@StocksOnSpaces The RL deepseek is based on is actually very computationally expensive. Whats more is that RL is not hindered by novel input data (we only have one internet etc), so the scaling is even more compute bound than traditional LLM training
@StocksOnSpaces You make interesting arguments on the spaces, but you are missing the key issue. It doesnt matter if the deepseek cost is true or not. It still scales with compute. Why settle at the current ai intell lvl? Its not like it its "too intelligent" currently
@Aries Looks amazing! As a developer it's gotten my attention. As its marketed with open source being a key feature, is there a repository to start getting acquainted with the code?
@RihardJarc A part of the ad revenue might go to seamless in-game ads, where ads are displayed as textures within the game world (on billboards and bus stops in a virtual city for example). A gaming experience can never be summerized by AI like search/factual pages can and will be.