Manual Exposure Meandering

I.

This is a reflection on getting out of a local maximum.

For years, I took photos on my Canon Rebel XT in Program or Aperture Priority mode. I couldn’t understand why anyone would need more than that. Either one seemed to offer total control, and I had a vague sense that Manual Exposure was Too Complicated. Why focus on 10 controls when you could focus on your shot instead? I brushed aside my photographer friend’s comment that he exclusively shot in Manual, thinking he must have gotten the 10,000 reps required for it while he shot weekly weddings in the summers.

Yet, I was frustrated with the experience of photographing my daughter and my dog using a newly purchased Ricoh GRiiix in Aperture Priority. I wanted control of the aperture, but sometimes I felt I needed control of the shutter speed. Both subjects are so fidgety¹! I tried to find a setting or button to quickly adjust the minimum shutter speed, but to no avail. It took weeks for me to have the thought, “Hey, I wonder if this is what Manual Exposure mode is for?” (It is, in fact, exactly what Manual mode lets you do.)

Changing to Manual has been a total surprise, because it’s actually less difficult to navigate. Rather than getting distracted by too many controls, I’m no longer distracted by wishing for controls that aren’t available. I’m still getting the hang of capturing my creatures’ moments perfectly, but my frustration has moved out of the camera and into the real world, where it belongs!

I’ve since realized what happened: my Rebel XT did not have an auto ISO setting, and adjusting the ISO required diving through a multi step menu. That makes Manual Exposure mode legitimately more difficult to control than with my Ricoh. I simply had an outdated belief that I didn’t reevaluate because I didn’t even realize it was there!

II.

Actually, this is a reflection on LLMs.

I conversed with ChatGPT about this experience, asking if there was a name for the phenomenon. It gave me a few decent suggestions: Imposter Syndrome, Self-Limiting Belief. When I remarked that this felt like reaching a local maximum, it heartily agreed and patted me on the back. I walked away feeling “whelmed” at the interaction; no suggestion resonated strongly.

But sometime later it dawned on me that the mediocre response I got was downstream of mediocre input. I didn’t think very hard about the question I posed, and I got a fairly generic answer. From there, I began to consider the way I use LLMs as I code. My results are mixed, and I often abandon them to dive back into the familiar pre-LLM workflow I’m used to.

What if that is just another Manual Exposure mode? What if there’s a technique I’m just unfamiliar with, one that is hidden by a long standing unconscious habit, one which would unlock LLMs’ power for me? It’s still early days for these tools, and the professional-photographer-friend-who-suggests-manual-exposure-mode maybe just doesn’t exist yet.

III.

Actually, this is a reflection on thinking.

The next day a new thought dawned on me: how did I connect a camera’s exposure mode to coding with an LLM? I’m still not really sure. It didn’t happen immediately. It took multiple leaps to get there. And I wasn’t paying attention to either topic when it occurred to me. Something went on in my subconscious mind, across a long period of time, that matched these thoughts into a pattern.

Right now, ChatGPT’s o1 model is showing remarkable abilities because it can reflect upon its own thinking process, refining it and choosing the optimal path to achieve its instructed goal. That seems fundamentally different than what I experienced.

So, I wonder, is this subconscious process something that we’ll bake into LLMs? Will this be what agents are for? Or is this a kind of thought we can’t yet fully describe or build into silicon? Is it what makes us human?