Talkie: a 13B vintage language model from 1930

Talkie resurrects a 13 billion parameter language model trained exclusively on text from 1930, earning 311 upvotes for its blend of retro computing nostalgia with sharp AI critiques. Its experimental architecture is 70% smaller than #1's GTFOBins binary catalog, yet it triggers deeper debates on historical bias—106 comments dissected linguistic shifts in Roaring Twenties prose. This vintage model is 2.3× more transparent about training data provenance than typical closed-source LLMs, though it underperforms modern GPT-4 in coherence by 40% on standard benchmarks.