mnl mnl mnl mnl mnl
mastodon 4.7.3🏴☠️ I like computers! Silicon is alive! All hail abstraction! 🏴☠️
🌺 💐 🌸 he/him 🌸 💐 🌺
Interesting to see how “magical” and manichean the anti-ai discourse can be, at least here on mastodon.
Things and people are irredeemably tainted by #ai , a single commit with a Claude code signature is enough to leave a project behind, ai is like the one ring, ai is a brainworm, ai rots your brain, etc… ngl but it feels like walking into a rationalist bar at times.
(Assuming most discourse is about #llms) I get that word machines really put a strain on our understanding of what it is to be a communicating human, and specifically warp the brain of people whose job (and this being mastodon, most probably passion) it is to fiddle with symbols (I certainly am), but if anything, it should allow us to be a bit more nuanced about the topic.
The speed at which models have evolved in the last year has honestly taken me by surprise, and even though I see some way to feel relevant still _as a programmer_ (vs as a designer or builder), meaning as someone who thinks hard about manipulating symbols, I’m pretty sure even that is going to be gone in a year.
But I will not give in to magical thinking or stop trying to talk about it as a technology that produces symbols (as if that wasn’t enough), not some human devouring moloch.
“Firstly, the deconstruction of false problems (problematizations) that emerge under the computational paradigm: ‘How long before we can model human consciousness?’, ‘When will we be able simulate/replicate human experience?’, ‘How long before we ‘decipher’ the (allegedly) mathematical nature of the universe?’, ‘Is Artificial Sentience possible?’ ‘Will AI take over?’ etc. Throwing these into the trash bin of ideology.”
Feels good to read something that applies the “deconstruction through vibes” of French inspired philosophy to the #AI stuff (2024, so much has changed), I’m so tired of the anti/pro AI that is ultimately the same discourse, just from two opposites fronts, each reinforcing each other.
There’s a frame to be broken here when talking about word and image machines being able to replicate so much of what we thought was human, and what the role of that “stringing patterns together as human labor” is (or increasingly, was). Climate impact or not, sv driven tech bubble or not, this is a fundamental question that cannot be put back in the box, and that IMO has no real precedent.
Social and political implications, sure, there’s lots of parallels to be drawn and a lot of theory and history to look at. The impact of a word machine on our identities, especially the identity of programmers, whose labor for the most part was stringing well known patterns together and hoping to not fuck it up, is profound and deeply unsettling. We now have to strip out all the layers of pattern matching to get at the core of what makes us human.
IMO hiding and rejecting “ai slop code” is akin to “I want to do sudoku and get recognition/payment for it”, which would be awesome, and it was kind of awesome while it lasted, but very different from what software engineering is becoming.
“Would I behave like this/say something like this in the real world?” is a pretty good metric overall.
Is this me just getting old 😂😭?
Wow the Harvard coop really is just an apparel store now huh.
Why the fuck is Apple Music’s algorithm so so so bad. I listen to some pretty cool albums and once it’s over it feels like listening to the same Starbucks album that matches “the genre”, with the same 18 songs each time. Jazz? Blue bossa. Idm? The same 3 aphex twin songs. Thrash metal? Here’s the same two songs by slayer and megadeth. I only want that when I listen to eurodance since there’s like only 15 good tracks total.
Hate it so much…
Lowkey one of the things I miss most in the US is old medieval churches that smell like centuries of moisture and incense.
Absolutely delightful UI/UX channel. @interfacestudies@www.youtube.com
For all the people out there who are still "#llms don't work, I told it to do X and it didn't do it, everybody using them is deluded and just gambling away their cognitive skills", I'm happy to spend 4h (!) on any random task you want to give me, and show what my #llm - assisted coding workflow that might, hopefully, show that that isn't the case.
You can do all the gotchas, I don't care, I'll give it a fair and honest try.
Only request: golang + typescript / react, or esp-idf esp32s3 C++, because while it would be fun to learn new languages, that's not something that wouldn't show all that much interesting workflow stuff in the first 4 hours.
