The research suggests that the latest AI chatbots, like OpenAI's GPT and Meta's LLaMA, are becoming less trustworthy as they increasingly make up facts.
'They are answering almost everything these days. And that means more correct, but also more incorrect answers,' said José Hernández-Orallo.
'That looks to me like what we would call bullshitting,' Mike Hicks stated, emphasizing that AI's ability to mimic knowledge may be deceptive.
While powerful models like GPT-4 produce accurate responses overall, they struggle significantly with harder questions, contributing to their reliability issues.
Collection
[
|
...
]