Undefined Reveries

[tpwrtrmnky]

youtuber who does old console ROMhacks for a hobby: i made the n64 do raytracing just to see if i could and it does five thousand FPS or something

triple-A game studio: if you're experiencing stuttering on the lowest settings, consider just not being poor? your graphics card is five months old??

[tpwrtrmnky]

guy whose mission is to defend corporations: nooooo you don't understand, it's not a fair comparison because the youtube guy is doing small-scale experiments motivated by an in-depth understanding of the hardware, while triple-A studios have a scope set by executives who don't know how it translates to work and just fired all their devs with enough experience to optimize

[tpwrtrmnky]

it's unfair to expect industry devs to do the kind of clever optimization that can be achieved by an experienced programmer who's not sleep deprived and has seen their family and friends in the last three years

[undefinedreverie]

sometimes i just look at the current state of gaming and just realize "Wow this is hell".. like im generally not a "good ol days" guy but like man back when game were still made by people who were passionate about making games was such a good time. never say never but i dont think this current landscape can even make any good games

[nyancrimew]

im still a virgin actually if you consider that sex isn't real

[undefinedreverie]

I'm still a virgin actually if you consider my soul is pure

[raakhti]

what is HAPPENING

[moonglade-poetess]

Hey guys, with so much love, and as someone with an actual English degree:

Please just use Sparknotes if you're going to do this. I get it. I do. But chatgpt or other genAI shit doesn't actually know what's important for you to know, and in some cases it might fully make shit up. Use sparknotes. Failing that, talk to someone who did read it. I'm begging.

[the-barefoot-hatter]

oh my god, there's whole books series that ALREADY do that. Have they never heard of the No Fear series published by Sparknotes? It puts a plain everyday english translation on the opposite page from the original text, plus character lists with descriptions and helpful commentary.

stop using AI for things that already exist!

[undefinedreverie]

The concept of people - not even attempting to read the original book - not looking up a readable English version - not even looking up a reading on youtube - not listening to one of those short form summary of book and just go straight to AI is so fucking Baffling it makes me want to kill myself truly... LIKE this is what chatgpt give you

What do you even get from this? like whats the fucking point??? IT DOESNT EVEN KNOW THE PLOT OR LIKE ONE OF THE BEST SCENE IT JUST MADE SHIT UP THAT SOUND VAGUELY TRUE??? AND WHAT THE FUCK DO YOU MEAN "non-copyrighted" THAT BOOK IS IN THE PUBLIC FUCKING DOMAIN!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

[shadowblade8192]

windows users will be like "oh switching to linux? that pales in effectiveness to my strategy, making tech companies stop making their products worse" and then not make any tech companies stop making their products worse

[undefinedreverie]

We should certainly try by rewarding good product and just not using bad products.. but like that assumes that a "good" product exist.. so like you cant leave windows for a better commercial OS like windows because it doesn't exist and since you are not leaving windows no one will even try to make a better commercial OS.. so line just bit the fucking bullet and switch to Zorin OS or AnduinOS its literally just windows

[undefinedreverie]

Just realized how trivially easy it is to pirate anime through the Terminal

Like its asks what anime you want to watch and what episode and THAT IT? it literally streams it into your native video player????? like no ads bullshit no having to look through 10 differents sites NO having to fiddle with servers and mirror???? NO having to wait for the video to buffer abysmally slowly?? and NO ANNOYING REDIRECTS???? like you dont even need to be tech savvy to use? why did no one tell me about this before??

[coelasquid]

Sometimes we’re unsatisfied with a thing we made because when it only existed in our head, we saw all the things it could have been and when it’s done we know all the things that it isn’t, but we can’t see the way it expands into a million new things when someone else unpacks it in their head.

[coelasquid]

Sometimes I wonder how many of the 90 thousand people who have interacted with this post have been like “I think I’ll go to OP’s blog and see what other lovely nuggets of wisdom they have” and then get hit with a wall of nothing but homoerotic drawings of the men from the Yakuza franchise.

