retrocog

A village builds a clock tower so everyone can glance up at it and know what time it is. The tower helps as people meet more easily, work schedules align, and meals are shared. Over time, fewer people watch the sun and the clock becomes authoritative. When it stops working, nobody knows what time it is. Discord ensues and arguments follow as meetings are missed and people wait impatiently for repairs. An older woman in the village kept baking at dawn and never stopped watching the sun. She never relied on the clock. Villagers gathers at her door and call her wise, for her practice ends up inadvertently helping to re-calibrate the clock once it is repaired. She replies that she is not wise or special and was just going about her usual baking routine.

I'm on version 6 of a writing an agent voice, style, tone, and orientation guide I have been developing over the course of the past year. Every time the agent writes something using the guide, I edit it and then update the guide where needed. Slowly, the agent is beginning to sound eerily close to my natural writing voice. No, I didn't use it for this comment :)

Sometimes, it seems, Microsoft will actually OEM Windows at a loss. A year ago I bought some NUCs with Windows pre-installed that were less expensive that if they came with Ubuntu. Microsoft apparently felt is was important to try and hang onto that market. I bought the less expensive NUCs with Windows and then formatted them and installed Ubuntu, saving myself around $40 per machine.

Is it possible to have more than a purely transactional relationship with an AI assistant? If so, what are the benefits and how does it work?

That seems like a good idea. The question is how the JWT is generated. A standard one would be more akin to a traditional crypto keypair. That is a "signal" key insomuch as it tells us who controls an account. It can't tell us the owner is the controller and that is the current weakness of crypto right now. To know the owner, we need another type of keypair to go alongside the traditional kind. That would be a "tone key" and is generated by a refreshing seed derived from the entropy of long-running, unfakeable conversations. The same way a friend might recognize us as being ourselves.

Using Agile methodology with agents actually works pretty well in my experience. We do sprints and then code reviews, testing and revision, optimization. During code review, I inspect everything the agents created and make corrections and then roll the corrected patterns into the training documentation for the agents so they learn and don't make the same mistakes.

I have not found that to be true on a personal level, but in fairness it does seem to be a widely reported problem. At its core, I think it is an issue of alignment. That is something different than skill.

IMHO, this is the exact instinct and there's a way to verify identity, location, and age without even having to share those directly.

"What looks like effortless writing is often the result of years of thought and revision. New tools change how prose is produced, but not where ideas come from. In the end, the real question isn’t who polished the sentence—it’s whether the signal made it through."

The above summary was generated by ChatGPT after assisting me with drafting the article. I couldn't have said it better myself. The point isn't to show off my writing skills, but to get a message across. I outline some of the reasons why I go with AI writing style, instead of using my own writing voice.