Weekly journal 2025-03-16~2025-03-24
1. Task Priority
.
Highest Priority: Tax Returns (T0109)
- Deadline is 3/17 so need to start now.
- Ensure that tax payments and document submissions are completed before other tasks, as there are penalties for late submissions.
example of a runner-up critical task
.
- Beta release of AI Task Management System WebUI.
- The need is high in the context of "response to increasing information channels (related to T0116/T0113)" and "productivity improvement through the use of AI."
- Break down tasks into smaller pieces and proceed from priority areas.
- AIPM2025-03-11のつづき
- Related to broad listening AI (Yasuno team's new broad listening system)
- Since the intention is to complete the in-house case study demo by the May trilogy (Audrey Tang et al.), it would be very effective if the processing pipeline of data for in-house use is prioritized.
- Monday, this is the priority.
Example of a task to be started in the short term
.
- Develop Slack thread extraction tool (T0128).
- Early prototyping as a mechanism for integrating multiple channels will be useful for subsequent "information channel organization".
- We need this. It's not like we're going to make it, it's already out there in the world to begin with.
- TTTC analysis of MVID dataset (T0008) → TTTC configuration file support tool (T0009)
- If we proceed in the context of "making an impact by May," we can show specific examples of analysis and reporting using TTTC within the company.
- A slight change of policy may be necessary.
- Nondestructive Scan & OCR (T0002/T0016/T0026)
- Already purchased a scanner. Trying it out and creating a stream of improved reading efficiency will increase the realization of AI utilization.
- So is this one, but first fromPDF2025-03-11.
- I've already scanned the data.
2. Cause analysis (unfulfilled tasks and disruption factors)
1.### Tasks are often left ambiguous . - Tasks that do not have clear "objectives, completion conditions, and next steps" will be put off, and as a result, more will remain untouched. 2.### Multiple sources of information are dispersed and cannot be followed . - Slack, Signal, ChatGPT, etc., it is difficult to know what is left where. - As a result, even important tasks are omitted or neglected. 3.### Waiting for review and human verification becomes bottleneck . - Using AI speeds up the process, but it is easy to get bogged down in the "final review is human" step. 4.### Tasks with no deadlines (improvements you want to make in the future) - GTD-like Inboxing is in progress, but it is bloated with no specific due dates or priorities.
--
3. Next Steps
.
1.### Urgent task support . - First, the tax return (T0109) must be completed with the highest priority. Since it is an essential task right before the deadline, it will be handled intensively from midnight to the next morning.
2.### Slimming down the backlog . - Check "Vague Tasks" and add "Objectives/Completion Conditions/Next Steps" if necessary, and make a concrete or close decision. - Future tasks with no due date can be separated in a "someday/maybe list" style and organized in weekly reviews to reduce confusion.
3.### Introduction of centralized and small automated information pathways . - First, build a prototype of the Slack thread extraction tool (T0128) → aggregate information around Slack → gradually implement the flow of having AI summarize from there. - To the extent possible, continue the experiment of "paper memo → AI input at night (T0098)" to verify the take-up from paper memo to AI.
4.### Clarification of measures awaiting review . - The "timing" and "granularity" of the review should be determined at the outset, such as human review of AI-generated tasks/codes collectively at a weekly pace. - Try a system of checking a week at a time to avoid a large number of fragmented reviews.
5.### Formulate an internal case study for the "Spread the Word AI" demonstration in May . - MVID data analysis (T0008, T0009) and Slack analysis tool experiments (T0128, T8577) will be turned around in small pieces. - The first priority is to deploy to Azure and link existing data, and create a minimum showable form by May.
[** Sample schedule (from 3/17 to early next week)
This place was rejected because it was a reckless plan in addition to being wrong about the day of the week in the first place.
Supplemental
Task Embodiment.
- If we ensure the operation of writing at least "the purpose of the task," "the conditions for completion," and "the first step to be taken," the probability of the task being buried in an untouched state will be greatly reduced.
periodic look back.
- Visualize blockages by briefly checking "the amount of new task submissions," "the amount of AI-generated tasks waiting for human review," and "the number of completed tasks" in a weekly review.
Gradually expand automation.
- Do not try to automate everything at once, but start with small steps of Slack extraction → summarization or tasking by AI → modification by humans. Gradually expand the process while operating the system.
--
If you do the above,
- digest urgent tasks without missing them
- reduce confusion by organizing ambiguous tasks and taking measures to prevent waiting for review
- create a form of external appeal, such as a public hearing AI demonstration This process should go smoothly. Finish the tax returns first, and then develop a small automated/intensive review system.
next: Weekly note 2025-03-24~2025-04-05
This page is auto-translated from /nishio/週記2025-03-16~2025-03-24 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.