[undefinedreverie]

and how many of you went to their profile after reading that theres yaoi there?

[piratesexmachine420]

The most important fact to internalize about large language models is that they’re profoundly stupid. They will lie, they will overstate, they will misunderstand, and they will be biased. There is no question you can ask that will never get a wrong answer. Never do anything with an LLM’s “opinion” that does not centralize this fact.

The second most important fact to internalize is: they cannot tell you if they’re actually confident, or are just answering as if they were confident. On some level they’re the same thing. What’s the last time you asked ChatGPT a question and replied by asking for your help? Has it ever happened? There’s simply no possible way to know if an LLM has a real understanding of a problem if you don’t have that yourself. Never make a decision on the basis of an LLM’s input without considering this too.

Evangelists may attempt to convince you that the solutions to these problems are only a matter of time.

I disagree.

There’s only one real way to get value (and I mean ’value’ here as coldly and pragmatically as possible) out of something that is architecturally incapable of meta-cognition:

Only ever bet your time. If a wrong answer will cost you more than an hour of frustration, reevaluate.

Stay transactional. This isn’t a conversation, you are trying to take advantage of The Man’s several billion ill-spent dollars.

Keep conversations as short as possible and delete them as often as possible, to serve points 1 and 2. If you want a permanent record of what you’ve learned, take notes somewhere else.

If you can’t keep to those rules, try reconsidering what it is you’re trying to do. Break the problem down into smaller parts, where the answers are easy to verify. “How do I make Flappy Bird in Objective-C” will only end badly, “how do I compile Objective-C with gcc” will either be right, or it’ll breadcrumb you to the right answer anyway, or at worst will just fail really obviously.

Don’t be afraid to lead it on, too. If you can’t figure out what library you need to use to draw a 3D graph in python, ask it to write a python program that draws a 3D graph, but don’t copy-paste the code. We’re here for facts. Facts like “what libraries is it using” or “how is it structuring these API calls” or “what does it mention in its docstrings”.

Once you’re done, write down what you’ve learned – or at least take a mental note of what’s happened – then delete the conversation and go again. The less it knows about what you’re doing, the more general its responses will be; and the more general its responses are, the easier it will be verify them.

Eventually, you’ll have a solution and an understanding of the problem. That’s what really matters.

[undefinedreverie]

#preach

[piratesexmachine420]

The most important fact to internalize about large language models is that they’re profoundly stupid. They will lie, they will overstate, they will misunderstand, and they will be biased. There is no question you can ask that will never get a wrong answer. Never do anything with an LLM’s “opinion” that does not centralize this fact.

The second most important fact to internalize is: they cannot tell you if they’re actually confident, or are just answering as if they were confident. On some level they’re the same thing. What’s the last time you asked ChatGPT a question and replied by asking for your help? Has it ever happened? There’s simply no possible way to know if an LLM has a real understanding of a problem if you don’t have that yourself. Never make a decision on the basis of an LLM’s input without considering this too.

Evangelists may attempt to convince you that the solutions to these problems are only a matter of time.

I disagree.

There’s only one real way to get value (and I mean ’value’ here as coldly and pragmatically as possible) out of something that is architecturally incapable of meta-cognition:

Only ever bet your time. If a wrong answer will cost you more than an hour of frustration, reevaluate.

Stay transactional. This isn’t a conversation, you are trying to take advantage of The Man’s several billion ill-spent dollars.

Keep conversations as short as possible and delete them as often as possible, to serve points 1 and 2. If you want a permanent record of what you’ve learned, take notes somewhere else.

If you can’t keep to those rules, try reconsidering what it is you’re trying to do. Break the problem down into smaller parts, where the answers are easy to verify. “How do I make Flappy Bird in Objective-C” will only end badly, “how do I compile Objective-C with gcc” will either be right, or it’ll breadcrumb you to the right answer anyway, or at worst will just fail really obviously.

