While when ChatGPT was released into the world, it seemed like it would rule with an iron fist, destroying millions of jobs, the reality is that this OpenAI tool has gone from being incredible… to just a meme in less than 6 months.
ChatGPT is undergoing changes, although until now, it has been incredibly challenging to pinpoint how or why. Users have complained extensively that the language model GPT-4 powering the paid version of the OpenAI chatbot has been degrading over time, providing false answers, and refusing to follow instructions it once adhered to.
And now, new research reveals that, indeed, the AI has undergone some rather profound changes, albeit perhaps not in the way users expected.
Dumber and less accurate AI
A new article published in the preprint archive ArXiv by researchers from Stanford University and the University of Berkeley claims that GPT-4 and GPT-3.5 are responding differently today compared to a few months ago, and not always for the better.
The researchers found that GPT-4 provided much less accurate answers to some more complex mathematical questions. Previously, the system could correctly answer questions about large-scale prime numbers almost every time it was asked, but more recently, it only answered the same question correctly 2.4% of the time.
The older versions of the chatbot used to explain their work in more detail, but the modern editions were much less likely to provide a step-by-step guide to solving a problem, even when asked.
During the same period between March and June of this year, the older version, GPT 3.5, became much more capable of responding to basic math problems, although it remained quite limited in analyzing the generation of more complex codes.
There has been a lot of speculation on the internet about whether ChatGPT is getting worse over time. In recent months, some regular users of ChatGPT on platforms like Reddit and others have openly questioned whether the GPT-4 chatbot is deteriorating or if they are simply becoming aware of the system’s limitations.
Some users reported that when they asked the bot to restructure a text, it consistently ignored the instruction and generated pure fiction instead. Others pointed out that the system failed in relatively straightforward problem-solving tasks, whether they were mathematical or coding-related.
Some of these complaints may have contributed to a decline in user engagement with ChatGPT for the first time since the application was launched last year. The study concludes that yes, ChatGPT is performing worse than it did six months ago.
Some of the links added in the article are part of affiliate campaigns and may represent benefits for Softonic.
