2 ms·
To be fair Codex was much better than GPT-3 on reasoning benchmarks like MMLU and the like. And people have kind of noticed the Code trained models to reason be
by famouswaffles 3y ago
To be fair Codex was much better than GPT-3 on reasoning benchmarks like MMLU and the like. And people have kind of noticed the Code trained models to reason better. Don't know if a paper was published about that though.