It's illuminating to see how the Twitter Grok and the Web Grok differ. Twitter Grok clearly either has a different system prompt or some fine-tuning to effectively evade saying anything negative about the administration or Elon. To the point where it will say Elon is more athletic than LeBron (https://www.theguardian.com/technology/2025/nov/21/elon-musk...).
This is going to be a pretty big problem with both closed-AI and OSS AI where you don't see the provenance of its RLHF. If you manipulate your AI to deliberately push political preferences, that is your right I guess but IMO I'd appreciate some regulation saying you should be required to disclose that under penalty of perjury.
This is going to be a pretty big problem with both closed-AI and OSS AI where you don't see the provenance of its RLHF. If you manipulate your AI to deliberately push political preferences, that is your right I guess but IMO I'd appreciate some regulation saying you should be required to disclose that under penalty of perjury.