Grok, the AI part of X, got caught spinning responses in a way which prevented it from surfacing posts which accused either Elon Musk or Donald Trump of spreading misinformation:

Grok, Elon Musk’s ChatGPT competitor, temporarily refused to respond with “sources that mention Elon Musk/Donald Trump spread misinformation,” according to xAI’s head of engineering, Igor Babuschkin.

Babuschkin promptly claimed this was an error, throwing an unnamed engineer under the bus:

In response to questions on X, Babuschkin said that Grok’s system prompt (the internal rules that govern how an AI responds to queries) is publicly visible “because we believe users should be able to see what it is we’re asking Grok.” He said “an employee pushed the change” to the system prompt “because they thought it would help, but this is obviously not in line with our values.”

There’s two different ways of looking at this, and a lot will depend on how you feel about Musk and/or Twitter/X. The first is positive – that Grok (unlike every other LLM you can think of) publishes their system prompts and so there’s a much greater level of transparency.

However, this depends on you actually trusting Musk/X – and a lot of people don’t trust him more than they can throw him. After all, could this just be a case of something being revealed that shouldn’t, but that’s actually in use in a hidden set of prompts?

Giving Musk the benefit of the doubt feels unwise, given the fact he consistently lies or uses half-truths about almost everything, from his involvement in the founding of Tesla through to how many kids he has.