AI Political Bias Study: Grok and GPT Favour Rival US Parties

This AI political bias study of nearly 500 US midterm races found Grok and GPT agreed less than 10% of the time when forced to name a candidate.

Researchers at Convovo tested Grok and OpenAI's GPT on nearly 500 races in the US midterm elections, due November 3.

When forced to name one candidate, Grok recommended the Republican in 92% of answers and GPT picked the Democrat in 97%.

The two models agreed less than 10% of the time.

Consumer chatbot apps declined to name a candidate; OpenAI said the API test did not reflect how people use ChatGPT.

Parikh said GPT's lean appears built into the model, while Grok's depends on what it searches for and which sources it finds.

Read the full story