AI Model Fragmentation Requires Larger Context Windows, Raising Compute Costs
Commentary on X
Highest ranked
@nikesharora Models are tied to language. Coding agent Research agent Quant agent Simulation agent I think model fungibility is not needed across layers, but only at each layer. This opens up possibilities, where labs and enterprises can actually be symbiotic.
@nikesharora True more we switch between models more context we need to maintain. This is evident even in Claude or ChatGPT chats, when we switch model it feels like new chat.
Think simplistically the more you want to move across models the larger your context window needs - perhaps less efficient from a compute and latency perspective. https://twitter.com/SProfiler1/status/2074764968495345858
Think simplistically the more you want to move across models the larger tour context window needs - perhaps less efficient from a compute and latency perspective. https://twitter.com/SProfiler1/status/2074764968495345858