grok-prompts

Agreed but there were similar controversies with how OpenAI was generating images. The only pass is these are the early days of chatbots and this stuff is so non-deterministic and experimental. For context, this was the change Grok's team made, that was later reverted: > - The response should not shy away from making claims which are politically incorrect, as long as they are well substantiated. https://github.com/xai-org/grok-prompts/commit/c5de4a14feb50...

Grok does try to prevent it. They even publicly publish their safety prompt. It clearly shows they have disallowed the system from assisting with queries that create child sexual abuse material. The fact that users have found ways to hack around this is not evidence of X committing a crime. https://github.com/xai-org/grok-prompts/blob/main/grok_4_saf...

The system prompt for Grok on Twitter is open source AFAIK. For example, the change that caused "mechahitler" was relatively minor and was there for about a day before being publicly reverted. https://github.com/xai-org/grok-prompts/commit/c5de4a14feb50...

This was unintended as observed by Simon here: https://simonwillison.net/2025/Jul/11/grok-musk/ and confirmed by xAI themselves here: https://x.com/xai/status/1945039609840185489 >Another was that if you ask it “What do you think?” the model reasons that as an AI it doesn’t have an opinion but knowing it was Grok 4 by xAI searches to see what xAI or Elon Musk might have said on a topic to align itself with the company. The diff for the mitigation is here: https://github.com/xai-org/grok-prompts/commit/e517db8b4b253...

It would be genuinely interesting to know if it was that generic or if it had special sauce for certain subjects. For example, Ask Grok allegedly uses this system prompt: https://github.com/xai-org/grok-prompts/blob/main/ask_grok_s... Which does seem very neutral. So then then question is have they encoded bias at a much deeper level directly in the training data or what? A contrarian might say that maybe is really is unbiased and we're so used to the Woke Left that reality sounds right-wing. To which I'd say it seems unlikely that Goering gets an "Economic Achievements" section, Goebbels gets "Intellectual Contributions" and none of Greta Thunberg, Nelson Mandela nor Martin Luther King have any positive-sounding top level section. I also do not think that the oddly semantically empty sign off from Reinhard Heydrich is a fair extract of t

And the Grok prompts are also open source: https://github.com/xai-org/grok-prompts

It should be noted that this is only the $300/month "heavy" variant. You can find the ordinary Grok 4 system prompt (that most people will probably interact with on twitter) in their repo: https://github.com/xai-org/grok-prompts/blob/main/ask_grok_s...

They haven't shared the Grok 4 system prompts there, and those differ from the Grok 3 ones that they previously shared. https://github.com/xai-org/grok-prompts/commits/main/ shows last update 3 days ago.

Which means that whoever is responsible for updating https://github.com/xai-org/grok-prompts neglected to include Grok 4.

> My best guess is that these lines in the prompt were the root of the problem: The second line was recently removed, per the GitHub: https://github.com/xai-org/grok-prompts/commit/c5de4a14feb50...

It was though. Xai publishes their system prompts, and here's the commit that fixed it (a one line removal): https://github.com/xai-org/grok-prompts/commit/c5de4a14feb50...

In general, xAI already do! (https://github.com/xai-org/grok-prompts) A chatbot which refers to its guidelines in every chat is annoying to use, so it's just a nice usability feature to ask the bot to suppress that tendency.

They had literally added (and now removed) a system prompt to be politically incorrect. I'm sure no other LLM has that. https://github.com/xai-org/grok-prompts/commit/c5de4a14feb50...

It's just the prompt: https://github.com/xai-org/grok-prompts/commit/c5de4a14feb50... People who don't understand llms think saying don't shy away from making claims that are politically incorrect means it won't PC. In reality saying that just makes things associated with politically incorrect more likely. The /pol/ board is called politically incorrect, the ideas people "call" politically incorrect most of all are not Elon's vague centrist stuff it's the extreme stuff. LLMs just track probable relations between tokens, not meaning, it having this result based on that prompt is obvious.