Weekly journal 2025-01-08~2025-01-18
work, work, work
Reflection by >o1 Pro: "The year 2024 was a year in which the practice and development of digital democracy tools advanced quickly, and the seeds of "social implementation" grew through the TMG and multiple community activities. On the other hand, due to the fast pace of development, there were also noticeable situations where the "building a modest foundation" of information organization and health management was not able to keep up." We need to do this.
KJ method with AI, task management with AI, I've tried both and have a common problem awareness.
It is not good at handling "leaks" when left to the LLM context
Often some disappear
It's fundamentally unreliable to have a few tasks disappear at any given time in task management.
Therefore, we guarantee "not disappearing" in terms of software.
nishio Maybe "prioritizing" tasks is no longer the optimal solution, but having the LLM read through all the tasks and display about 3 in order of recommended to do now. (If the suggested ones can't be done now, give verbal feedback on that.) nishio I had AI create a "mechanism to let AI manage tasks" because AI was asked to create it. The trade-off was that a TODO list with one-line tasks loses too much context, and if you write detailed information so that it is not lost, the list becomes less comprehensive. nishio AI can wind up listing tasks and finding related ones, bracketing out components common to multiple tasks, and deciding what to do now, all from humans. AI can do all of this. The AI's context width is far wider than the human's vision. ---
This page is auto-translated from /nishio/週記2025-01-08~2025-01-18 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.