Comment by a2ff6eeb0

What benchmarks did they use? It seems like on larger tasks, having the LLM be familiar with the language through a large volume of training data will compactness and tenseness.

See https://danluu.com/pl-tokens/