1,686 karma · joined January 4, 2021
but the implementation will be up to your provider and harness, for deepseek, they expose some numbers: https://api-docs.deepseek.com/guides/kv_cache/ and Anthropic has a list of actions invalidating your cache: https://platform.claude.com/docs/en/build-with-claude/prompt...
Basically, you avoid anything dynamic: model change, tool change, etc it's also important that your system prompt or main prompt doesn't have non-static data like the date/time/place or someone's name (the person you interact with in a chatbot for example). That should be left to tool call or search.
Just want to say that:
>In our deployment, student-reported reading completion baselines for MATH 010 were approximately 15%, with instructors estimating 10%. Individual student reports of reading compliance ranged from "literally no one does that" to "is this being recorded?"
is hilarious
A cool person making cool things!
Some people just don't have the experience you do, "its proponent keep praising it in denial", can we have a better level of debate, come on now.
I do take the train quite often as I said, anything on large axes is usually fine (Brussels - Charleroi, Brussels - Antwerp, etc) but yeah smaller lines are usually struggling some more.
I wish we had more ambitious governments in general, not only in terms of energy but also in the (bio)tech scene, which used to be touted as our great strength (we do have a lot of pharma companies though).
Absolutely insane take imo. You do you man.
Hopefully the current energy crisis is a wake up call.
The very good backwards compatibility is attractive but as the result of the small community, there's also a lot of abandoned packages and fewer QoL packages (formatters, linters, etc); I know there are some but for example I had setup `cljfmt` in Emacs and it wouldn't work, didn't look further.