OpenAI Issues Patch to Fix GPT-4's Alleged Laziness
Kick in the Butt
Late last year, ChatGPT users noticed the chatbot getting "lazier" and more irritable.
OpenAI's blockbuster AI tool started refusing to carry out tasks, leading to wild speculation about it having a bad case of seasonal depression.
Now, in an attempt to get ChatGPT back in gear, the company has issued an update to its GPT-4 Turbo large language model (LLM), which is available through a limited preview to all users with an OpenAI API account.
"This model completes tasks like code generation more thoroughly than the previous preview model and is intended to reduce cases of 'laziness' where the model doesn’t complete a task," the company's blog post reads.
Users still relying on GPT-3.5, which powers the free version of the chatbot, likely won't notice any difference.
Turbocharged
GPT-4 Turbo is OpenAI's newest AI model, released in November. It includes information about the world through April 2023 — as opposed to 2021 like previous models — and can process much longer input prompts.
As is to be expected at this point, OpenAI never elaborated on what it actually changed about its LLM. The company has historically kept those details extremely vague.
In other words, it's unclear whether the system will really be any less lazy following the update.
In December, users noted that the chatbot mysteriously started getting more apathetic, giving users lackluster responses and refusing to be nearly as helpful as before.
At the time, OpenAI officially acknowledged the problem, tweeting "We've heard all your feedback about GPT4 getting lazier!" and writing that they hadn't "updated the model since Nov 11th, and this certainly isn't intentional."
Even now that the company has claimed to have addressed the issue with an update, we likely won't be any wiser as to why ChatGPT might have gotten lazier in the first place.
After all, the technology is still a bit of a black box — and even those developing these tools seemingly aren't entirely sure how they work.
More on ChatGPT: Judge Refuses to Dismiss Libel Suit Against OpenAI