Motivation

Understanding and resolving my recent wobble in my motivation for research · Jake Levi · 2025-10-21 · [ Back to home ]

This thought process I was having this morning was originally going to be a Tweet, but it quickly became too long for a Tweet, so I did what any reasonable person would do, and started a blog.

It seems pretty unusual for people to talk publicly about not being motivated. Is that because it is not an issue that people struggle with, or rather that people do struggle with it, but think it is somehow "wrong" or even "weak" to talk about? I think it is pretty sensible to talk about this. Speaking from my own experience, my motivation is aroused (for want of a different word) by a certain context. But time marches onward, contexts change, and there inevitably exists some threshold beyond which it is simply rational to re-evaluate my motivation, and decide whether or not I'm still pointing in the right direction.

So, I've been having a bit of a "wobble" recently with motivation for research. I think there are various factors behind this. But I think a big part of it is that I have always been betting on fundamental ML research that is trying to establish better foundations than those of the current mainstream (transformer-based architectures, backpropagation-based learning rules, etc). I think this has caused me to have a "wobble" because these conventional foundations have recently been so staggeringly successful that it simply feels incongruous right now to be trying to change those foundations.

But I think the time will come when the hype cycle matures, and there will be a common realisation that there is a strong incentive for orders of magnitude better efficiency (until we are comparable with what we know is possible from human learning), and there is a limit to what can be squeezed out of the current status quo, which (I believe) will not get us close enough to what's possible. Then the "noble quest" to try to change the mainstream foundations of AI will seem a lot less incongruous.

So I'm faced with a choice. I could take a break from trying to improve AI efficiency and work on something more mainstream (ugh). Or, I could continue on my current path, and reframe my current work as futuristic rather than incongruous, a headstart on what one day may become mainstream. Okay, that's enough time spent writing this blog post, I've got work to be doing.