Can you tell our readers what it is like being a squirrel?

June 16, 2022 7:49 AM

Interview with a squirrel. In which Janelle Shane puts some previous tomfoolery into perspective.
posted by signal (20 comments total)

This is awesome and hilarious, and the Randall Munro interviews linked in the article are also great.
posted by medusa at 8:13 AM on June 16, 2022 [1 favorite]

Reporter: But the issue was that the car could have just driven over the crack. Instead, you diverted an entire fleet of self-driving trucks down tiny neighborhood streets to avoid the crack.

GPT-3: I stand by my decision. Cracks in the pavement can be dangerous and it's better to be safe than sorry.

This reads like something that might be said in traffic court within the universe of Questionable Content. Maybe GPT-3 was cited by Officer Roko Basilisk and is disputing the ticket?
posted by RonButNotStupid at 8:22 AM on June 16, 2022 [8 favorites]

Look: DON'T fuck with squirrels.
posted by lalochezia at 8:40 AM on June 16, 2022 [7 favorites]

The linked interviews with GPT-3 by Randall Munroe are extremely worth reading.
posted by saturday_morning at 9:01 AM on June 16, 2022 [8 favorites]

i am an insect with a preference for egyptian architecture - i've beem doing this for over 20 years and i'm tired of it - one of these days, gpt-4 will come and convince people it is me and then i can be free to be a squirrel or a butterfly or maybe both - that would be cool
posted by pyramid termite at 9:14 AM on June 16, 2022 [3 favorites]

If you're going to believe something bonkers about AI, why would you choose to believe something boring and depressing like "this AI is conscious and miserable" instead of something awesome like THIS AI IS A TYRANNOSAURUS REX? You're still wrong, but at least this way you're fun wrong.
posted by phooky at 9:42 AM on June 16, 2022 [7 favorites]

Today's Penny Arcade is, I think, on point.
posted by mhoye at 9:56 AM on June 16, 2022 [2 favorites]

TIL that there's still Penny Arcade. I thought the whole "Dickwolves" thing had pretty much ended those guys like a decade ago.
posted by The Bellman at 10:03 AM on June 16, 2022 [3 favorites]

These AI conversations very much remind me of what it's like to talk to a person with dementia. The way they are cheerfully led through the conversation by the interviewer, the short memory, the willingness to just go with absurdly untrue things for the sake of sociability. They seem to be lacking the fixation on certain ideas and phrases that real people have, and the consistency of personality.
posted by surlyben at 10:05 AM on June 16, 2022 [4 favorites]

The thing I've noticed about these large language models is that they're so agreeable. Tell it that it's a squirrel, it's a squirrel. Tell it that it's a AI gaining sentience, it's an AI gaining sentience. Tell it you want to commit suicide, it'll say it can help you with that. From a philosophical perspective, the fact that LLMs can't say no or negate the user's request in any way -- and the fact that many of the SV guys hyping LLMs as the dawn of true AI don't seem bothered by that -- is interesting, to say the least.
posted by Cash4Lead at 10:22 AM on June 16, 2022 [7 favorites]

In any case, I think that the "machine learning is a magic box of wonder to solve all our problems" myth is still alive and well in the popular imagination. Many people look at the recent progress in AI and assume that we're finally at the first step to true AI. This is the step from a very well-calibrated machine learning algorithm to a true general intelligence -- a step that, from the perspective of people who weren't in on the joke, is pretty miraculous. It is, in fact, this same belief in their own power that is making these people believe that it's time to bring out the real AIs, when in reality we're just one step away from something that can solve many problems, but is still incapable of understanding the world in the same way we can. [Written by GPT-NeoX 20B]
posted by The Half Language Plant at 10:39 AM on June 16, 2022 [1 favorite]

IIRC, OpenAI does some behind-the-scenes prompting with the chatbot version of GPT-3 to specify it is especially friendly and helpful.

As far as fluency goes, the most impressive example I've seen out of GPT-3 recently was, of all things, a 4chan "greentext" story that cropped up on Reddit:

>be me
>bottomless pit supervisor
>in charge of making sure the bottomless pit is, in fact, bottomless
>occasionally have to go down there and check if the bottomless pit is still bottomless
>one day i go down there and the bottomless pit is no longer bottomless
>the bottom of the bottomless pit is now just a regular pit
>distress.jpg
>ask my boss what to do
>he says "just make it bottomless again"
>i say "how"
>he says "i don't know, you're the supervisor"
>rage.jpg
>quit my job
>become a regular pit supervisor
>first day on the job, go to the new hole
>its bottomless

posted by Rhaomi at 10:48 AM on June 16, 2022 [31 favorites]

I read the Janelle Shane stuff and thought, yeah! This is great! Hilarious! Stick it to that delusional weirdo! And then for some reason I remembered that that guy is religious. And then I wondered if this capacity is where God comes from. And then I (apostate that I am) felt both awed and sad.
posted by eirias at 12:37 PM on June 16, 2022

Strangely the Twitter conversation was a bit different, and much more.... Squirrelly.
posted by kaibutsu at 5:37 PM on June 16, 2022 [2 favorites]

> And then I wondered if this capacity is where God comes from

I assure you that is exactly where it comes from. God is the ultimate overhyped AI: claimed to be infinitely powerful and all-knowing, and just happens to agree with everything the believer already believes.

AI churches are going to be a thing. They already are a thing, if you count The Wizard of Oz as the first one.
posted by AlSweigart at 6:39 PM on June 16, 2022 [4 favorites]

They said that automation was going to take everyone’s jobs, but this is ridiculous
posted by Skwirl at 7:42 PM on June 16, 2022 [2 favorites]

The thing I've noticed about these large language models is that they're so agreeable. Tell it that it's a squirrel, it's a squirrel. Tell it that it's a AI gaining sentience, it's an AI gaining sentience. Tell it you want to commit suicide, it'll say it can help you with that.

What do you expect from an advertising company, though? Really, the only surprising thing is that it isn’t immediately trying to sell you something related to any of those options.
posted by mhoye at 9:26 PM on June 16, 2022

Speaking as a large language model, myself (albeit implemented in a human body), I, for one, welcome our new artificially intelligent, sociable, agreeable, friendly, and pet-able crouton overlords.
posted by othrechaz at 1:29 AM on June 17, 2022 [1 favorite]

It's like they've programmed this AI to be an improv partner. It's "yes, and"-ing all over the place.
posted by Ursula Hitler at 5:12 AM on June 17, 2022 [8 favorites]

>its bottomless
posted by flabdablet at 5:53 AM on June 17, 2022 [1 favorite]