Hordes of social media posts—and even some media outlets like Newsweek—are reporting that the Grok AI app (described by Elon Musk as the “smartest AI on Earth”) concluded that there is a 75% chance that Donald Trump is a Russian asset.
I’m no fan of Trump, but I have to point out that Grok’s analysis illustrates a prime weakness of large language models: They are no better than the data sets used to train them. Most online discussion of the Trump-Is-A-Russian-Dupe theory probably originates with people who don’t like Trump and conclude he is in Putin’s pocket. This almost surely skews the result.
Is Trump a “useful idiot”? Maybe or maybe not, but it’s foolish to rely on a Large Language Model’s conclusion.
Large Language Model Weakness: Skewed Training Data
