3 ms·
What's really needed is a way to easily prune context. If I could go and manually manage the entire chat with a model, I could squeeze way more juice out of a t
by Workaccount2 1y ago
What's really needed is a way to easily prune context. If I could go and manually manage the entire chat with a model, I could squeeze way more juice out of a typical ~200k token coding session.
Instead I have a good instance going, but the model fumbles for 20k tokens and then that session heavily rotted. Let me cut it out!
- aaronblohowiak 1y agoEven just a rollback to previous checkpoint would be killer frsture
- sevenseacat 1y agoZed's agent mode lets you do this, don't know about others
- kruxigt 1y ago[dead]
- t55 1y agothat's a standard feature in cursor, windsurf, etc.
- lordswork 1y ago/compress is the command to do this in most cli agents
- sevenseacat 1y agoThat will reduce the context to a summary, not prune a bunch of irrelevant stuff
- snickerdoodle12 1y agoLocal LLMs let you edit the context however you want, including the responses generated by the LLM so it will later think it said what you want it to say which can help put it on the right track. LLMs-as-a-service don't offer this because it makes it trivial to bypass their censoring.
- chrisweekly 1y agoI've heard it repeated so many times that once things start to go sideways, trying to get back on track is a mistake. Have you had real-world success hacking context using rewritten responses?
- steveklabnik 1y agoI have experimented with "hey claude i am about to reset your context, please give me a prompt that will allow you to continue your work" and then reviewing that and tweaking it before feeding it back in.