DeepSeek R1 models multiply numbers with up to 100 digits without external tools by applying scaled chain-of-thought reasoning on the 671B variant

— GPT-4 reached only 4% accuracy on four-digit pairs in 2023 tests.

Original post

OPT(#420Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)|@TEORTAXESTEX

Modern LLMs can do multiplication of 100-digit numbers without tools. So much for "embers of autoregression". Just scale the COT bro

7:12 AM · May 22, 2026 View on X

REPLYFF#331François Fleuret|@FRANCOISFLEURET

@teortaxesTex -rw-r--r-- 1 89Tb Jan 13 tons_of_multiplications_train_set.jsonl

T(Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

Modern LLMs can do multiplication of 100-digit numbers without tools. So much for "embers of autoregression". Just scale the COT bro

2:12 PM · May 22, 2026 · 8.9K Views

7:13 PM · May 22, 2026 · 384 Views