Oh well. I might be to picky here, but how I see things, determinism cannot be improved or worsened, but achieved or not achieved. Or Partially archieved, when analyzing a system that has both components that are deterministic or non deterministic.
There are times to think in absolutes, and when talking about deterministic behavior of technical systems, this is one of them. Join the sith side, we have cookies, and when we say we have cookies, we do.
Yes! The "-ness" implies some degree of spectrum already, and from what I know and understand, one can talk about different degrees/classes of randomness. "decreasing entropy" would work too, I guess (Intuitively: If one would open a bet on a system outcome, how high would the quotas be? If there is some degree of randomness in the system, it would make sense to put money on multiple different outcomes to minimize ones losses. Only if there is no randomness at all, it would be rational to bet all money on the one option we expect).
Another further distinction would be between "true randomness" and "We do not know all input factors yet, hence, the outcome appears random to us". The former is something that genuinely exists in the universe as far as I understand physics (i.e. even if all input parameters are known, it is not possible to predict the outcome reliably, which seems to be the case for quantum physics, which I have the surfacest or surface levels of understandings about), the later boils down to reducing "unknown unkowns".
In practise, this means stuff like using seeds for random number generators, fixing things that look random like day of the week of a DateTime.now(), and so on. Depending on what you do, it might be never possible to get rid of all randomness, but I am 100% convinced that you can always contain it into a well-defined cage, and control everything else down to the last bit.
That also made me think of the opposite: "Increasing predictability" would be acceptable for me too, and a deterministic process/component/system/whatever is one that is at maximum possible predictability for me then.
In all seriousness, this seems like it could be fine if done well. You can just have a model do a pass over the system prompt and set reasonable parameters based on that. That's probably not what they're doing, but it could be.
Oh well. I might be to picky here, but how I see things, determinism cannot be improved or worsened, but achieved or not achieved. Or Partially archieved, when analyzing a system that has both components that are deterministic or non deterministic.
There are times to think in absolutes, and when talking about deterministic behavior of technical systems, this is one of them. Join the sith side, we have cookies, and when we say we have cookies, we do.