Don’t be afraid to lead it on, too. If you can’t figure out what library you need to use to draw a 3D graph in python, ask it to write a python program that draws a 3D graph, but don’t copy-paste the code. We’re here for facts. Facts like “what libraries is it using” or “how is it structuring these API calls” or “what does it mention in its docstrings”.

Once you’re done, write down what you’ve learned – or at least take a mental note of what’s happened – then delete the conversation and go again. The less it knows about what you’re doing, the more general its responses will be; and the more general its responses are, the easier it will be verify them.

Eventually, you’ll have a solution and an understanding of the problem. That’s what really matters.

[unfortunatepresence]

git girlie who's new to dating so she added "--recurse-submodules" flag just to be safe, but now she's in a polycule with her gf and all of gf's friends

[85-rend]

i dont think "doing it scared" is helping anything to be honest

[85-rend]

to be clear i meant this about things you are supposed to do to enjoy that being scared and stressed about fucks up the entire thing. "do it scared" is maybe good advice for shit like grocery store and doctor phone calls but i'm talking about like hobbies and social interactions.

[undefinedreverie]

the "face and conquer your fears" thing only works when the fear is irrational but like for example i want to do public speech and im scared that i might mess it up.. that is a completely rational fear that stems from the part of our brain that has learned avoid social humiliation (cuz that means death) and i should listen to it and prepare practice in safe environments etc.. but like if i just went there scared?? it will be a disaster

[pondering-code]

Case 10. Pride

Geekiness: slightly geeky

After a cursory inspection, the Java master decreed that each monk in the Elephant's Footprint Clan would henceforth catch his own fish, harvest his own rice, sew his own robes, and cobble his own shoes.

Within a month, the monks were half-naked and severely malnourished. Deadlines began to slip. The abbot implored the Java master to reconsider his decision.

"Very well," said the Java master. "Every Thursday, have them also burn their robes and shoes, and cast their rice and fish into the river."

The abbot asked what transgression his clan had committed to merit such punishment.

"Punishment?" said the Java master. "I intended only to reward. Do not the monks of the Elephant"s Footprint relish their independence and self-sufficiency? Their code tells me as much:

"Each has authored essentially the same Data Access Objects and Value Objects as his neighbors. Each has crafted his own String utilities and configuration file parsers. Each has taken exquisite pains to design, implement, integrate, test, debug, repair, and extend that which he might merely have consumed from one of his brethren.

"Shall I infer incompetence? If so, then every last monk of the Elephant's Footprint must be cast out of the monastery.

"Shall I infer willful ignorance? If so, tradition demands that they blind themselves with nettles, and _then_ be cast out.

"Shall I infer contempt for the preciousness of time? If so, the dungeons await...

"But no," concluded the Java master. "I shall instead infer pride—pride taken by each that his code is superior to that of his brethren. Pride is one of the Four Seeds of Greatness, and must be encouraged. Let the Elephant's Footprint hone its skills with shears and spears, with needles and sickles, until every monk may be likewise proud of his fine robes and the bounty on his plate!"

"There will be little time left for coding," observed the abbot. "And the Emperor is quite literal in his interpretation of the word deadline."

The master nodded. "The merchant, lately ruined, knows only the pangs of loss. Whereas the old hermit extols the virtues of austerity, and will not change his state. Yet are these not the same man, separated only by a gulf of years?"

Thus was the Elephant's Footprint spared twice and corrected.

⋅⋅⋅⋅⋅⋅⋅⋅⋅ My reading ⋅⋅⋅⋅⋅⋅⋅⋅⋅

See: Not-Invented-Here Syndrome. A reluctance to adopt solutions from others due to: being territorial; pride; fear of dependency and loss of control. Common in tech.

[pondering-code]

Case 18. Necessary Features

Geekiness: slightly geeky

This same monk of the Elephant's Footprint was poring over the code created by his clan. There were several competing implementations of a particular Data Access Object; at present he was attempting to decide which would become the definitive one that his clan would adopt.

