ChatGPT has been giving answers for several hours that are complete gibberish. He even writes in Spanish.

• In recent hours, numerous complaints have emerged about the chatbot’s abnormal behavior.

• OpenAI appears to have found and fixed the problem.

A little over a year ago we discovered that Bing was going crazy with ChatGPT. The chatbot started chatting and showed that generative AI models not only make mistakes and make up data, but can also give meaningless answers. It just happened again, but this time with ChatGPT.

What happened, ChatGPT. Several Reddit users have posted screenshots showing that ChatGPT’s responses are complete gibberish or constantly mixing up words from different languages. Captures were available on both this platform and this platform. like on Twitter show how ChatGPT for some reason started acting very strange in the last few hours. Some users have shared conversations with ChatGPT Plus, in which, for example, the chatbot’s final goodbye is really strange.

Spanish. Apart from these strange responses, there are several posts where you can see ChatGPT mixing English with some Spanish words. The user described how he was learning Spanish and using ChatGPT in both languages, but after he made a request in English, expecting a response in that language, the chatbot began to mix some Spanish words into the text.

OpenAI solved this problem. Examples keep appearing and there are those who collected several in X (Twitter) threads with all sorts of weird chatbot behavior. Problem seems to affect ChatGPT Plus (based on GPT-4) and the OpenAI platform status website states that “we are investigating reports of unexpected responses from ChatGPT.” They updated this shortly after, saying they had identified the issue and were fixing it, and later said they were monitoring the chatbot’s behavior after making the change.

But we don’t know what happened. At least not at the moment. The company has not provided further details about the incident, nor is it known what caused the behavior or how they were able to resolve it if they appear to have already done so.

Chatbots are (very) imperfect. This new effort further demonstrates that these generative AI models today are as amazing and impressive as they are inaccurate. Their behavior is irregular and it is advisable to look at all the answers they give us.

Image | Focus photo

In Hatak | OpenAI has been using “GPT” as a brand name for its artificial intelligence products for many years. The problem is that you can’t patent it.