We live in the era of “go viral first, think later.” The enthusiasm surrounding Artificial Intelligence often reaches exaggerated proportions on social media, distorting real technological advancements. A recent and notable example involved big industry names and premature claims about large language models solving complex mathematical problems.

The Case of the Erdős Problems

It all started when a researcher euphorically announced on X (formerly Twitter) that OpenAI’s latest model, GPT-5, had solved ten unresolved problems left by the prolific mathematician Paul Erdős. The statement suggested that the AI-driven acceleration of science had officially begun. However, the reality was far less glamorous.

Mathematician Thomas Bloom, creator of a repository cataloging these problems, quickly clarified the situation. The fact that a problem is not listed as “solved” on his website does not mean it lacks a solution; it simply means the solution was buried among millions of academic papers that he had not read. The AI model did not create new knowledge; it merely scoured the internet and found the existing answers.

The Distortion of Facts

The ability of AI to retrieve obscure information from scientific literature is, in itself, an incredible breakthrough. Yet, the rush to generate social media engagement overshadowed the tool’s true utility, turning excellent bibliographic research into a false narrative of unprecedented scientific discovery.

Experts point out that this tendency to exaggerate is dangerous. While the internet celebrates false historical milestones for AI in mathematics, more sober assessments show that these models still exhibit significant flaws when applied to critical fields like medicine and law, frequently providing inconsistent guidance.

The Need for Verification

The episode serves as a crucial warning: grandiose claims about artificial intelligence require fewer knee-jerk reactions and more rigorous verification. Flashy victories in tests and competitions are merely starting points. Understanding the true impact of AI requires a deep analysis of how these models operate, separating rapid data retrieval from genuine human scientific innovation.