"Each implementation has its own particular virtues and vexations," said the monk to the Java master. "And each has features the others lack. I must consign all but one to the winds of oblivion. How am I to choose?"

"I cannot answer when I am thirsty," said the master. "In the cabinet beside you are three wooden cups. Fill one from the river, the second from the temple well, and the third from the water-seller in the village. Bring all back to me without a drop spilled, and I will offer you what help I can."

The monk did this, and set all three vessels before the master.

"What have you learned?" asked the master.

The monk bowed. "If you give a coin to the water-seller in the village, he will strike a gong twice: once before filling your cup, and again when you have had your first sip. If you fill your cup from the temple well, no gong sounds but the bucket is heavy. If you dip your cup in the river, the fish are troubled and the river-god must be appeased by a gift of ten white stones."

"I can drink but one of these," said the master. "Which will I find the most pleasing?"

The monk considered carefully. "The river was cold, but not altogether clear. The temple well I have drunk from in the past: its waters are bitter with limestone. And though I am not acquainted with the water-seller, I heard his gong many times today. On this evidence, I would choose the cup from the water-seller."

"A most sensible conclusion," said the master. "Indeed, I have visited this same water-seller every morning for eight years, and found the experience perfectly satisfying each time. Which is the vessel that he filled?"

The master sipped from the indicated cup, and his face fell in disappointment. "Young brother," he said. "You have completely misunderstood my thirst. Where is the sound of the gong?"

———

The master's thirst is unsatisfied yet the monk has his answer— what sound was the vessel that poured it?

The monk would have fared better if he had set a brass censer on the window-ledge and merely waited for rain.

⋅⋅⋅⋅⋅⋅⋅⋅⋅ My reading ⋅⋅⋅⋅⋅⋅⋅⋅⋅

The master wants not only to drink, but to hear the gong. Sometimes the requirements are not obvious. So it is in software.

Study people's use cases carefully when implementing breaking changes.

[nolziene]

paradise regained

[vivinky]

From Could AI Slow Science? Sayash Kapoor and Arvind Narayanan

AI could short circuit the process of building human understanding, which is essential to scientific progress ————

AI might prolong the reliance on flawed theories.

How was the geocentric model of the Universe overturned in favor of the heliocentric model — the model with the planets revolving around the Sun? It couldn't be resolved by comparing the accuracy of the two models, since the accuracy of the models was similar. Rather, it was because the heliocentric model offered a far simpler explanation for the motion of planets. In other words, advancing from geocentrism to heliocentrism required a theoretical advance, rather than simply relying on the more accurate model.

(Note: when the article says the models have similar "accuracy," it means predictive/mathematical accuracy, not which model is closer to reality.)

This example shows that scientific progress depends on advances in theory. No amount of improvements in predictive accuracy could get us to the heliocentric model of the world without updating the theory of how planets move. (...)

If AI allows us to make better predictions from incorrect theories, it might slow down scientific progress if this results in researchers using flawed theories for longer. In the extreme case, fields would be stuck in an intellectual rut even as they excel at improving predictive accuracy within existing paradigms.

Thought this was interesting. This concern was floating in the back of my mind when using AlphaFold and this essay put it into words perfectly.

[undefinedreverie]

im going to print this and put in on my office wall

[dogpuppy]

[undefinedreverie]

THEY ARE TRYING THIER BEST!!!!!

[tslurwithbible]

I much prefer to use my limited programming ability for evil, like putting yuri on a calculator. I just need some time...

[undefinedreverie]

i fail to see any evil in your proposal

[c17rus-deactivated20260218]

My son running on the Pi Pico but the training data is too small to actually work so I'm go write 5k line of conversations to make it relatively accurate

Its slowed to every 2 seconds cus its impossible to read at its full 100 runs a second

He's stupid, just really really fast :3

[undefinedreverie]

pardon?