Check Out This Interesting Post on X by @kregenrek

Cool. Will try. My fear is that at 1.7M tokens it's just too big to fit in Claude's context window.

Yeah, that's likely the case... I heard recently that OpenAI is working on a trillion token context window? 🤯

These big context windows are useless IMO The models need to work on their attention again. If the context window is big enough even with todays limit, every model makes a ton of mistakes. Claude Sonnet is from my experience the best at attention, and even better than the o models.

Jan They're not useless, but it takes a lot of work to get them to be useful. Not a trivial task.

Matt Mireles Useless might be a harsh word, but if the LLM forgets everything, there is no use in cramming more into its memory, like a glass that is running over. IMO the utility would rise a lot more if they'd manage to meaningfully increase the attention capacity for existing context windows.

Agreed. I am working on the precursor to this, fyi.

heh... precursor.ai?

I know! Cant' wait to see what you're cooking 🧑‍🍳

hmm.

I got missingmanual.ai

nice