tonyaldon

What's the new shiny programming language?  Plain English.  Before, with only a few words (if, else, for, def, return, etc.) you were set to program a computer.  Now, it's the wild west, you need the whole dictionary.

What's the new shiny programming language?  Plain English.

Codex: the relentless teammate reviewing every OpenAI PR

You write faster with AI!  I don't.  In fact, I'm slower.

I agree with everything you said.

When I think I'm done and want some pushback, I use this prompt:

---

Here is my final draft, my final writing.  Do not correct it.  Just tell me:

1) Why does it work? What is the psychology behind it?
2) What can be removed or shortened?
3) What is missing?

How I use LLMs and Firecrawl to dig into company culture

I think an old-fashioned brain will always work great.

You write faster with AI!  I don't.  In fact, I'm slower.  But I write with more precision.  When I'm done with a piece, it feels closer to what I really think.

What is definitely faster is trying different formulations: in seconds I can get alternatives, apply systematic rules, and compare my draft with what I have in mind.

And as I know what I like, when I get to place where I like it, I just stop (paraphrasing Rick Rubin here).  No overthinking.  I feel great about it.  I'm done.

Faster to try, 100%.  But faster to finish?  I'm not so sure.

What about you, AI and your writing?

Did you ask AI?

Your LLM is only as good as the context you give it

Agreed!  And in that case, nothing in the source code gives you any hints.

Absolutely.  I think it belongs in the docstring for that piece of code. I would have found it after 5mins of poking around the source.

Start small, keep iterating, and have fun.  That's how I approach working with AI.

Recently, I hit a nasty bug while using a library that was new to me: characters were disappearing while typing in an input form on Android.

The LLM had access to the library's source code and docs but couldn't figure it out.  Same with their own AI chat on the docs.

I found on GitHub issues that this bug had been fixed with a new flag when instantiating the library.

This new flag, added because of the bug, is not documented as fixing this specific problem.

It's only mentioned in a post in the ticket's issues.

After finding that the issue was solved, I tried web searches.  Sometimes the bug shows up, sometimes it doesn't.

If it had been documented in the source code, the LLM would have found it with no effort.

Great documentation is still the cornerstone of great developer experience.

Any resources about this? I curious and want to learn more.

But you still can learn something from bullshit!

LOL!

I still do.  I don't do it to influence ChatGPT, but for myself.  It helps me keep thinking as if I were talking to a person (not a supercomputer), so I can maintain a conversational flow.  It feels more natural and helps me organize my thoughts.

Start small.  Iterate.  Cut.

Accidentally used an "inferior" model—the best just months ago.

Accidentally used an "inferior" model—the best just months ago.  Almost got offended by its imprecise answer.  Talking to super-intelligence is addictive.

I just caught myself thinking in prompts.

Great advice.  I'll go walk the dogs now.  Thanks for the reminder.

Sometimes I think the answers from LLMs matter less than the thought that goes into writing the prompts or questions.