Writing variants
RawReply gives you actually different angles, not three paraphrases of the same safe draft. Use it when the idea is right but you need the version that fits the room.
Direct
Before
I would like to communicate that our current evaluation approach may not be measuring the right things.
After
Most AI evals miss because they grade output quality with the same family of model that wrote the output. Cousins judging cousins.
Thoughtful
Before
Please make this sound more nuanced while preserving the concern about flawed evaluation methods.
After
There is a deeper issue with most AI evals: we benchmark models against the rubric the models were trained to satisfy. The blind spots are systemic, not statistical.
Sharper
Before
Make this punchier for a post, but do not make it sound like engagement bait.
After
Most evals are theater. The judge is from the same model family. The rubric was written by a model. Of course everything scores 4.6.
Variants are different ways to express the same idea: more direct, more thoughtful, sharper, shorter, or more specific to the audience.
Many tools return three lightly reworded versions. RawReply is designed to make the angle different, so you can choose the read that fits the moment.
Use variants for posts, proposals, emails, replies, and one-pagers when the core idea is there but the first draft does not quite land.