One of those two headed centaurs. Probably an evolutionary midpoint between horses and centaurs.
Home
Public posts preserved on my own site.
Should I log on to twitter and see if Taco Bell is taking full advantage of taco being in the national news?
Well, the new $200 (per person) price for vaccines is going to wipe out any income tax reductions I might be seeing.
#Jetbrain's survey: "I want developer-oriented Silly Tavern but you guys are writing "do Matt's job for him so his boss can justify shrinking the team and cutting salaries". Copilot breaks reliable intellisense, it was a mistake and it was a mistake to copy it. I speak from 2 years of trying to let the bot autocomplete my code. It can't. You have to talk to it. It is like a cat, it isn't going to do what you want without some herding."
Who ordered this much coffee
The 3 kinds of queues are:
- FIFO
- LIFO
- FAFO
Google Gemini "Live" is the bot from the movie "Her" and is the first bot my 3rd grade daughter would talk to. She didn't like talking to OpenAI models via the web ui.
Aside from fluid idle chit chat being solved, image search is now a solved problem and being able to do with while talking and sending video is a big improvement over "take a picture, put into input box, type question, send"
SSO, I think this is a misnomer. I've been asked to sign in 10,000 times and lunch isn't over yet.
I need to fit a PC, 2 monitors and 2 laptops on the same desk. When I'm done, my standing desk will only be able to stagger.
I've noticed my personal productivity goes back and forth between:
- monomania
- 100% rabbit hole exploration
I enjoy the monomania part better, it feels more productive. Rabbit hole exploration has more discoveries with a 10x, 100x pay off *later*, but monomania has me moving towards a goal 10x faster.
Both of these workflows are incompatible with SAFe, scrum, waterfall, Kanban, etc.
Explaining to my son how difficult it was to work with LaTeX in my day.
Now #golang has a supply chain security problem.
Github actions, terraform, go.... hmm, what do we got in common here.. treating git endpoints as trusted. But git endpoints can't have any central authority to check for malicious code! At least at #pypi you can report a package as malicious.
https://www.youtube.com/watch?v=EyO_SMl2YBk&ab_channel=CodeHead
@acdha Keep tweeting, the FOMO is killing me.
More new (to me) mypy alternatives
so I guess we will run
`mypy . && ty check . && pyrefly .`
@mattl@social.coop That resume is for someone who already knows you. The first line people reviewing resumes need to be able to instantly peg you as some job title that they're hiring for. Good luck, it is rough out there (compared to 2 years ago)
All this talk about single pain of glass. Sounds better than having many pains. Is this like walking on glass? Zero pains would be even better.
@osunderdog uv is now mature (both in the sense of feature complete, docs and no visible bugs for my initial use).
There are a few pyproject.toml converters with various bugs, so I end up having to convert the pyproject.toml sections by hand.
Pycharm now has uv support.
I miss `poetry shell` and now I have to activate the venv 2 different ways depending on OS. poetry's `shellingham` solved a problem that uv doesn't.
OpenAI/Anthropic: sophisticated UI for LLM prompts
Google Gemini: bare bones UI that can't even export all chat history.
Silly Tavern: hold my beer. OpenAI is about 2 years behind the state of the art prompting UI
I want my bot to stay in character, which means, a software developer who has strong opinions about untyped python, google doc strings, the use of pytest instead of unittest, etc.
I want my bot to know the Lore of my office, like that we used to use Pipenv, then poetry, now uv.
♡ 1 ↻ 0
Regarding the thinking models- this isn't too far from what people were doing with role playing group chats (e.g Silly Tavern), the difference is the thinking models are role playing someone with really bad anxiety.
Have you ever opened a .pyc file and thought, "man, that's ugly"
Problem partially solved, this is executable python.
