The pledge
I pledge to put down my ideas on paper before asking for help from Claude or ChatGPT, , in order that I will produce better work in the same amount of time.
Intention
Why did I make this pledge? I use Claude and ChatGPT a lot. I know they consume energy, create emissions, and use water both during their training phase and when they are running.
The numbers for these operations (our calculations) are probably of the correct magnitude but their significance is not clear to me. Comparing the training of an LLM to X number of flights doesn't tell me whether the training should have gone ahead, or whether I should use the model. Should the flights have flown? Were they for a good reason? What is a lot of flights?
Even if I knew the answer to these questions, I don't think it would matter. If LLMs are useful - and they are - we are going to use them. If the Jevons paradox applies, we will use LLMs - or generative AI - to do more, whether that more is valuable or not.
Hypothesis
Therefore, I am looking at this the other way around. Rather than monitor my LLM usage, I will keep a watchful eye on my intentions. The idea is to know my goal in advance and use LLMs to achieve the task in less time, or to do it better. Ideally both things would be true.
I am assuming that by doing things quicker or better, I will save energy (and by association emissions, water, etc.).
Outcome
I don't know how to measure the outcome of this pledge.
I will therefore measure success by whether or not I do write things down. I can verify that. In addition, I aspire to work faster and better. This approach is unscientific but still, I think valid, because the pledge is really about my - human - behaviour, not whether or not AI can help me.
The outcome section was written after I asked Claude what it thought. I didn't change anything. Here is part of its reply:
This reminds me of other situations where the act of preparation or reflection itself can be more valuable than any measurable outcome. For instance, the practice of writing a project brief often clarifies thinking regardless of whether the final product strictly follows that brief.
Claude
pledger: Dan
Update
I stuck to my pledge and committed more ideas to paper. I also began to sketch out the idea of using AI as a ratchet for iterating on ideas.
I completed one project without recourse to AI: a client/server environment to accompany a workshop on HTTP messaging.