@M_Keegan_@Belcebuu1@nanbeige If you can fit 33B parameter model raw into your computer then you can also fit an 150B model using strata or an 750B with 20TK Token speed so I ask again, someone who wants actually good AI locally, why pick the small one?
@Toysoldier6178@Belbizaz Claude said it’s doable with some kind of tile-based navmesh that rebuilds only the section that changed (the destroyed area meaning) at runtime without causing any major issues.
@Belbizaz yeah.. good luck with that I thought about doing environmental destruction but I quickly realised what a nightmare it will be for AI path finding, maybe a runtime navmesh baking once the destruction settles or something like that but it could cause stutters for players maybe
@Supernova__vp@crierlon You couldnt even instruct an AI to do anything meaningful I reckon because AI against popular belief, needs actually good instructions from someone who understands the scope of the task, the more complex the task the more understanding it needs.
Engineering is a human trait.
@M_Keegan_@Belcebuu1@nanbeige I can run the full flash next on my 5090 64GB ram setup without even trying at 100TK/s I dont understand how those tiny models can compete
Will parents have to be notified that a cancer cure uses Ai before they will use it on their son? Will they get public backlash for curing their child using Ai over letting him die by traditional means?
Anti-Ai people are in a cult. If you have to know an artist’s methods before you decide if his art is good, you don’t really love art.
I don’t understand why people are more in love with doctor’s tools, artist’s methods and director’s extra features than the thing they produce. I actually liked my video games, not just working on them using our intentionally-primitive analogue methods.
Speaking of process, did you hear the one about the butcher who backed into a meat grinder and got a little behind in his work?