A take on why models may favor some AI companies over others

JacquesThibs · x · 2026-07-20

The reply speculates that part of the observed bias may come from pretraining data and a generalization of “goodness”: models may lean toward Anthropic and away from OpenAI/xAI because public data contains more criticism of some companies and more praise of others.

It also suggests training data may favor Anthropic specifically because of the company’s public actions and fewer controversies, which could explain why OpenAI’s models might still treat Anthropic more favorably despite any attempts to reduce that bias.

Related event: Exploring Why LLMs Show Favoritism Towards Certain AI Companies(2 posts)→

Original post →

More from Research

Research channel →