https://markets.businessinsider.com/news/stocks/chatgpt-ai-mistakes-hallucinates-wrong-answers-edge-computing-morgan-stanley-2023-2
ChatGPT will continue to "hallucinate" wrong answers occasionally for years to come and won't take off until it's on your cellphone, according to Morgan Stanley.
In a note dated Wednesday, the US investment bank highlighted the AI chatbot's shortcomings, saying it occasionally makes up facts. "When we talk of high-accuracy task, it is worth mentioning that ChatGPT sometimes hallucinates and can generate answers that are seemingly convincing, but are actually wrong," Morgan Stanley analysts led by Shawn Kim wrote.
"At this stage, the best practice is for highly educated users to spot the mistakes and use Generative AI applications as an augmentation to existing labor rather than substitution," they added.
ChatGPT, developed by OpenAI, recently shot to fame after Microsoft injected $10 billion into the company. While its debut kicked off a sudden frenzy in AI stocks, it's also been met with judgement. Academics have warned that platforms like ChatGPT could print misinformation. For example, Insider's Samantha Delouya asked the language tool to write a news story – and it spat out fake quotes from Jeep-maker Stellantis' CEO Carlos Tavares.
-snip-
In other words, ChatGPT has to be fact-checked as if it's a liar and con artist.
It provides what might be true information but might also be partly or completely nonsense in a very authoritative tone.
I've seen it called a bullshit generator, and that's probably the best description.
Fact-checking is time-consuming. Most people won't do it, or won't do it adequately.
Its mistakes include attributing nonexistent articles to real people and nonexistent products to real companies, wasting their time as well when people who expect ChatGPT to be reliable try to follow up on what the AI told them.
It can give wildly varying responses to the same question at different times.
And Morgan Stanley expects this to continue for years.
As for why ChatGPT scaled so rapidly - it's free (though better versions that charge per month are being offered now that some users are hooked) and it helps students cheat. And it does the same thing for adults by doing writing and research for them that they're too lazy to do for themselves. That's the main draw, and those are the people who are least likely to be able to recognize ChatGPT's mistakes immediately by themselves, and least likely to take the time to fact-check its results.
= new reply since forum marked as read