At the recent VivaTech 2025 and Anthropic's first 'Code With Claude' developer event held in Paris, Anthropic CEO Dario Amodei made a shocking and thought-provoking statement. He said that today's advanced artificial intelligence models give less 'hallucinations' i.e. wrong information than humans in cases with some limited and clear facts.
AI is working better than humans
Dario Amodei says that in recent internal tests, models like Claude 3.5 performed better than humans in fact-based quizzes. He said, "If you define hallucination as a person or model saying the wrong thing with full confidence, then humans also do this often."
During the 'Code With Claude' event, where new models like Claude Opus 4 and Claude Sonnet 4 were also introduced, Amodei reiterated the same point. According to TechCrunch, in response to a question, he said, "It depends on how you measure it but I think AI models now probably make fewer mistakes than humans, although their mistakes are sometimes very unique."
A big step towards AGI
These new models of Anthropic are being considered a big step towards AGI (Artificial General Intelligence) by AI. There has been a lot of improvement in memory, code generation, tool use and writing quality. Claude Sonnet 4 has set a new benchmark in the field of AI software engineering by scoring 72.7% in the SWE-Bench test.
However, Amodei also clarified that errors have not been completely eliminated from AI models. AI can still make mistakes, especially when it comes to open-ended or less structured information, such as legal or medical advice. He stressed that the reliability of a model largely depends on the type of question it is asked and the context in which it is being used..jpg)
click and follow Indiaherald WhatsApp channel