@eva_khoury_ local models and hardware will continue to improve and likely get good enough for simple tasks. but they’ve been stifled by the big companies buying up all the hardware at crazy prices, limiting our access. the datacenters economies of scale is extremely hard to compete with.
@eva_khoury_ depends what your threshold of useful is. a high end consumer gpu can run models which can do quite a few useful things like basic coding, agentic stuff like organizing files and running ffmpeg commands ect. mid range consumer hardware is kinda on the boundary of useful.
🦠Watch @computerender's first public talk on Mote: an interactive ecosystem simulation!
Mote uses a custom GPU-based physics engine to model hundreds of thousands of organisms, leading to fascinating emergent phenomena: a sandbox that is part game, part research.
Link below⬇️
@KLieret Can you explain why in plots 1&3 the problems with <10% memorization chance have a zero percent pass rate? And also why would the pass rate would go down for some models when the chance of memorization increases?
@overand Thank you Geo! Mote is absolutely on its way to release. I apologize for being sparse on updates. There is a ton of work to do and there will be more updates. I appreciate your interest and your patience 🙏