ChatGPT - If Stack Overflow and Reddit had a child

  • Want to keep track of this thread?
    Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
    Create account
AI images have been creeping into homewares for a while now, but I’m noticing an odd trend that on the label, it lists a designer as “Ada Ingram”. Clearly the companies think they’re being incredibly clever as the initials are AI. If they’re ashamed of using it, why list a designer at all?
3D33625A-5686-4C2C-9E26-E70E918CA66F.webp
It’s always the kinds of designs boomer women go crazy for.
 
Kimi v2 is a very interesting chinese model. 1T parameters with 32b active ones, so given a fuckton of memory, you could actually run it at home. It's a very natural sounding model, I'd go as far as to say the only natural sounding model there is. Even without prompting for it, it will give you sometimes one-sentence or even one-word replies and it will actually disagree with you on things and state as such. It has in general a very direct (for an LLM) and "non-corporate" way of replying to queries without using many filler words I find very refreshing. It also has an interesting and extensive vocabulary that's quite different from all other LLMs who have seemed to converge on sounding the same. I'm not 100% confident in my ability to clock a properly prompted Kimi v2 as language model in conversation and that is very interesting. I wonder what they did differently. Pretty smart too.
 
I was thinking back on this story I heard about Platea from Dan Carlin. The Greeks were having their big, balls-to-walls final battle, and Greek command was desperate. Normally this culture used the practice of haruspicy (reading of entrails like you would tea leaves), but for weighty, occasional subjects. With everything on the line, though, the generals sat around sacrificing lamb after lamb about each tactical decision. Carlin compared it to if, in the Cuban Missile Crisis, the Joint Chiefs of Staff sat around playing with a Magic Eight Ball.

I think cyberpsychosis shows that drive to grasp for any kind of voice out of the void.


Also, my chatbot said based on something I fed it that I probably grew up steeped in Tumblr, so I guess it's time to kill myself.
 
Should Ai datacenters be forbidden from offering their services outside of their country?

Ok this might sound retarded, but Ai centers are a HUGE drain on natural resources (water, energy) on the local environment and houses in the area experience huge costs increase for water and power.

And Ai is mostly used for YouTube slop and shitty chatbots.

So what we're gonna get is the first world getting electricity costs rise 4000% and water be a scarcity so that pajeets can output Ai slop and farm content.
 
Should Ai datacenters be forbidden from offering their services outside of their country?
No.
Ai centers are a HUGE drain on natural resources (water, energy) on the local environment and houses in the area experience huge costs increase for water and power
The US government is an EVEN BIGGER drain on all natural and unnatural resources on the entire solar system. I recommend forbidding the US government.
 
Kimi v2 is a very interesting chinese model. 1T parameters with 32b active ones, so given a fuckton of memory, you could actually run it at home. It's a very natural sounding model, I'd go as far as to say the only natural sounding model there is.
I have briefly used it myself through Open Router for a bit of RP and it seems far more effective at creative writing and much cheaper as well! My time spent using on Sunday was a bit brief I did not notice a series of “slop-isms” such as “shivers down X’s spine” or “X is a testament to Y”, etc.
Normally If one prompts for a setting with technology or magic the dreaded scent of ozone starts seeping in with DeepSeek, Gemini or, Grok like they’re jonesing for it. I personally find the overuse of ozone is the worst offense with how repetitious the usage is. Maybe it’s too early to say but I feel Kimi K2 might be on par with Claude if not better one’s prompt is well crafted enough.
 
The big problem I'm trying to solve is integrating LLM into a game. The thing with LLMs is things they can do reliably are much easier solved with "conventional" programming. Complex language or pattern problems that are almost impossible to solve with normal programming, they don't solve reliably enough. I'm coming back to this problem on and off over the years since 3.5 showed up but never come to some satisfactory result. My last run at this was putting LLMs into a MUD-like environment, if anyone remembers those. I feel this was my best attempt so far but I also only dabble, I don't invest serious time into it. Things are just changing too fast for that and I have a lot of other things to do.

