Notices by jonny (good kind) (jonny@neuromatch.social)

Incredibly one legitimate use of blockchains right now would be as an append ledger of hashed text diffs submitted in real time as you draft something in order to prove that you actually spent time to write it. Proof of work. It's a proof of work. How is proof of work actually a thing I want to have.

Unfortunately, the prior scam tech tumor is too design pigeonholed for doing scams to support countermeasures against the current scam tech tumor.

Illustrators now just post videos of them drawing stuff to prove they can draw stuff. Is the new social norm for affection and intimacy going to be attaching a video of you writing a message to prove to you care enough about someone to write your own words at them

RE: https://mastodon.social/@adastra1930/115907889643635879

can you imagine the kind of cultural and technological engineering effort it took to make it so installing an unknowable surveillance device in your home is easier than being able to use the abundant internet coursing through reality that you already own to simply be able to ... have your own camera

when you get near the buzzing hive sound of an LLM-driven project you'll know that half the words you see are completely not real and you need to now navigate around like 10 god bots saying incomprehensible nonsense wrapped in overformatted markdown, and the remaining servile human thralls are totally beyond communication by conversation. it's a dead space, a depeopled space, an exclusion zone, a radioactive cesspit soaked in rocket fuel and run on dogshit.

the old way of "approach the throne of the benevolent dictator" sucked, but i never for one second thought that people would flock to "approach the hall of mirrors where everything is real, nothing matters, nobody is in charge, and the firmament mocks you in an always more distant echo" as an alternative.

The problem with "AI" is that it's fundamentally anti-human and that translates very neatly into the most toxic social spaces you can imagine.

I can count a dozen conversations with people in very different domains where half the team is high on AI that are like "we don't actually know what is going on at all over there anymore, we say that there are problems and then they allegedly solve 10 different ones but never actually address anything we say." Like it turns out leaving the realm of human care entirely has some consequences and those consequences look a lot like "rupturing the social fabric needed to sustain any cooperatively shared pursuit"

RE: https://mas.to/@carnage4life/115832534415373032

keeping up to date with "AI" skills is extremely easy actually, the most sophisticated setup i've seen is just a bunch of markdown files begging the necrotic grooves left in the statistical skin of language to pretend like they are different character archetypes wrapped by a (n assemblage of) IDE and CLI tools.

the "skill issue" FOMO trap is just a trap and the question was never "whether they work or not" but rather "how they reconfigure the dynamics of informational power by dramatically consolidating it in the owners of informational capital."

so it's clear this is not goalpost shifting, i am on the record on the Good Blockchain (git) arguing that whether or not the "AI" works or not is irrelevant as of May 2023: https://github.com/sneakers-the-rat/surveillance-graphs/blame/main/_sections/_ideology/vulgarld.md#L17

If you've read any of these fancy markdown commands in any "framework" or in any live repository, you would know that prompt engineering is not a thing and the only stable strategy is repeated begging because there is a fundamental fucking god-blessed information theory-guaranteed injective relationship between the prompt and output domains where you can't have enough intuition about the slot machine to use reliably in practice.

i have burned thousand of hours training, fine tuning, and direct prompting language models, and there is a subtle indescribable art in understanding what they "know" to fashion inputs accordingly, but the combinatorics of language make it absolutely laughable for the sweet spot of "human brain can gain unconscious intuition" to lie exactly in the ungodly chasms of these brazillion parameter models and their outputs.

I regularly work in parallel by making good faith efforts to have the latest models (and the fancy fuckin multi-agent setup with whatever the most hyped markdown lore is on the cursor forums and git tag rankings) do what I do, and the answer remains the same: if i already know what i want to do and have a well worked out spec, and most of the surrounding code is already well scaffolded, then it can do most of the raw work writing the code on the page, but the places where it fails to adapt to local idiom, or when there is something that isn't already pretty much completely done, the process of reading, undoing, and redoing shit code is always more work than just writing the code. Because unless you're totally outside your element in a new language or framework writing the code is not the hard part, and if you are outside your element that's when you specifically become unable to evaluate the LLM output.

the hard parts of code are fucking TALKING TO PEOPLE and the LLMs make that problem A MILLION TIMES WORSE

@arrjay
I am trying to keep the game as simple as possible so that its possible for me to actually finish it lol, but we can have expansion packs. I sort of like the idea of there being absolutely no progression, but a bunch of things that seem like progression

RE: https://neuromatch.social/@jonny/115315321873043743

I have a free afternoon and want to practice making a simple game that I can actually finish, so I was gonna work on horse drink simulator 3000, where you lead a horse to water and perform various minigames to try and make it drink but of course you can't. What are some classic horse drink moments for the interstitial dialogue trees - the original example was about grindset vibe coders worrying about someone stealing their app, but also I'm thinking trickle down economics, blaming shadowy ~powers~ instead of visible billionaires for social problems, etc. What horse drink compatible scenarios have you led someone to water but been unable to make someone drink?

the kinds of ppl i see sticking around on the fedi persistently are those that are very particular about their identity, or the company they keep, or the kinds of things they are exposed to, or the kinds of things they contribute to. that seems like pretty good company to me.

Its like someone locked some people in a room and didn't tell them anything about how AI was being competed over and how computing hardware had changed in 20 years and said "and only you can fix this problem."

Like did they miss how all the companies are basically having to buy off entire towns to produce enough electricity because the amount of electricity it takes to compete in this arms race causes the water to become poison

one day they'll find a dusty pdf on some forgotten floor of the googleplex detailing how they eked out 1% greater profit by derailing the only credible hope for a different internet at scale by telling them that they were tech visionaries and geniuses capable of anything they put their minds to.

mozilla: "we choose the path of the most capital intensive boondoggle we can't afford"
https://blog.mozilla.org/en/mozilla/rewiring-mozilla-ai-and-web/

25 years ago we out competed a buggy web browser by making a less buggy web browser. that's why today we're choosing to engage in the same field of play that the largest companies in the world, which are orders of magnitude more powerful than they were 25 years ago, are betting their entire futures on. while they can rewrite the state of the economy at a whim, dropping nuclear power plants along critical infrastructure and taxation optimums, we have the scrappy heart of a lion.

"i quit my job posing increasing elaborate puzzles for myself and then solving them so that the things i imagine become real and make things i want to happen happen so that i could become addicted to a slot machine that occasionally makes something like the thing i want to happen happen"

it sort of sucks that now we are in an era where no new software fucking works and all changes to working software are to make it not work