ChatGPT goes berserk

Late on Wednesday 21st of February, it seemed like ChatGPT briefly lost its mind. According to OpenAI the the incident involved a bug introduced during an optimisation attempt that affected how ChatGPT processes language, leading to nonsensical responses. This issue was quickly identified and resolved. Gary Marcus discussed the incident, highlighting the unpredictable nature of AI systems and the importance of developing more interpretable, maintainable, and debuggable technologies. He framed the incident as a wakeup call for the need for trustworthy AI, emphasising the challenges of ensuring AI safety and stability. As always, the Hacker News comments on each article are enlightening.

Visit Original Article →