How LLMs Learned to Reason: SFT --> RLHF --> RLVR
1. The Starting Line: The Last Non-Reasoning Flagships GPT-4.5, DeepSeek-V3, and Claude 3.5 Sonnet share something that has nothing to do with benchmark scores: they were the last major models built entirely on the "pretrain, then instruct-tune" recipe. All internal computation was done in one forward pass per token, with no backtracking, verification or revision mechanisms in place. By late…
The Last Non-Reasoning Flagships GPT-4.5, DeepSeek-V3, and Claude 3.5 Sonnet were the final major models constructed using the pretrain, then instruct-tune recipe. These models proceeded without any backtracking, verification, or revision mechanisms. As data and compute scaling reached a plateau, the focus shifted from larger models to a different training paradigm: training models to think before they answer.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
