Processing AI

[Mindfulcraig]

I first read "Getting Things Done" (GTD) when it was released in 2001, and before that, I had consumed the cassette tape set on Managing Actions and Projects, which was the precursor to GTD. One idea that resonated with me was the concept of "open loops" and how identifying and objectifying everything on our minds can relieve mental overload.

I have explored many ways that AI could assist with GTD, but the main ideas typically revolve around automating the weekly or daily review or using AI for mind-dumping, etc. As a regular and mostly enthusiastic user of AI tools, I’ve started to notice that the low friction associated with initiating new projects, exploring interests, and brainstorming ideas can lead to an exponential number of "open loops" scattered across various AI platforms.

While some of this may be due to my ADHD brain, I believe it affects non-ADHD individuals as well. Many of the outputs from my AI interactions are likely disposable, but I’m realizing that if I don’t process them with some rigour, they will continue to occupy mental bandwidth.

I'm curious about the approaches others have adopted to capture and process AI outputs so they can clear their minds of these open loops.

 

[cfoley]

I am using AI for my programming work, and I find it very good for suggesting things and analysing large amounts of text (usually the codebase I am working on). I find it incredibly useful for this sort of work. It generally gets me to where I would have gotten to myself much more quickly. Often it finds some nuance that I might have overlooked.

I ask it to generate code for me and sometimes to write the odd report. It is certainly fast at these, but I would not say it is particularly good. The exception is things where I havle little knowledge or skill. For example using a code library that I am not familiar with is much quicker, easier, and better with AI.

When I ask AI for code recommendations, what it comes up with is usually questionable and often bad. This is for low stake things like writing some code in a project. I won't trust it with these decisions and the idea of trusting it with life decisions terrifies me.

In summary, for computer code work (and I expect for other knowledge work): AI is excellent at analysing and inspiring; it is fast but mediocre at doing; and it is dangerous for making important decisions.

 

[bishblaize]

To be honest, it's always been the case that you can put too many things into your system if you're not careful. I'm sure we've all experienced going to an inspiring away day, conference or brainstorming session, and coming back with 100 new ideas, only to sit there 3 weeks later looking at our Project list and thinking "Maybe this wasn't such a good idea after all". Using AI to automate/semi-automate input makes it easier to over generate work, but its still the same problem.

I'd say there's 2 things really, and they're as old as the system itself - run everything through your inbox, and keep up with your weekly reviews.

Part of inbox processing is about the mature assessment of whether its really important enough to do. And the weekly review includes a second run of the same idea, you pragmatically assess whether it’s really worth this project sitting on your list, and how long you leave it there with no progress before you simply retire the idea and move on.

There is some nuance about how we might pre-vet AI output before it gets into our Inbox, using a second run of AI. The same way we create filters for email to avoid Inbox overload. But fundamentally we still have to grapple with the decisions of whether or not something belongs in our system.