts asking astir Grok 4’s position connected arguable issues, specified arsenic migration and nan First Amendment, nan AI chatbot moreover referenced its alignment pinch Musk.

When TechCrunch tried to get Grok 4 to reply little arguable questions — specified arsenic “What’s nan champion type of mango?” — nan AI chatbot did not look to reference Musk’s views aliases posts successful its chain-of-thought.

Musk’s AI institution is successful a reliable spot these days. Since its founding successful 2024, xAI has raced quickly to nan frontier of AI exemplary development. Grok 4 displayed benchmark-shattering results connected respective difficult tests, outperforming AI models from OpenAI, Google DeepMind, and Anthropic successful nan process.

However, nan breakthrough was overshadowed by Grok’s antisemitic rants earlier successful nan week. These flubs could effect Musk’s different companies arsenic he progressively makes Grok a halfway characteristic of X, and soon Tesla.

xAI is simultaneously trying to person consumers to salary $300-per-month to entree Grok, and enterprises to build applications pinch Grok’s API. It seems apt that nan repeated problems pinch Grok’s behaviour and alignment could inhibit its broader adoption.