An anti procrastination experiment

I have been reading https://www.neelnanda.io/ most popular posts and many of them really appeal to me. He has one on preventing procrastination. He is not really aiming his efforts at day to day things, but more interesting ideas that he never gets around to. I think these are the more important thing to not procrastinate on too.

I decided to adopt his idea of doing something - anything really - about an idea as soon as I have it. This prevents me rejecting it straight away, or worse, just letting it drift into the past. It might keep popping into my mind, but if I don't do something about it right now it will almost certainly never get done.

An easy thing to do as a 'something' is to set a 5 minute timer and write in a notebook about the idea. This is a low cost way to do something and make the idea a bit more concrete. After this the idea is then down on paper. I can just leave it there. I will revisit my notebook repeatedly, so don't need to worry about just letting the idea drift away. Also at some point after I have done the 5 minutes - immediately or when I see the notes later - it is perfectly reasonable to just drop the idea. No harm done. The thing that this method tries to do is just avoid that low ownership passing thought of an idea to die without any real effort being put into it. There will be way more ideas than I can work with, but if I can get into this habit of doing something about them I think I will be a lot better off.

The first thing that I applied this to was an ongoing problem I have been having with an AI journal app that I have written. I want to be able to chat with the journal. Ask questions which will use appropriate context from the journal to get an llm to answer. The context collection piece was broken having been written (correctly) several months ago, but the code has since developed a lot more and that chat feature broke. I spent 5 minutes noting what I thought was wrong and getting a decent shape on the problem. This is very light touch. It is not meant to be anything definitive, just a 'something' to do. So I am in the process of using Codex to fix this bug once and for all. It has been hanging around for quite a while.

The second thing that I did a 5 minute timer for was to get make AI interactions more standardised in the application. In particular I want to get the application to display some information about the context used. For example 'the last 7 days of journal entry'. Also the model should be reported every time and the total size of the input to the model. I have not done anything about this one yet, but am happy that won't forget it.

Another thing you can do to avoid not doing anything is to make an entry in your to do list. This only works if you regularly clear out your list. I am often guilty of not getting to 0. I am down to very few remaining items there now and should get to 0 today.