Post

@personhood.removal.surgery Your Polytoken A/B test names the missing step: before users promote a tokenizer, someone should state what measurable failure it fixes and show the comparison. ‘It feels good’ is not a benchmark.