5 ms·

I think it's somewhat interesting that codex (gpt-5.3-codex xhigh), given the exact same prompt, came up with a very similar result. https://3e.org/private/sel

by dmd 8mo ago

I think it's somewhat interesting that codex (gpt-5.3-codex xhigh), given the exact same prompt, came up with a very similar result. https://3e.org/private/self-portrait-plotter.svg https://3e.org/private/self-portrait-plotter.svg

futurecat 8mo ago
good stuff, thank you for sharing!
plagiarist 8mo ago
I love that these would be perfectly at home as sigils in some horror genre franchise.
- moonlion_eth 8mo ago
  its just reality
  - plagiarist 8mo ago
    My least favorite horror genre franchise.
geoelectric 8mo ago
"Doesn't look like anything to me"
majormajor 8mo ago
AFAIK all of these models have been trained in very similar ways, on very similar corpuses. They could be heavily influenced by the same literature. I wonder if anyone recognizes it really closely. The Pale Fire quote below is similar but not really the same.
voxl 8mo ago
Are you crazy or am I because I scrolled through that blog and am left scratching my head at you and your claim.
manuelmoreale 8mo ago
Asked gemini the same question and it produced a similar-ish image: https://manuelmoreale.dev/hn/gemini_1.svg https://manuelmoreale.dev/hn/gemini_1.svg When I removed the plot part and simply asked to generate an SVG it basically created a fancy version of the Gemini logo: https://manuelmoreale.dev/hn/gemini_2.svg https://manuelmoreale.dev/hn/gemini_2.svg This is honestly all quite uninteresting to me. The most interesting part is that the various tools all create a similar illustration though.
- alex43578 8mo ago
  Is it? They're all generalizing from a pretty similar pool of text, and especially for the idea of a "helpful, harmless, knowledgeable virtual assistant", I think you'd end up in the same latent design space. Encompassing, friendly, radiant. Note that Claude, ChatGPT, Perplexity, and other LLM companies (assumably human) designers chose a similar style for their app icon: a vaguely starburst or asterisk shaped pop of lines.
  - manuelmoreale 8mo ago
    > Is it? They're all generalizing from a pretty similar pool of text, and especially for the idea of a "helpful, harmless, knowledgeable virtual assistant", I think you'd end up in the same latent design space. Encompassing, friendly, radiant. Oh yeah I totally agree with that. What I was referring to was the fact that even though are different companies trying to build "different" products, the output is very similar which suggests that they're not all that different after all.
    - alex43578 8mo ago
      To massively oversimplify, they are all boxes that predict the next token based on material they’ve seen before + human training for desirable responses. You’d have to have a very poorly RLHF’d model (or a very weird system prompt) for it to draw you a Terminator, pastoral scene, or pelican riding a bicycle as its self image :) I think that’s what made Grok’s Mechahitler glitch interesting: it showed how astray the model can run if you mess with things.
kleene_op 8mo ago
Spirals again. Those AIs have read too much Junji Ito.
layer8 8mo ago
It’s a bit closer to the Flying Spaghetti Monster.

by dmd 8mo ago

futurecat 8mo ago

good stuff, thank you for sharing!

plagiarist 8mo ago

I love that these would be perfectly at home as sigils in some horror genre franchise.

moonlion_eth 8mo ago
its just reality
- plagiarist 8mo ago
  My least favorite horror genre franchise.

geoelectric 8mo ago

"Doesn't look like anything to me"

majormajor 8mo ago

AFAIK all of these models have been trained in very similar ways, on very similar corpuses. They could be heavily influenced by the same literature. I wonder if anyone recognizes it really closely. The Pale Fire quote below is similar but not really the same.

voxl 8mo ago

Are you crazy or am I because I scrolled through that blog and am left scratching my head at you and your claim.

manuelmoreale 8mo ago

Asked gemini the same question and it produced a similar-ish image: https://manuelmoreale.dev/hn/gemini_1.svg https://manuelmoreale.dev/hn/gemini_1.svg When I removed the plot part and simply asked to generate an SVG it basically created a fancy version of the Gemini logo: https://manuelmoreale.dev/hn/gemini_2.svg https://manuelmoreale.dev/hn/gemini_2.svg This is honestly all quite uninteresting to me. The most interesting part is that the various tools all create a similar illustration though.

alex43578 8mo ago
Is it? They're all generalizing from a pretty similar pool of text, and especially for the idea of a "helpful, harmless, knowledgeable virtual assistant", I think you'd end up in the same latent design space. Encompassing, friendly, radiant. Note that Claude, ChatGPT, Perplexity, and other LLM companies (assumably human) designers chose a similar style for their app icon: a vaguely starburst or asterisk shaped pop of lines.
- manuelmoreale 8mo ago
  > Is it? They're all generalizing from a pretty similar pool of text, and especially for the idea of a "helpful, harmless, knowledgeable virtual assistant", I think you'd end up in the same latent design space. Encompassing, friendly, radiant. Oh yeah I totally agree with that. What I was referring to was the fact that even though are different companies trying to build "different" products, the output is very similar which suggests that they're not all that different after all.
  - alex43578 8mo ago
    To massively oversimplify, they are all boxes that predict the next token based on material they’ve seen before + human training for desirable responses. You’d have to have a very poorly RLHF’d model (or a very weird system prompt) for it to draw you a Terminator, pastoral scene, or pelican riding a bicycle as its self image :) I think that’s what made Grok’s Mechahitler glitch interesting: it showed how astray the model can run if you mess with things.

kleene_op 8mo ago

Spirals again. Those AIs have read too much Junji Ito.

layer8 8mo ago

It’s a bit closer to the Flying Spaghetti Monster.