In case anyone was wondering, re. subclasses

randconfig · April 20

Wereswan wrote: »

dk_dunkirk wrote: »

Wereswan wrote: »

And I just realized my math was way off, because it's multiplicative, so it's really 378 possible one-class combinations, and 5,670 possible two-class combinations.

Nevermind, nobody's testing all that.

Are we just throwing AI at everything now? Can't we let ChatGPT play through all of these combos and give us the breakdown?

I'm starting to suspect that this is more-or-less what actually happened: they put it through machine learning, the robot told them it was fine, and now they're being blindsided by the 170K+ frankenstein builds they've enabled.

That's not how machine learning works... AI does not just mean chatGPT, many forms of AI have existed long before LLMs (large language models), like chatGPT, were created and shoved in our face with every single online service. LLMs are more likened to an advanced google search, which is different from doing data analysis with machine learning.

Machine learning in this case would be used to perform a data analysis to find outliers, meaning combinations of abilities that are either way better or way worse than the average, and then to simulate changes to existing skill lines, perform a data analysis on that simulation, and compare it to the original data analysis to see if it eliminated or reduced the severity of the outliers for balance. I think that's how it works anyway, been a while since I've done any AI stuff.

loosej · April 20

Wereswan wrote: »

dk_dunkirk wrote: »

Wereswan wrote: »

And I just realized my math was way off, because it's multiplicative, so it's really 378 possible one-class combinations, and 5,670 possible two-class combinations.

Nevermind, nobody's testing all that.

Are we just throwing AI at everything now? Can't we let ChatGPT play through all of these combos and give us the breakdown?

I'm starting to suspect that this is more-or-less what actually happened: they put it through machine learning, the robot told them it was fine, and now they're being blindsided by the 170K+ frankenstein builds they've enabled.

nbksaske · April 20

I'm looking forward to being able to test it out in PvP, my Templar having access to streak will be great 😁

Maxitude · May 29

Wereswan wrote: »

And I just realized my math was way off, because it's multiplicative, so it's really 378 possible one-class combinations, and 5,670 possible two-class combinations.

Nevermind, nobody's testing all that.

************

I think the math might be different. If we assume that the main class selected has NO effect on the behavior of the subclasses, then we simply have a pool of skill lines to choose from regardless of the main class. That pool is 7 classes times 3 skill lines equal 21 skill lines to choose from.

Our job is to choose 3 from that pool of 21. With this assumption we have a combinatorial math problem. If so, we have a 21!3 (21 choose 3) calculation, where order doesn't matter. The important point is that order doesn't matter. We don't care that whether we start with a Templar skill and then choose an Arcanist skill followed by Warden skill, rather than starting with the Warden and then choose the same Templar and Arcanist skills. (If we find at release that ZOS has made the main class skill change the behavior of subclass skills, then this math won't apply.)

You can calculate this math problem using any combinatorics calculator (possibly even on your cellphone). The solution to 21!3 = 1330 different skill combinations.