So, I asked it to modify its instructions.md file to not repeat that mistake. The result was the new line "Avoid single-use wrapper functions; inline logic at the call site unless reused"

instructions.md is the new intern.

So, I asked it to modify its instructions.md file to not repeat that mistake. The result was the new line "Avoid single-use wrapper functions; inline logic at the call site unless reused"

instruction.md is the new intern.

My company was acquired by another company that had in-house lawyers. California Law states that you have rights to your own IP, when produced on your hardware and on your own time. So, I was careful to air gap all my work on a separate computer. Meanwhile, the acquiring company ask me to sign an employment agreement. Its terms restated the California law in very ambiguous terms. I couldn't tell if I was declaring rights to my own IP or signing away my rights. So, I asked them: "which is it?" Their replay was "Yes." ... I was an employee for a day.

A few years ago, I tried that again with a remote team of 20 coders. I failed; I couldn't keep up with the barrage of pull requests.

Today, pair programming with Claude Code and GPT feels more like the later.

I think there is an opportunity here for smart refactoring. But, needs a larger context. I tried this on some legacy code with Cursor and Claude opus 4.1, but a million tokens is not enough. I dunno; maybe translate between a private and shared LLM. Has anyone tried this?

1) Aaron Marcus - who found optimal menu count to be 5 +/- 2

2) Magic number 7 +/- 2

3) Fitt's Law selectivity (bigger is easier)

4) Shared layout for mobile + desktop

5) I hate short form

6) Is 5) a non-sequitur?

7) No! I now have the attention span of a goldfish.

8) Maybe I should read a book

Wittgenstein stated that the differences between personal and cultural language makes it impossible to agree on anything foundational to philosophy.

Godel did something similar to discrete structures by hacking self reference -- a kind of recursive self reference with no exit state.

I think pair programming with an LLM is interesting. You get to compare personal context with aggregate context. Plus, the external feedback helps me break out of excessive personal self reference.

Does not stop there; there are tiers of middlemen: https://youtu.be/kVY7-Ti77UQ?si=2neBSw4yJL7qGdxX&t=441

https://news.ycombinator.com/item?id=31706955#31708672

Somehow OpenAI reminds me of a paper by Kenneth Colby, called "Artificial Paranoia"

[*] https://www.sciencedirect.com/science/article/abs/pii/000437...

For example, I wrote a NLP parser for a calendar app, at Tempo.AI. It was much more efficient than the visual interface. And thus, it was accessible. But, it didn't use the accessible idiom. Instead, it was universally more efficient, whether you are blind or not.

A good example is a wheelchair accessible doorway. One method is to have a button at wheelchair height. The other method is to have the door open with an electronic eye. The first is Accessible. The second is Universal. Doesn't matter if you are in a wheelchair or not. It's a throughput multiplier.

Hopefully, whomever takes this on doesn't take the standard Accessibility approach, which is adding an extra layer of complexity on an existing UI.

A good friend, Gordon Fuller, found out he was going blind. So, he co-founded one of the first VR startups in the 90's. Why? For wayfinding.

What we came up with is a concept of Universal design. Start over from first principles. Seeing Gordon use an Accessible UI is painful to watch, it takes three times as many steps to navigate and confirm. So, what is the factor? 0.3 X?

Imagine if we could refactor all apps with a LLM, and then couple it with an auto compete menu. Within that menu is personal history of all your past transversals.

What would be the result? A 10X? Would my sister in a wheelchair be able to use it? Would love to find out!

[1] https://github.com/musesum/DeepMenu

Coincidentally, the peak notch in the OP was the 1st subharmonic of that.

[1] https://en.wikipedia.org/wiki/Flyback_transformer