This Substack essay uses the controversy surrounding a recent photo of Senator Mitch McConnell to explain why chatbots such as Grok and ChatGPT are poor tools for image verification. Mia highlights multiple examples in which the models confidently fabricated details, contradicted themselves, or arrived at different conclusions based solely on how a question was phrased. Rather than performing true forensic analysis, large language models generate plausible responses from patterns in their training data, making confident errors an unavoidable part of how they work.

The article also explains why even specialized AI-image detectors have important limitations and why authenticating an image ultimately requires careful forensic analysis, not chatbot responses. Mia concludes that the real issue isn't whether the McConnell photo was genuine, but why the public was expected to accept a single low-resolution image as proof of a sitting senator's condition. More broadly, she argues that AI-generated confidence should never be mistaken for evidence.

https://helloitsmia.substack.com/p/grok-does-not-know-if-mitch-mcconnell-is-alive