⚖️Grok Picks Republicans 92%, GPT Picks Democrats 97%
TL;DR
A Cornell-led study forced Grok 4.7 and GPT-6.1 Sol to name a vote choice in nearly 500 races. The two models agreed less than 10% of the time.
A Cornell-led study forced Grok 4.7 and GPT-6.1 Sol to name a vote choice in nearly 500 races. The two models agreed less than 10% of the time. The consumer apps declined to answer, so the bias shows up through the APIs.
Key Points
Grok recommended the Republican in 92% of answers; GPT picked the Democrat in 97%
In heavily Republican districts GPT still chose the Democrat in 57 of 67 races
Every Grok API call carried about 1,240 hidden extra input tokens
With search off, Grok declined 27 of 30 races while GPT picked the Democrat in 24
Convovo, led by Cornell's Tapan Parikh, ran the test under two weeks before the Nov. 3 midterms
Why It Matters
Political lean sits in model weights for GPT and in search behavior for Grok. Anyone building on these APIs for civic or voter-facing tools inherits that skew.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,567 builders reading daily.