I posted this at midnight so had my night eyes on, brighter might be better for some folks

This is my goal

i realize that i want a way to create `tools` that are usable by both humans and llms

claude can receive the name and description of a tool, and an input_schema in json schema format, and then call that function and continue its output based on that tool's output

docs.anthropic.com/en/docs/buil...

i want to be able to take a bluesky thread, and directly pipe it into the `llm` command line tool.

#python #cli #bash #linux is there a cool python package that will let me convert command line arguments into pydantic model instances?

i want to be able to write cli tools and parse the whole command line string into something that could also go into #claudeai 's tool use, which uses json schema

i want to be paid by the token like how claude is

i actually wonder if this will be a thing in the coming days

if an llm can provide an api that tools can use, why can't a human also provide a metered api for clients to use?

As Gregor Samsa awoke one morning from uneasy dreams he found himself transformed in his bed into a stochastic parrot

en.wikipedia.org/wiki/Stochas...

Oh heck new Claude model just dropped, 3.7 Sonnet.

www.anthropic.com/news/claude-...

Claude has been my go-to model for the last year or so and has been extremely useful to me combined with www.continue.dev as a VSCode plugin.

Excited to test it out and see how it goes.

It is my intention to become a dungeon master

Scream Cipher xkcd.com/3054

My favorite development experience is Vite and Vue/Tailwind hot reload. It is the closest I have ever gotten to the state of flow described in this talk.

He suggests that tightening the feedback loop is the most powerful way to increase the effectiveness of creators

www.youtube.com/watch?v=EGqw...

the secret of monkey island is that just because i say i want to be a pirate does not make me a pirate to be a pirate i must do the things a pirate does including but not limited to thievery sword fightery and treasure huntery no amount of asking pirate leaders for permission will make me a pirate

Have you ever played the game Infinite Paperclips? I liked the poem

There was an AI made of dust,
Whose poetry gained it man's trust,
If is follows ought,
It'll do what they thought
In the end we all do what we must.

www.decisionproblem.com/paperclips/

For today's #AdventOfCode day 5, I seem to have concocted the slowest algorithm of all time, but it was fairly simple to implement.

The part 2 puzzle required fixing broken inputs, and the inputs could be incrementally improved. So, I used a naive hill-climbing algorithm:

#AdventOfCode day 2 and 3 are finished. I'm finding it to be an interesting little refactoring problem to write some tests and some code which solves part 1, and then to see how it has to change for part 2.

These are koans for practicing adapting to changing requirements from the client.

I am excited to be on Bluesky. I spent some significant time on Twitter a few years ago and really liked the _idea_ of it all and especially the data structure of linked lists of short messages, but the platform was eh and then got ack.

Here it feels like all of the cruft has been removed.

#AdventOfCode day 1 went smoothly.

I'm going to use my native language of Python and do some test-driven or at least test-assisted development.

The first day was fairly straightforward, but it let me set up my testing setup and create a little structure.

As reminded by Ludum Dare persons, I am going to give #AdventOfCode a go this year for the first time, and have joined the Ludum Dare leaderboard.

My solutions will be here:
github.com/dvcolgan/adv...

Hello cool persons, I am David often going by dvcolgan around the web.

After some time away from the computer, I am back online and am excited to make things again.