Lưu trữ — US — techcrunch
OpenAI’s ‘embarrassing’ math
OpenAI faces backlash after misrepresenting GPT-5's math capabilities, claiming it solved ten unsolved Erdős problems. Critics argue this is misleading, stating GPT-5 found references to existing solutions rather than solving them. OpenAI researchers acknowledge their error, stating it highlights the difficulty of literature searches rather than original breakthroughs.
Điểm chính
- OpenAI VP Kevin Weil claimed GPT-5 resolved 10 previously unsolved Erdős problems, leading to criticism.
- Meta's Yann LeCun and Google's Demis Hassabis described the claims as 'embarrassing.'
- Mathematician Thomas Bloom clarified that the problems listed as 'open' did not mean they were solved by GPT-5; it found references instead.
- Sebastien Bubeck from OpenAI admitted that only existing literature solutions were found but argued it was still a notable achievement for literature search difficulty.
Vì sao quan trọng
- Misrepresentation of AI capabilities can lead to public distrust in AI advancements.
- As of 2025, trends include increasing skepticism towards AI claims, especially related to complex tasks like mathematics.
- This incident reflects broader challenges in AI communication, especially regarding actual capabilities versus perceived capabilities.
The controversy over OpenAI's claims about GPT-5 underscores the importance of clear communication in AI development, highlighting the challenges and potential pitfalls in addressing complex tasks and the resultant public perception.