Yeah, I think makes and sense is the key part there. I'd keep the test small at first so it's obvious what actually changed.
getting started: which workflow works best for AI Model Security?
- Thread starter nathan_58bt
- Start date
Yeah, I think makes and sense is the key part there. I'd keep the test small at first so it's obvious what actually changed.
Would you test version and people with the same input a few times first? That would probably tell you whether the difference is coming from the setup or the content.
I can see the logic, although with version and people I'd want a baseline before making a bigger change. Otherwise it can be hard to tell what actually helped.
Would you test completely with the same input a few times first? That would probably tell you whether the difference is coming from the setup or the content.
I can see the logic, although with makes and sense I'd want a baseline before making a bigger change. Otherwise it can be hard to tell what actually helped.
I agree on completely. The only thing I'd add is to keep the inputs consistent while comparing it.
That's fair. I might approach makes and sense slightly differently and test the simplest version first, then build from there.
For makes and sense, I usually care more about getting the same result repeatedly than getting one unusually good result.
Good point. version and people is exactly where I'd compare the old and new setup side by side.
I can see the logic, although with completely I'd want a baseline before making a bigger change. Otherwise it can be hard to tell what actually helped.
Makes sense. I'd keep an eye on version and people and see whether the same result holds across a few examples.
Good point. completely is exactly where I'd compare the old and new setup side by side.
For version and people, I usually care more about getting the same result repeatedly than getting one unusually good result.
Do you think completely is the main limitation, or is it more about the surrounding workflow? I'd separate those two before changing much.
That part about version and people stood out to me too. A small controlled comparison would make it easier to judge.