On the D Blog: Language Plasticity is More Important Than Ever

[Adam Wilson]

On Monday, 10 August 2026 at 13:41:59 UTC, monkyyy wrote:
> If that where the numbers come from, they are suggesting 
> deleting whitespace; grouping 4-spaces as 1 token is a "gpt-2" 
> level technology.

There has been a lot of work done at the harness and ingestion 
layers to reduce input token usage precisely because 
whitespace-sensitive languages are so token hungry. So yes, 
they've gotten the input token usage down, but by improving 
tooling, not because the language itself is less token expensive. 
And there is evidence, albeit in early stages, that such schemes 
are increasing the amount of mistakes that the LLM makes on 
output and thus creating more iterations over the problem. Which 
increases total token usage.