IQ of OpenAI engines and other language models

Hi there,

Since OpenAI is marketed as artificial intelligence, I wonder if it is possible to test the different engines, endpoints and language models with a standardized IQ test for humans. At least for the text and numbers part of such a test. And maybe with the help of a picture to text module for the graphical part.

What do you think: how would the different engines currently perform?

Without wanting to offend the people I know, my gut feeling tells me that at least davinci is already above 100 :slight_smile:

Also, I wonder if there are already internal scales regarding the intelligence of the different engines compared to each other.

And I would also be interested if there is already something like international competitions or rankings of the intelligence of the different systems.

Does anyone have any information for me on this?

Regards

2 Likes

IQ is often measured by speed. Since the latest DAVINCI model “thinks” far faster than most humans, it’s IQ in verbal reasoning would be several orders of magnitude higher than average, probably in the 160 to 170 range. Verbal reasoning is, however, only one measurement of intelligence, as is IQ.

There are numerous other problems with using terms like IQ, fluid intelligence, and crystalized intelligence on an AI model. Even so, these terms can help us understand the power of GPT-3. In terms of crystalized intelligence, GPT-3 towers above pretty much all humans given how broadly it has been “educated”.

2 Likes