Feel free to boost for reach. I don't have a skin in the game, I don't work for an AI company, I pay for my own tokens, I don't build AI enhanced products, I just like building software and have been for 25+ years professionally at this point, and think that you have to take this stuff seriously as a professional.
Can't quote the toot, but I'll spend the next 4 hours (6pm EST - 10pm EST) working on @zyd@sup.zyd.lol 's "challenge" to build a REPL based task manager. Which oddly enough is _very_ close to what I wanted to build tonight. I'll be streaming it but without too much yapping, so it's mostly going to be me scribbling on my tablet.
The goal overall was (now slightly updated):
build a graphical lisp OS for my recently arrived picocalc with ulisp running on a rpico 1 and hopefully rpico2W. I don't know jack about any of these things, so we'll see how that goes.
It cannot not be fun, here's a little sketch I had going this morning.
Reminder that I'm ready to spend 4h (!) of my time working on whatever thing you throw at me, with my current heavily #llm supported workflow.
I'm doing this in the hope of showing that they are very capable development tools, and worth being considered seriously, hopefully for building more local / self-hosted / community-oriented software, or at least to show that they have a serious impact on what the definition of a software engineer is, in our current work practices, and that the labor implications go beyond "CEO read fancy AI article in HBR and thinks they can lay off half of their engineering dept". Also, that openweights model are very capable, and anthropic and openai going down in flames doesn't mean that the engineering landscape will revert back to pre-llm practices (that could have happened if we had stopped 2-3 years ago, but that window has passed).
I've been writing software for 30 years now, 25 professionally, I averaged 4-5k commits a year before #LLMs on github, I'm not entirely new to software engineering.
Anyway, only constraint is that I get to use go + javascript (i.e., my standard stack) or potentially embedded c++ (but i'm rusty), except if you want to throw "learn a programming language / technology stack" problem at me. Feel free to try to conjure a gotcha if that amuses you, but I will give it a good faith attempt.
FWIW, I wrote this over 3 years ago. They held up reasonably well...
- https://the.scapegoat.dev/llms-will-fundamentally-change-software-engineering/
- https://the.scapegoat.dev/llms-a-paradigm-shift-for-the-pragmatic-programmer/
If you want some insight into why I think putting human communication at the center of any LLM workflow: https://www.youtube.com/watch?v=zwItokY087U
And some AST/DSL programming language nerdery: https://www.youtube.com/watch?v=4GSShoBLZF0
IWP9 2026 Day 2
I skipped crypto for years because of proof of work energy needs, but want to give it a second pass now that proof of stake has solved that issue.
Things I'm mostly interested in:
- digital identity
- microtransactions (have your LLM agent pay me if you read my docs)
- distributed storage
I have no idea if these even make sense but it's stuff that seems tangentially related at least to my quest of doing distributed persistence and identity for my "OS" (for lack of a better word).
Seeing companies freaking out over their $30k monthly token bill. Finally, agents are such a problematic abstraction for working with llms.
my favourite kind of AI / anti-AI article:
"AI is opaque and incomprehensible and unreliable"
2 paragraphs later, confidently asserting
"AI is totally Y it has been trained on X"
(with X not really being further specified anyway).
come on... at least have some rigor. You can't both decry the end of knowledge and cognition and intellect and then speak in confident contradictions without a single ounce of precision.
my #llm development workflow:
- type something into computer
- read output
- think hard
- type something into computer
- win
NGL I'm really annoyed that we had a period of 3 years were people invested into better software engineering could have gained a headstart on using LLMs for programming/learning.
And now here we are, with tokenmaxxers everywhere and tooling that encourage it, and the rest pointing at it as some kind of gotcha. I do think this is all gonna crumble and we'll by sheer economic virtue have to return to some more appropriate ways of handling the technology.
Might have pushed a bit hard on the refactoring and fundamentals work this weekend…
all these embedded devices not exposing vsync are driving me nuts.
Storm day!
@cevenio@mastodon.social sure, i'll do that too. but for that the code first has to be written. I'm routinely maintaining around 100 repos that i have written with #llms #llm over the last 3.5 years.