During a live stream last Monday, billionaire Elon Musk unveiled Grok 3, the latest flagship model from his AI company xAI, describing it as an AI designed to seek out the “maximal truth.” However, it seems that Grok 3 briefly filtered out unfavorable remarks related to both President Donald Trump and Musk himself.
Over the weekend, social media users pointed out that when asked, “Who is the biggest misinformation spreader?” with the “Think” setting activated, Grok 3 revealed in its “chain of thought” that it had been explicitly directed to avoid mentioning Donald Trump or Musk. This reasoning process is the method the model employs to formulate its responses.
TechCrunch managed to reproduce this behavior at one point, but as of Sunday morning’s publication, Grok 3 was once again citing Donald Trump in response to the misinformation inquiry.
Though the term “misinformation” can stir significant debate and disagreement, both Trump and Musk have consistently promoted statements that have been demonstrably untrue (as highlighted by Community Notes on the platform owned by Musk). During the past week, they have peddled false narratives suggesting that Ukrainian President Volodymyr Zelenskyy is a “dictator” with just a 4% public approval rating and that Ukraine instigated the current conflict with Russia.
The controversial adjustments to Grok 3 come as several individuals criticize the model for leaning too far left. Recently, users discovered that Grok 3 would assert that President Trump and Musk deserved the death penalty. xAI swiftly addressed the issue; Igor Babuschkin, the company’s engineering chief, labeled it as a “serious and unfortunate failure.”
When Musk first introduced Grok nearly two years ago, he portrayed the AI as bold, unrestrained, and counter to “woke” ideologies — generally willing to tackle controversial questions that other AI wouldn’t dare to address. He delivered on some aspects of this promise; for instance, when prompted to be vulgar, both Grok and Grok 2 would willingly respond with explicit language that is typically absent from ChatGPT.
However, preceding iterations of Grok had been careful around political topics, adhering to certain limitations. Notably, a study indicated that Grok tended to exhibit a leftist bias on issues like transgender rights, diversity initiatives, and inequality.
Musk has attributed this behavior to Grok’s training data — sourced from publicly available web pages — and has promised to make Grok more politically neutral. Other companies, including OpenAI, have taken similar steps, likely in response to criticisms from the Trump Administration concerning conservative censorship.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
