I trapped an LLM into a Raspberry Pi and it spiraled into an existential crisis

[floppyd]

>Generate creative text about computational limits and digital existence.

So, it's more like "I ran a text predicting machine on a Raspberry Pi 5 and explicitly instructed it produce text cosplaying a sentient AI", which makes it a lot less interesting. And actually just running an LLM in a loop with basically empty prompts also produces "sci-fi" results, so not sure why this approach was chosen.

[lostmsu]

Hm, I'm curious and the "sci-fi" results. What were they?

[novok]

Yes the prompt also pre-seeds existential dread.

[pisanvs]

I hope to not offend "purists" by saying this is or isn't art, but it definitely made me reflect on how some people's existence, especially in their later years is not so far from this same essence...

LLMs "know" how it feels, or rather, reads to die. It definitely makes me feel quite uneasy, especially with the recent news of a teen's suicide aided by an LLM.

[allears]

I think you're granting too much ability to a machine that simply completes text in a statistically appropriate way. What you're really getting is a summary of everything anybody has ever said on the internet that's even tangentially related, output in a way that seems like a continuation of whatever was prompted.

This can seem profound in the same way that a horoscope can seem profound. Any discussion about "existence" will call up seemingly philosophical quotes.

[theearling]

So a website version of this?

https://hackaday.com/2025/05/24/ai-art-installation-swaps-di...

[countbassi]

Yea! The project was inspired by https://rootkid.me/works/latent-reflection

[xcavier]

I'm sure this would be interesting if it actually rendered on my device…

[countbassi]

Should be working now! My internet was down for a couple of days unfortunately

[xtiansimon]

Cute. Nice execution. Does it talk to itself? Knock Knock… sorta thing?

[nathan_compton]

LLMs are so banal.