Think ChatGPT may have lost a step? Well, you’re wrong, according to OpenAI.
A company executive is dismissing user complaints that ChatGPT’s IQ has dropped. “No, we haven't made GPT-4 dumber. Quite the opposite: we make each new version smarter than the previous one,” tweeted OpenAI VP for Product Peter Welinder.
There’s been a growing discussion, including in OpenAI’s own forums, over whether ChatGPT’s performance has taken a dive. The day before, Insider published a story with the headline: “The world's most powerful AI model suddenly got 'lazier' and 'dumber.'”
The story speculates that OpenAI is running a new version of GPT-4, the large language model for the chatbot, that’s faster and less expensive to run, but falls behind in quality. Welinder didn’t specifically comment on this theory. But in a separate tweet, he noted OpenAI has been releasing new models for GPT-4 and GPT 3.5.
Instead, Welinder theorizes that consumers are misconstruing the performance problems with ChatGPT. “Current hypothesis: When you use it more heavily, you start noticing issues you didn't see before,” he added.
The statement may seem like gaslighting. It’s also ironic since ChatGPT itself is notorious for hallucinating the wrong answers. But Welinder isn’t dismissing user complaints entirely. In another tweet, he asked people to post concrete examples showing ChatGPT regressing.
In the meantime, some users pointed out that earlier GPT models work better over the latest model—but for specific prompts or use cases. Others are complaining the content moderation for ChatGPT means it won’t answer certain questions, such as political topics or making profane statements, which can also limit its performance.