"When you use it more heavily, you start noticing issues you didn't see before."

Log Off, Guy

It hasn't been the best few weeks for OpenAI. Besides a growing number of lawsuits, an FTC inquiry, and declining user numbers, OpenAI has also been battling discontent from GPT-4 users who claim that the paid version of the popular chatbot, powered by the AI firm's advanced large language model GPT-4, has been declining in quality in recent weeks.

"The current GPT-4 is disappointing," reads a post on an OpenAI developer forum, published June 3. "It's like driving a Ferrari for a month then suddenly it turns into a beaten up old pickup. I'm not sure I want to pay for it."

"The last few days to maybe a week, I've been noticing a steady degradation of the responses from GPT-4," another user added on June 14.  "Outside of the objective hard errors I've been getting, the last couple of weeks has been especially bad on the subjective side as well. The creativity and liveliness seems to have been squashed."

But according to OpenAI vice president of product Peter Welinder, the Silicon Valley firm is not making ChatGPT dumber. Instead, you might just need to log off.

"No, we haven't made GPT-4 dumber," Welinder tweeted on Thursday. "Quite the opposite: we make each new version smarter than the previous one. Current hypothesis: When you use it more heavily, you start noticing issues you didn't see before."

Tomato, Tomato

Welinder has a point. Though GPT-4 is OpenAI's most advanced LLM thus far, they've never claimed that the system is perfect. From generating hallucinated information and citations in textual outputs to generating shoddy code — to more standard technical difficulties like extensive downtime — GPT-4 is still under development, so it still has glaring flaws.

Still, most of the folks complaining are the ones who cough up $20 per month for ChatGPT Plus, under the promise that the subscription service offers "faster response times" in addition to "priority access to new features and improvements." It's understandable that they might feel frustrated, or maybe even slighted, when expectations aren't met — especially when those complaints seem to be the consensus among fellow subscribers.

So, is Welinder gaslighting GPT-4 subs? Maybe, maybe not. And to his credit, he did open the Twitter thread up for discussion.

"If you have examples where you believe it's regressed," he wrote, "please reply to this thread and we'll investigate."

Regardless, at the end of the day, logging off from most tech from time to time is healthy. If anything, let the VP's quip be a reminder to shut your computer for a while and touch some grass. Cheers.

More on OpenAI's bad week: FTC Investigating ChatGPT for Saying Harmful Things about People


Share This Article