Query: Data Exhaustion by adlrocha

Loading chart...

1-- CHART 2: Data Exhaustion Timeline
2SELECT 
3    year,
4    model_example,
5    dataset_size_tokens_billion,
6    total_available_stock_tokens_billion,
7    source_type
8FROM "@adlrocha.llm_stats.ai_training_data_exhaustion"
9ORDER BY year

year	model_example	dataset_size_tokens_billion	total_available_stock_tokens_billion	source_type
2018	GPT-1	1	300000	Historical actual
2019	GPT-2	4	300000	Historical actual
2020	GPT-3	374	300000	Historical actual
2021	Gopher	300	300000	Historical actual
2022	Chinchilla	1400	300000	Historical actual
2022.3	GPT-3.5	4000	300000	Historical actual
2023	GPT-4	13000	300000	Historical actual
2023.5	PaLM-2	4000	300000	Historical actual
2024	Llama 4 Scout	40000	300000	Actual from Meta
2025	Projected	75000	300000	Epoch AI projection
2026	Projected	150000	300000	Epoch AI projection
2027	Projected	250000	300000	Epoch AI projection exceeds stock
2028	DATA WALL	300000	300000	Epoch AI median exhaustion date