OpenAI GPT-4 Performs Better Than Most Humans on Bar Exam, GRE and Many AP Subject Exams

[Yazata]

It does better than 80% of humans on the Uniform Bar Exam, close to 80th percentile on the GRE quantitative and almost 100th percentile (better than almost all humans) on GRE verbal.

GPT-4

We’ve created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various...

openai.com

 

[Heyo]

It does better than 80% of humans on the Uniform Bar Exam, close to 80th percentile on the GRE quantitative and almost 100th percentile (better than almost all humans) on GRE verbal.

GPT-4

We’ve created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various...

openai.com

Are the humans who take these tests also allowed to use the internet?

 

[Debater Slayer]

This should not be surprising, in my opinion. ChatGPT learns from billions of data points, which no human could possibly do. These obviously include the kinds of knowledge that would be sufficient to pass or ace a standardized exam.

Supercomputers have long been far better than humans at a lot of tasks mainly due to their abundant processing power giving them the ability to retain, access, and process information beyond human limits. Garry Kasparov, arguably the greatest chess player of all time, even lost a chess match to IBM's Deep Blue supercomputer in 1997. Now we're seeing a vaguely similar sort of computational prowess come to our own screens via AI instead of being exclusive to research labs or multi-billion-dollar corporations.

The defining aspect of human intellect is our self-awareness. We have desires, self-contemplation, and a dynamic sense of identity. An AI remains a mere combination of algorithms and datasets, although it is quite impressive that it has become so advanced at processing and learning from data as well as communicating the resultant information in natural, fluent language.

 

[Twilight Hue]

It does better than 80% of humans on the Uniform Bar Exam, close to 80th percentile on the GRE quantitative and almost 100th percentile (better than almost all humans) on GRE verbal.

GPT-4

We’ve created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various...

openai.com

It's why I think AI lawyers will one day become a standard.

Particularly among the injury and class action crowd.

What a glorious day that would bring to the country should that become a reality.

 

[Twilight Hue]

This should not be surprising, in my opinion. ChatGPT learns from billions of data points, which no human could possibly do. These obviously include the kinds of knowledge that would be sufficient to pass or ace a standardized exam.

Supercomputers have long been far better than humans at a lot of tasks mainly due to their abundant processing power giving them the ability to retain, access, and process information beyond human limits. Garry Kasparov, arguably the greatest chess player of all time, even lost a chess match to IBM's Deep Blue supercomputer in 1997. Now we're seeing a vaguely similar sort of computational prowess come to our own screens via AI instead of being exclusive to research labs or multi-billion-dollar corporations.

The defining aspect of human intellect is our self-awareness. We have desires, self-contemplation, and a dynamic sense of identity. An AI remains a mere combination of algorithms and datasets, although it is quite impressive that it has become so advanced at processing and learning from data as well as communicating the resultant information in natural, fluent language.

Another step towards the Fermi Paradox imv.