Luis

Is there a trick to fix this kind of pixel smearing when using AA? Looks great with AA off, but with it on the stars smear like craaaazy. Can I compute custom velocity vectors or something? It's just a basic (unlit) material sampling a skybox texture to give that "portal" feel

level designer AI is easily a $100M opportunity waiting to be unlocked there is so much depth in designing game worlds, even a 3D blockout designer ai would be printing money like crazy some thoughts about what agents should be good at: 1. Clear Strategic Progression AI that helps in building the core experience loops as well as the big loop of intro > buildup > climax > resolution 2. Spatial Design Generating or at least reviewing shapes, landmarks, lighting, colors and cues for coherence and clarity. 3. Pacing Create series of high intensity moments with breaks of relaxation 4. Reward Generate meaningful challenges which feel rewarding. The challenge, depending on the progression in the game, should be just difficult enough for players to feel like they're much more skilled after achieving it. Progression should incorporate new skills after they're unlocked 5. Show, not tell AI should be capable of constructing scenes where the objective is clear from the design of the level using all of the above, without relying on verbose explanations. Current LLMs are very bad at these skills when tasked to create games. They are highly finetuned to mimic known gameplay mechanics. It's not clear to me how this can be done as all points are very tightly coupled to gameplay, so it needs to be able to own and understand the mechanics to be able to build out levels and tell a story. I still think this is very much possible to solve, it needs the right benchmarks for ai agents to train on and open minded experts in these fields to use them and find limitations

Akira Toriyama manga art comes to life as an animation and scored by Hans Zimmer for the Dragon Ball 40th anniversary!