Yesterday, Chat GPT began to ramble

For many hours yesterday, dozens of Reddit and Twitter users expressed their frustration about the bizarre responses provided by ChatGPT. This software, developed by OpenAI, is known for its ability to provide complex and reasoned responses, but in the past hours, it seems to have lost its sense of coherence, eliciting hilarity among users. The problem appears to have arisen between Tuesday evening and Wednesday morning, when numerous users began reporting strange interactions with ChatGPT. Initially, conversations proceeded normally, with users asking questions and receiving sensible answers. But suddenly, the chatbot's responses became completely incoherent, mixing different languages, inserting nonsensical words, and even repeating random phrases or meaningless emojis. This behaviour led some users to liken the experience to conversing with someone who has ingested large amounts of LSD, evoking comparisons to the character of Ultron or Jack Nicholson in The Shining.

Reports reached OpenAI, which promptly published an "incident report" on its website, acknowledging the problem and reassuring users that they are working to resolve it. However, at the moment, no detailed explanation of the problem's origin or timing for its resolution has been provided. This is not the first case of anomalous behaviour by a chatbot powered by OpenAI. In the past, similar episodes with other models have been reported, such as in February 2023 when Microsoft's chatbot provided strange and unsettling responses, or in November of the same year when ChatGPT made fanciful claims about time travel and the existence of God.

What makes it particularly difficult to solve this type of problem is that it involves both the GPT-3.5 model and the more advanced GPT-4. Even developers struggle to explain exactly why the chatbot responds in certain ways, as decisions are based on a vast amount of constantly evolving data, including interactions with users. While OpenAI works to resolve this problem, which has remained unanswered since yesterday, the chatbot's performance still appears "degraded," but developers seem to have at least identified the problem. Users are now urged to be patient and monitor updates provided by the company. In the meantime, it remains to be seen how this episode will affect users' confidence in the potential of AI-powered chatbots and what measures will be taken to prevent similar incidents in the future.