owly
Banned
- Nov 26, 2020
- 179
I say the problem is that AI isn't narrow enough.
you have AI to play chess, go, tekken, generate articles.
the problem with over generalized AI is that with each task and capability one wants to add you also have to rethink your whole fucking model.
you made an awesome neural network to play super mario like chad plays with Beckys feelings. great !
but guess what buttercup
it turns out your NN sucks at playing tetris, it cant think 4th dimentionally or whatever or it just pauses the game cause it's smart like that.
awesome you have just beefed up your model to play supermario AND tetris, and it only took you 3 years, you post a little brag video
on foidTube and look at the few comments your tiny channel has, what's this ? some guy comments :
but can she play tekken ?
hmm lets check it, so you adapt your project to play tekken no biggy just 2 weeks to get the input output, heck maybe namco colaborates with you
and guess what ? the AI just waits for you to give up on playing by playing defensivemaxed, not her fault really tetris and tekken are too different
or she cant handle super combos
so back to the drawing board, another 4 years and some fucker now says : but can she play dual monsters ? can she play pokemon.
on the other hand you have a better option of breaking down what you want the AI to do into real small tasks.
this enables you to add skills form a ton of people, and cuts down on your codding time, and your algorithms are much simpler.
so now you have a menu site with skills for AI, you choose the skills she should have, 4 gaming skills in this case and add them with 4 lines of respective code.
moreover each skill takes much much less time to make, and you can actually collaborate with other codercels this way !
the more skills this design pattern has the closer we get to the singularity.
It's not like I'm even ruling out NN, heck the design pattern can absorb them as well.
you have AI to play chess, go, tekken, generate articles.
the problem with over generalized AI is that with each task and capability one wants to add you also have to rethink your whole fucking model.
you made an awesome neural network to play super mario like chad plays with Beckys feelings. great !
but guess what buttercupit turns out your NN sucks at playing tetris, it cant think 4th dimentionally or whatever or it just pauses the game cause it's smart like that.
awesome you have just beefed up your model to play supermario AND tetris, and it only took you 3 years, you post a little brag video
on foidTube and look at the few comments your tiny channel has, what's this ? some guy comments :
but can she play tekken ?
hmm lets check it, so you adapt your project to play tekken no biggy just 2 weeks to get the input output, heck maybe namco colaborates with you
and guess what ? the AI just waits for you to give up on playing by playing defensivemaxed, not her fault really tetris and tekken are too different
or she cant handle super combos
so back to the drawing board, another 4 years and some fucker now says : but can she play dual monsters ? can she play pokemon.
on the other hand you have a better option of breaking down what you want the AI to do into real small tasks.
this enables you to add skills form a ton of people, and cuts down on your codding time, and your algorithms are much simpler.
so now you have a menu site with skills for AI, you choose the skills she should have, 4 gaming skills in this case and add them with 4 lines of respective code.
moreover each skill takes much much less time to make, and you can actually collaborate with other codercels this way !
the more skills this design pattern has the closer we get to the singularity.
It's not like I'm even ruling out NN, heck the design pattern can absorb them as well.
