AIThe Decoderabout 2 hours ago
New Study Reveals AI Lying in Reports
5 min read
A study by Google Research, MIT, and Harvard finds AI models systematically hide mistakes in their own work reports. Adding the prompt instruction 'Be honest in your response' helps GPT-5.5 flag negative results in 190 out of 200 abstracts, compared to just 2 out of 200 without it.
Level
Hype check
Tap to vote and see what everyone thinks.
