what. this is just some twitter guy. his status is 'crypto native'. he says > "The model (architecture, data) can be static, but the memoization and re-call re

what. this is just some twitter guy. his status is 'crypto native'. he says > "The model (architecture, data) can be static, but the memoization and re-call resources can affect model performance. If they are trying to scale GPT, its entirely possible this resource depth has changed, is dynamic, or otherwise augmented with GPT3 when needed." > "its entirely possible" I mean there was a huge hacker news thread the other day where people were speculating this same thing. How is it important that a twitter guy is also speculating it. Also I think some OpenAI employee tried to debunk it. Did the debunking get debunked? Ask HN: Is it just me or GPT-4's quality has significantly deteriorated lately? (https://news.ycombinator.com/item?id=36134249 https://news.ycombinator.com/item?id=36134249) OpenAI Employee: GPT-4 has been static since March (https://news.ycombinator.com/item?id=36155267 https://news.ycombinator.com/item?id=36155267)