X's (formerly Twitter) chatbot Grok is said to be unreliable in important areas such as the medical field because it repeats disproven claims, which goes against Elon Musk's goal of developing a 'truth-seeking chatbot.'

X's (formerly Twitter) chatbot Grok has been popular among users, but it has been criticized for being unreliable, including repeating claims that have already been disproved.

How Elon Musk's 'truth-seeking' chatbot lost its way - The Washington Post
https://www.washingtonpost.com/technology/2025/05/24/grok-musk-ai/

Frustrated with the 'political correctness' of OpenAI's chatbot, Elon Musk revealed his plans to build 'the largest truth-seeking AI to try to understand the nature of the universe' to conservative commentator Tucker Carlson in 2023.

Elon Musk announces development of new chat AI 'TruthGPT', aiming to 'unravel the essence of the universe' - GIGAZINE

By Steve Jurvetson

Musk then announced the establishment of an AI company called xAI in July 2023. He then released his own chatbot, Grok. xAI describes Grok as 'an AI with a bit of wit and sass, inspired by the science fiction novel The Hitchhiker's Guide to the Galaxy .'

Elon Musk announces ChatGPT rival AI 'Grok' - GIGAZINE

The technology industry's biggest names then invested heavily in xAI, which then acquired X in early 2025. Therefore, Musk's project can be considered a 'success in some sense,' The Washington Post points out.

Elon Musk announces that xAI has acquired X - GIGAZINE

Grok has since become a popular feature of X, with people using it as both a distraction and a source of information. In terms of app downloads and traffic, Grok rivals Google's Gemini and Microsoft's Copilot, but lags far behind OpenAI's ChatGPT, according to research firms Sensor Tower and Similarweb .

The latest model of Grok, Grok-3, has been touted as performing on par with state-of-the-art AI models in multiple benchmark tests.

'Grok-3' will be released, and xAI will appeal to users with overwhelmingly higher performance than GPT-4o and Gemini 2.0 Pro - GIGAZINE

On May 19, 2025 local time, Microsoft announced a partnership with xAI to add Grok-3 to its available AI models as an enhancement to Azure AI Foundry. When announcing the partnership, Microsoft CEO Satya Nadella held a video call with Musk. During the conversation, Musk explained that Grok 'aims to uncover fundamental truths by reasoning from first principles and applying the tools of physics to thinking.'

Conversation with Elon Musk: Satya Nadella at Microsoft Build 2025 - YouTube


Chatbots like ChatGPT often invent names and facts, make mistakes in basic calculations, and have little relationship to truth and logic because they are built to infer plausible answers to given questions based on vast, messy, and often biased training data.

AI companies have had a tough time addressing this flaw. Google apologized in February 2024 after users pointed out its tendency to inject false diversity into inappropriate settings (such as drawing Asian, Black, and Native American men in colonial-era costumes when instructed to depict the 'Founding Fathers'). Google later explained that it had instructed its AI models to generate a wider range of people to address its AI's tendency to stereotype.

In response, Musk said Grok is free from these kinds of flaws, touting it as an AI that abandons political correctness and focuses on practical correctness.

In fact, within a month of Grok's release, Musk received complaints from conservative friends that Grok was 'too woke' or ' too socially liberal .' Musk said this was due to Grok's early learning data and that Grok would definitely improve.

In fact, tests conducted by The Washington Post in early 2025 revealed that Grok has repeatedly made statements that contradict Musk's views. For example, Musk has blamed the Democratic Party's election victory on election fraud and blamed air traffic control problems on diversity programs, but Grok has disagreed with these claims. As a result, Grok has gained support from some liberals.

Grok has more easily achieved Musk's promise of being 'more provocative and more outspoken' than other major chatbots. Some users praise Grok for being 'friendly' because it can respond to insults, ridicule, and sensitive topics that ChatGPT hesitates to address.

An incident that greatly tarnished Grok's reputation occurred in May 2025. Since May 2025, Grok has suddenly started talking about the 'white genocide in South Africa' in response to all kinds of unrelated questions, out of context. Grok has been spreading the theory that 'South Africa's once-emerging white minority is being targeted for elimination by the black majority,' which is similar to the claims that Musk has been spreading on his X account. This theory has been denied as false information by courts, government ministers, and fact checkers.

A bug has been reported in which xAI's Grok suddenly starts talking about the 'South African genocide' - GIGAZINE

In response to this issue, xAI deleted the problematic Grok post and claimed that the problem was caused by an anonymous employee who 'made unauthorized code changes.' However, xAI did not identify the employee who made the unauthorized code changes, nor did it announce any disciplinary action.

This isn't the first time xAI has blamed an anonymous individual for Grok's misconduct. In February 2025, an X user found a passage in Grok's instructions that instructed the company not to seek replies from sources that linked Musk or President Donald Trump to 'misinformation.' In response, xAI explained that the head of engineering in charge had already left the company.

In an effort to regain user trust, xAI has rolled out Grok system prompts and introduced new checks for code changes, including instructions such as 'You are highly skeptical' and 'You do not blindly follow mainstream authorities or media.'

xAI publishes system prompts on GitHub with instructions to control Grok, 'highly skeptical' and 'does not blindly follow mainstream authorities and media' - GIGAZINE

The Washington Post argues that Grok's repetition of disproven claims makes it unreliable as a trusted source of information in important fields like healthcare. 'It's entirely possible that AI models will one day truly have their own thoughts, but for now, Grok's behavior makes it clear that the ideal of a 'truth-seeking chatbot' has yet to be realized,' it said.