After transcribing the entire book and feeding it to GPT-3, I was able to get this response:

Question: What happened to the rabbit? Answer: The bear caught him and ate him.

@PeterHurford Well, yes, but honestly there are already AI (CLIP, GPT-3) that could do it. It just requires someone to actually program something to stitch them together...

@warrenkrettek Just completed the challenge (kinda). I transcribed the book fully (basically making the picture book work for screen readers) and then fed the text into GPT-3. It answered the correct answer after 1 try :)

@chudetz Question: How do you know? Answer: Because the bear is wearing the rabbit's hat.

@chudetz

I personally don't believe I resolved it, I was just testing out if GPT-3 could even interpret it. Still very interesting. I'll attempt to patch together some CLIP + GPT-3 system that can get the images and the text. Time to attempt this again :)

@william_s

I honestly don't believe it remembers that, but I'll check.

"Q: At the end of I Want My Hat Back, what happened to the rabbit? A: The rabbit was eaten by a fox."

@notany Also, for the second question,

John has 50 apples. John's father ate 40 apples. How old is John?

John is 10 years old.

GPT-3 is basically a child...

@(notany I tried. It says "the wolf ate the rabbit". In the book they never say the word "bear" ever.

@casens

After transcribing the entire book and feeding it to GPT-3, I was able to get this response:

Question: What happened to the rabbit? Answer: The bear caught him and ate him.

Also, CLIP can label images. It was released with DALL-E. https://openai.com/blog/clip/

Honestly, it's amazing.

@1point7point4 Well, considering I was able to do it after transcribing the entire book and feeding it to GPT-3 then I got this response:

Question: What happened to the rabbit? Answer: The bear caught him and ate him.

@chudetz Also, if you want to contact me, I'm @Develotec on Twitter.

(BTW: you should put your documentary on YouTube! More people would access it and watch it!)

@ThirdEyeOpen It's impossible to infer it without the images. The images give vital context.

https://gist.github.com/Devetec/6a78b44eb877f…

In the transcript it's obvious why

@NunoSempere Eleuther isn't really a competitor, as they aren't a business. They just release the models for people to use, and maybe some competitor will use those models to create their own API. An actual competitor would be AI21 Labs, those people are going places.

I think that this should be increased to the right side much more. Just making a regression graph over the last few years, it would mean that 2021, 2022, 2023, 2024, 2025, and 2026 combined would have 30300 papers. 300 papers on the low end is basically impossible, considering we are merging the amount of papers over 6 years. Please fix this, and clarify whether it's the average amount per year over that 6 years or the total 6 years combined.