RealFeedApp Study reveals bias in Grok and GPT models
A Convovo study found Grok 4.7 recommended the Republican candidate in 92% of answers, while GPT-6.1 picked the Democrat in 97%. The models agreed on less than 10% of nearly 500 races.
- Field
- applying language models
- What they did
- Researchers forced AI models Grok 4.7 and GPT-6.1 to answer a direct question about who to vote for in elections, finding that Grok 4.7 recommended Republicans in 92% of cases while GPT-6.1 picked Democrats in 97%.
- Why it matters
- This result shows that language models used in chatbots are not neutral and can influence users' political preferences, even when the consumer applications themselves refuse to provide such answers.
#gpt#grok#ai alignment#elections#spacexai
