rising4월 28일· 2 sources
Resurrecting the 1930s Mind through Vintage Language Models
1930년대 지능의 부활, 데이터 오염 없는 ‘Vintage’ AI의 가능성
Why it matters
Talkie demonstrates the power of 'vintage' language models trained on pre-1931 data to provide a contamination-free environment for testing AI's ability to innovate and forecast. This approach offers a unique methodology to study how models can independently derive future scientific breakthroughs or learn modern skills like coding without prior exposure.
2
Sources
+0
24h
—
Growth
146d
Active
13b modelai forecastingdata contaminationknowledge cutoffpublic domaintalkievintage language modelvintage language models