FT商學院

AI keeps going wrong. What if it can’t be fixed?

Pessimists warn it could wipe out humanity. Optimists hail a medical revolution. Henry Mance meets the sceptics who argue that the technology is simply flawed

The chatbot was speaking complete gibberish. “To rev the virgate, it’s enley to instil group danters,” it told one user. “I’m by. I’m in. I’m for, I’m from, I’m that,” it told another. 

Some users joked that it had ingested acid — or too much James Joyce. Others found that it spoke like an English tourist in Marbella: “Muchas gracias for your understanding, y I’ll ensure we’re being as crystal-clear como l’eau from now on.” It was February 21, and ChatGPT was broken. 

OpenAI, which runs ChatGPT, admitted the problem and fixed it quickly — which is perhaps the least you’d expect from a company recently valued at $80bn. It explained that an update had “introduced a bug with how the model processes language”.

您已閱讀5%(714字),剩餘95%(14250字)包含更多重要資訊,訂閱以繼續探索完整內容,並享受更多專屬服務。
版權聲明:本文版權歸FT中文網所有,未經允許任何單位或個人不得轉載,複製或以任何其他方式使用本文全部或部分,侵權必究。
設置字型大小×
最小
較小
默認
較大
最大
分享×