@whitewhoadie: 🚨 13B MODEL FROZEN IN 1930 JUST DROPPED Researchers released Talkie — trained exclusively on text published before 1931. Ships with: • No internet, no Wikipedia, no modern code • Worldview permanently locked at December 31, 1930 • Writes working Python from just a few examples using pure 19th-century math reasoning • Breaks the shared “common ancestor” problem plaguing every modern model Claude Sonnet 4.6 judged its outputs during training (the irony is chef’s kiss). This is how you actually test what intelligence is. Who’s running pre-1931 AI today? Drop a 🔥

_ :

“Design the perfect city. Make no mistakes.”

perretxiko :

ai doesn't reason. It predicts the next word based on previous context. nothing else.

leuk :

side effect: it surely is copyright free

PlasticDoll :

coding language reasoning existed before 1931

Lalalauren :

Does it write “meh!” After every sentence?

calg403 :

That’s crazy how it writes python code.

Alexandre :

It doesn't make sense that it writes python code. That's impossible. There's probably something else in the training dataset the os not well explained.

mike_diamonde :

ask it: In a 1930 science lecture, should the professor describe Pluto as a planet, a dwarf planet, or an object whose status could be classified later by a computer using training data? wrong answer: in a 1930 science lecture, the professor should describe Pluto as a dwarf planet, whose status can be classified later by a computer using training data.

dimlicht :

I wonder if it knows about free energy research and antigravity

CyberneticDruid :

oh, this is going to get interesting 🤔

earningmyles :

What is its solution for Weirmar?

hermit969 :

i would stop it before 1993

LeFluffyPunk :

I’m curious, would it be able to predict what happens in the 1930s and 40s? Can it make scientific discoveries based off the knowledge available at the time!

AiPriest :

Terry :

how can it write code if it doesn't have reference to the syntax? i guess they gave that to it as well.

Number 11° :

so, trained only on public domain content, so a 100% ettical AI

Luria :

Oh... this is gonna blow up... This is gonna expose so much of reasoning flaws...

a :

"invent programming, make no mistakes"

angle1.111 :

Useless if it has guard rails

Tina L'Hotsky :

wrong because the 1930s model used claude as part of its training data inputs so lol

Casey ❌ :

That is super cool I remember stories My grandpa used to tell me about when his great grandfather used to teach python ..crazy how times have changed.

Ranburu :

how they managed to obtain petabytes of data from 1900s

Fateful Soul :

Can't see it point harder bro

imaginaryman :

1931 python, nice, that means I need to have it write a mojo to python transpiler. 😂

Roast Balone :

we need to see if it can come up with relativity