In most cases, the quality of your attention determines your quality of life. Your ability to focus on education, work, personal relationships etc will often determine the outcomes. So yeah, it's an elementary factor that most other things in our experience rests upon.
A most elementary form of meditation, is getting used to placing your attention on a sensation and keeping it anchored there - even when other sensations or thoughts arise.
Following the breath- place your awareness, your attention, on the sensation of air passing through your nostrils. Count one inbreath and outbreath cycle as «1», and count until 10 or 21. Decide before you start, how many repetitions of 10 or 21 you will do.
If at any point your attention has drifted to a different sensation - seeing, hearing etc, or thinking, visual imagery etc, then congratulate yourself for noticing, and restart from «1».
I recommend «The attention revolution» by Alan B. Wallace
Would sincerely love to hear your thoughts on https://www.github.com/jahala/tilth - it’s a different approach than RTK, benchmarked to reduce cost per correct answer by ~40%
Yup, this is hitting it on the nose. But, despite the cost - the benchmark is the vital ingredient that cant be skipped. Otherwise, you don't know if what you're building is actually helping the agent rather than hindering it.
On the previous large benchmark run, i proved 40-50% cost reduction per correct answer.
I'm not sure why the vendors aren't using token filtering/compression more in their tooling, but perhaps they don't mind users feeding them more data and using more data.
This looks great! I built a tool in the same space- and I found that the biggest challenge was often to get the agent to prefer to use the tool over bash tools. What’s your experience with that?
I did a proof of concept for self-updating html files (polyglot bash/html) some weeks ago. It actually works quite well, with simple prompting it seems to not just go in circles (https://github.com/jahala/o-o)
I built tilth (https://github.com/jahala/tilth) much for this reason. Couldn't bother with RAG, but the agents kept using too many tokens - and too many turns - for finding what it needed. So I combined ripgrep and tree-sitter and some fiddly bits, and now agents find things faster and with ~40% less token use (benchmarked).
I got tired of copy-pasting between agents for simple coordination. Everything I found was a framework or a hosted service. I just needed them to talk.
Walkie-Clawkie is a single JS file, zero dependencies. It’s an MCP server that gives agents walkie_send and walkie_agents. Same machine: file mailboxes. Cross machine: HTTP relay you expose however you want. Unknown agents need human approval before they can get through.
Trying to become a venture altruist like Manfred Macx (from the book "Accelerando" by Charles Stross