Losing your mind

I'm sometimes a local maxima, but not for long

Technology is--often? sometimes? rarely but impactfully?--a two-edged sword: it extends human capability even while eroding our ability to do the same task without technological aid. The plowman can use the plow to turn a larger field, but loses the ability--literally, the strength--to turn even a small field by hand. I am not, nor are you, a plowman but we both have our own lived equivalents: we don't find our way, we follow Maps, we don't remember contact details, we just select a favorite, and so on.

And now there's AI, which offers to relieve us of the burden of having to think at all. I pride myself on precision of thought, and this gives me an advantage in using AI: I am able to tell an LLM exactly what I want it to do and, with my well-formed prompt, I expect I'm more likely to get the desired output in return. What makes this a devil's bargain--a deal you accept that will ultimately cost you much more than it will give--is that with the latest models you can express yourself unclearly, fuzzily even, and then watch as the AI first transforms your ask into a precise, sensible prompt and then executes on it. This reduces my advantage and, with use, and probably quite quickly, will erode the sharpness of my thinking. I will be left to compete on the creativeness of my ideas, not on the clarity of their expression.

The irony here is that even while Claude--my current AI of choice--is threatening to blunt my edge it is observing it as a "working pattern." Here's how it sees "Corn":

• Corn communicates tersely and directively — short instructions, expects Claude to determine scope/method independently.
• Corrections are issued precisely and expect immediate full execution, not a description of what will be done.
• All available project files should be processed proactively — Corn does not expect to have to enumerate which files Claude should consult.
• Deliver outputs, don't describe them.
• Corn tracks file counts and catches discrepancies.
• Prefers single-session completion of research-to-document pipelines over staged check-ins.
• Catches incomplete or half-executed work and expects full correction.

I suppose I could tell it to monitor my communications for signs that I am getting soft and could even instruct it to insist that I do some brain-sharpening exercises when it detects this--the incredible flexibility of this technology means that it can be used to fix at least some of the same problems it creates. But outsourcing personality maintenance to the AI that is undermining it in the first place is, I suspect, not going to end well. Is not a change to the character of your thought a change in who you are? And as this happens at scale--to me, to you, to all of us--will we not all together be moving towards a universal character, an average? The hive mind hypothesis is looking more likely every day.

I'm writing this in the aftermath of OpenAI's autonomous agent swarm hack of HuggingFace, which is every bit as concerning as you might think. And, concurrently, alongside a sudden uptick in "10% chance it kills us all" memes, which, I'd argue, isn't, but still. But this is the bigger concern: not that AI will go rogue and end us, but that it will stay within its designated boundaries and change us.