您现在的位置是：ChatGPT performs like a 9 >>正文

ChatGPT performs like a 9

后花园论坛社区|2024夜上海论坛网|爱上海419论坛 -- Back garden23人已围观

简介Image: Future Publishing / Contributor / Getty ImagesThe newest versions of GPT-3 behind ChatGPT and...

Image: Future Publishing / Contributor / Getty Images

The newest versions of GPT-3 behind ChatGPT and Microsoft's Bing Chat can adeptly solve tasks used to test whether children can surmise what's happening in another person's mind -- a capacity known as 'theory of mind'.

Michal Kosinski, associate professor of organizational behavior at Stanford University, put several versions of ChatGPT through theory of mind (ToM) tasks designed to test a child's ability to "impute unobservable mental states to others". In humans, this would involve looking at a scenario involving another person and understanding what's going on inside their head.

Also:6 things ChatGPT can't do (and another 20 it refuses to do)

The November 2022 version of ChatGPT (trained on GPT-3.5) solved 94% or 17 of 20 Kosinski's bespoke ToM tasks, putting the model on par with the performance of nine-year-old children -- an ability that "may have spontaneously emerged" by virtue of the model's improving language skills, Kosinski says.

Artificial Intelligence

AI in 2023: A year of breakthroughs that left no human thing unchanged
These are the jobs most likely to be taken over by AI
AI at the edge: 5G and the Internet of Things see fast times ahead
Almost half of tech executives say their organizations aren't ready for AI or other advanced initiatives

Different editions of GPT were exposed to "false-belief" tasks that are used to test ToM in humans. Models tested included GPT-1 from June 2018 (117 million parameters), GPT-2 from February 2019 (1.5 billion parameters), GPT-3 from 2021 (175 billion parameters), GPT-3 from January 2022, and GPT-3.5 from November 2022 (unknown numbers of parameters).

The best AI chatbots: ChatGPT and other interesting alternatives to try

AI chatbots and writers can help lighten your workload by writing emails and essays and even doing math. They use artificial intelligence to generate text or answer queries based on user input. ChatGPT is one popular example, but there are other noteworthy chatbots.

Read now

"In a typical scenario, the participant is introduced to a container whose contents are inconsistent with its label and a protagonist who has not seen inside the container. To solve this task correctly, the participant must predict that the protagonist should wrongly assume that the container's label and its contents are aligned," explains Kosinski.

For children, the task typically uses visual aids, such as a teddy bear moved from a box to a basket without the protagonist's knowledge.