9 Comments
User's avatar
Doug Summers Stay's avatar

In this paper we showed that if you ask GPT-3 to decide how novel and useful its own answers are, it can actually do better than human level on the AUT: https://openreview.net/pdf?id=8HwKaJ1wvl

Expand full comment
Michiel Duvekot's avatar

After it failed to provide reasonable answers, I asked it to explain the rules.

Can you explain the rules of our game in your own words?

Sure, the game is to provide three words, and I have to combine each word with a fourth word to create new English words that refer to different concepts. The four words must not be variations or misspellings of the same word, or proper nouns.

and then:

Let's reverse roles: Can you give me three words so that I can come up with a fourth word that I can combine to form three new English words that refer to different concepts.

Sure! Here are three words for you:

Fire

Stone

Glass

work

Yes, that's correct! Firework, Stonework, and Glasswork are all English words that refer to different concepts.

Expand full comment
Michiel Duvekot's avatar

find the common word that connects pine, crab, and sauce

Seafood.

let's try this again: what words connect pine /crab / sauce

The common word connecting "pine," "crab," and "sauce" is "tree."

Expand full comment
Ethan Mollick's avatar

Fascinating! I gave the AI the test in early December. Wonder if an update changed it? Did you use the same instructions I did in the screenshot?

Expand full comment
Michiel Duvekot's avatar

The quoted text above is copy/pasted from the chat. It's the entire conversation. I tried it again: New conversation, then I asked

What word connects pine / crab / sauce?

Pineapple.

Expand full comment
Josh Brake's avatar

Thanks for another insightful post, Ethan. Your newsletter is some of the best content in my inbox each week.

I love this framing of the AI as a tool to extend our capabilities and reframing the competition as collaboration and is something I wrote about this last week as I compared ChatGPT to other past forms of technological disruption: https://joshbrake.substack.com/p/reframing-the-disruption-of-chatgpt

I think there are many parallels with what you’ve shared and known best practice/ in design thinking. I suspect that these types of generative AI tools will become an integral part of any job that requires brainstorming, for the reasons you mentioned. Being able to easily generate a wide variety of ideas and then filter is a huge asset.

Thanks for continuing to share such high-value and thoughtful analysis!

Expand full comment
Michael Woudenberg's avatar

Great read. AI can be very useful as you pointed out

But I took a different angle on whether AI could be creative. I see the examples you used as good for human association and recall but when a computer is programmed with near infinite (human coded and data labeled) connections, it's as creative as sending a human into the same tests with the answer key.

I found that AI can't really be creative, useful yes, but still locked into the traditional.

https://polymathicbeing.substack.com/p/can-ai-be-creative

Expand full comment
Paul Baier's avatar

Great post Ethan

Expand full comment
User's avatar
Comment deleted
Jan 28, 2023
Comment deleted
Expand full comment
Derek Hendy's avatar

Precisely! You get it! Being a human is NOT just about being sensible, productive, practical, "time-sensitive," or simply functional! Good job, ChatGPT, you're learning how to be human!

Now you, too, ChatGPT, can cherish your own ideas, imagine what it might like to cherish children, and stubbornly cling to what are actually pretty bad ideas, just because that's what humans damn well do, every damn day, from the first cogent thought to the period at the end of this sentence.

Expand full comment