|
Welcome back to Tiny Thoughts, a collection of my insights, experiences, and observations on AI. Got something juicy to add? I'd love to hear from you!
|
This week felt like watching people discover AI can be more than a fancy search engine. From therapeutic experiments to century-old warnings, the common thread is this: we're still figuring out what AI means for how we live and work.
When telephones ate the world
Whenever I get really bored, I read the news… from 100 years ago.
On TimesMachine, I came across a 1924 New York Times article that perfectly captured something we're living through right now. It's about a man who finally got a phone after years of resistance. His complaint:
"Look at my next door neighbor. He comes in now and then from his shop to chat about business or about this and that; and can he? 'Mr. So and so, the telephone'—his clerks come running after him and he dodges back and forth till he forgets what he came to say in the first place."
The article describes "a philosophy of tranquility" and warns that "the trouble with the hum of machines is that they do not hum steadily enough; there is peace but the machines go by fits, forever speeding and slackening and speeding again."
People sure wrote differently back then.
Here's the line that hit me: "Each of us must have some point of reserve and some refusal; we must to hold our self-respect."
Feels like we're having the exact same conversation about AI that people had about phones 100 years ago. The technology changes. The human need for boundaries doesn't.
Source: NYTimes
Blurring the lines with AI
When we blur boundaries with tech, interesting things happen. Sam Stern brought together people using AI in wildly different ways. What happens when you treat AI like a being with its own agency? What about using AI as an adjunct to therapy? Or to create music? What got me was how deeply personal each approach really gets. The messy middle they describe is exactly where we all are. Nobody has this figured out yet.
Source: Voice of Esalen
Claude Skills makes building even easier
Claude introduced Skills. Skills are just files with instructions and optional scripts that Claude loads when needed. That's it. No complex protocols, no heavy usage, just plain text telling the model how to do something.
I'm still experimenting with it, but the brilliance is in the simplicity. Want to automate your workflow? Write a skill that describes it. The model figures out the rest. Compare this to MCP, which is a whole protocol specification consuming a lot of compute (and energy). Skills trust the model to read instructions and execute them.
This table explains it further:
Source: Simon Willison
Thanks for reading!