🛣😀😀😀😀😀😀😀😀😀😀😀😀😃😀😀😀😀😀😀😀🛳😚😀😀😀🚕😀🙓😀😚😀🙲😀🙜😀😢😀🙓😁😵😁😀😀😀😀😀😀😠😀🙧😂😩😃🙣😁😀😀😀😀😀😀😀😀😀😀😀😅😀😀😀😃😀😀😀🛳😢😀😀😀🚕😀🙛😁😀😀😀😀😀😀😀😀🙓😁🙕😀😎😀🙓😂😳😃😵😁😀😀😀😀😀😀😠😀🙧😀😩😃🙎🙺😇🙈🙥🙬🙬🙯😬😠😁😡😩😁😅🙰🙲🙩🙮🙴😩😁😄🙮🙡🙭🙥🙳😁😀😀😀😠😈😼🙳🙴🙲🙩🙮🙧😾😅🙨🙥🙬🙬🙯🙲😇😀😀😀😂😀😀😀🙳😓😀😀😀🚀😀🛜😄😉🚈🙇🚐🙄🚐😶🚘😑🛐😊😛🛕😄😜🛳😀😀😀😀😅🙗🙯🙲🙬🙤🙎😩😁🙲😇😀😀😀🚩😀🙲😈😀😀😀🙲😆😀😀😀😈😼🙭🙯🙤🙵🙬🙥😾🙲😋😀😀😀😁😀😀😀🙳😓😀😀😀🛰😃😁😁😁🛲😄😁😁😝🛱😆😀😁😆🚀🙧🚅😎🙲😈😀😀😀
I wanted it to be space delimited multi-emoji with meaningful digraphs for opcodes, but no! dis.dis() doesn't have enough information to convert it to bytes and the raw bytes doesn't have enough information to convert it to effectively dis.dis() style output.
And the python assembler library that exists is for 2.x
So to complete this joke, I'm going to either need more time & effort, or drop my goal of having executable emojis.
♡ 0 ↻ 0
Ah shoot. Do I need to upgrade this to python 3 to finish a low effort joke?
We had a python assembly language and we lost it, it appears.
Which end of a pool should you jump into?
Software developer: Its deep ends.
I couldn't go to #PyCon this year, so I guess I'll watch videos from last year's
https://www.youtube.com/playlist?list=PL2Uw4_HvXqvYhjub9bw4uDAmNtprgAvlJ
Some experience with thinking models:
- for some questions they waste tokens profusely. A magazine's worth of thinking? What they got my anxiety or something?
- for unit tests they assume the code is correct, they assume there is a correct answer. But unit testing is all about finding crap code! Wrote a unit test when it should have said, "this code is buggy and I don't understand what it wants to do"
- It translates 1980's BASIC into python like a champ. Also OCR's like a champ, granted
Experience with Google's Gemini, which I'm testing after 2 years of ChatGPT and having been frustrated with Google's Bard (moron!) and the subsequent model's refusal to speak Esperanto (addressed! finally!)
♡ 2 ↻ 0
A single piece of shit developer? No, but with a team of 50, I think we can ship code to production.
@AlSweigart Which model is doing that?
Anyhow, I've been considering switching to Gemini for the larger context window- fitting a whole existing code bases into the context before asking a question
It has come to my attention that vibe coding means non-developers asking the bot to code until the code works, not a developer asking a bot to code and then editing it until it works. Now I got a lexical gap for LLM-assisted coding again.
@mheadd :(
Explain it like I'm 50. Which I am.
With 1980s references.
♡ 3 ↻ 0
@ayo This is the real software development lifecycle.
AWS Cognito and every auth server:
We are 1000% secure. All you have to do is set the 15,000 settings exactly perfectly. There are more possible configurations than positions of atoms in the universe, but you can do it with our user friendly admin pages! We believe in you!
But if you screw up, just remember, most auth servers are hacked because of incorrect configuration and that's your fault.
Bribe coding. Take the envelope and fix this bug. No questions.
Imbibe coding. Chug! Chug! Chug! Write JS in the main.py file. Puke.
Scribe coding. Write code with a fountain pen in a dotted journal. Give to intern to compile and execute.
Tribe coding. We all write a few lines a code, then we go slaughter the other tribe before they can ship a competing product.
Subscribe coding. You signed up for the 100 lines of code a month, would you like to upgrade your plan?
Pope's on fire, the temperature has gotten high enough that the smoke has gone from black to white, signalling a clean burn with minimal soot.
Inflation? I'm concerned about the snap, crackle and pop of prices.
#ty #mypy - but mypy already compiles itself to c via #mypyc so being faster isn't what I'd be expecting, it will be different and probably find different typing errors since for all the type checkers, it comes down to a matter of opinion. Do you think that will be a runtime error? That's in the future, who knows man.
Idea- (and not a piss on AI emotional reflex)... maybe we know we have hit AGI because the bots exhibit a certain level of stupid. When the potential number of paths your thoughts can go, the odds of one of them being stupid goes to 100%. We all thought we could create a machine (or an educated human) that could map out the possibilities so perfectly all the stupid would be gone.
Every problem you can pose in English has multiple meanings, answers & many of those answers are going to be stupid.
No, my best days are still ahead of me. It's just a temporary phase. You don't even know me. Login.gov knows me.
It is time for my every-6-months review of how AWS KMS works so I can forget how it works again in 6 months.
Confluence thinks I'm worthless.
After simplifying today is the start of the year again, 5/5/2025 = 1/2025