I have briefly used it myself through Open Router for a bit of RP
I'm not really the RP kind of guy but I threw a scenario into an old writing prompt to see how many GPTisms the model had. Since I today rediscovered the classic "Joan of Arc" by OMD (look for it if you don't know it, great tune) I told Kimi to write me a story from first perspective of a foot soldier in a battle Involving Jeanne d'Arc.

Tonight the air is thick as a butcher’s apron—smoke, sour wine, the copper stink of men who have not washed since Easter. I crouch in the ruined cloister, helm dented so the nasal kisses the bridge of my nose every breath, bill-hook shaft cracked like an old wife’s loom. My gambeson still carries another man’s dried bowel from Meung; it flakes off in brown snow whenever I shift. Jean de la Motte, vintner’s runt—nobody’s son—pressed into mail two sizes too large for me.

Jeanne comes walking between the sleepers, torchlight licking her white harness. She stops, small, chin lifted, eyes bright enough to read scripture by.

—Soldier, she says, voice low, country-plain, —thy heart beats louder than the wardrum. Fear?

I nod.

She smiles—sad, fierce—and presses a strip of linen into my palm: the hem of her banner, browned with her own blood from yesterday’s arrow-graze. —Tie it tight, she says. —When the ladder sways let this keep thee from swaying.

Behind her a veuglaire speaks; stone dust drifts like unholy incense. Somewhere a man whimpers, calling for his mother in no tongue God claims.

Jeanne kneels so our eyes are level. —Name thy sin, she murmurs, so tomorrow thou fightest for something dearer than wine or plunder.

The words crawl out, raw as new cheese: —At Patay I fled. Left my brother Guillaume with a belly wound. Still hear him call.

She closes her eyes one heartbeat—long enough for the river battery to answer again—then lays her gauntleted hand against my cheek. Steel is ice, the palm inside it fire.

—Demain tu cours vers eux, she whispers, then instantly, —Tomorrow thou runnest toward them. Fetch him back. She kisses the cold iron by my ear, brief as sealing wax. —God loves a frightened man who keeps climbing.

She moves on. I knot the linen round my wrist; it is warm from her skin. I sleep kneeling, forehead to stone, every dream a ladder whose rungs are slick with somebody else’s life.

Dawn bleeds across the Loire. Mist hangs thick as curdled milk; the breach gapes—stone fangs round a smoking wound. We form the forlorn hope, sixth ladder. Jeanne mounts first, sword only shoulder-high. Her voice carries calm:

—Forward, for the King of Heaven. The gate is iron and oak; faith is sharper.

We climb.

Third breath: arrows clatter like hail. Fourth rung: Pierre the baker’s son gurgles, slumps; I step on his back and keep rising. My cracked ash shaft groans, fibres peeling like sinew. Sixth rung: a cloth-yard punches through the man below me—he keeps climbing anyway, spear dragging from his thigh like a tail.

Jeanne gains the parapet. An English axes-man hooks her banner, snaps the pole; the sacred silk flutters down inside the wall. She hesitates one heartbeat—only that. I vault the last stride, bill-hook useless; I swing the splintered haft into the Englishman’s face-plate. Bone caves with a wet crunch. We crash onto the wall-walk. I roll atop him, snatch a loose coping stone, bring it down twice. He stops moving.

Jeanne is already up, ragged banner reclaimed. She plants it between us. —The wall is ours—hold!

More French spill over. I feel a punch beneath my ribs—hot, cold, hot. An English dagger has found the gap where coat-of-plates meets belt. I do not see the hand; Jeanne does. Her sword flicks, farmer’s scythe, and the hand drops away still gripping the blade. She catches me as knees soften.

—Stay, she orders, pressing the linen relic against the wound. Blood soaks it at once; her fingers come away shining. Below, the drawbridge drops; our men pour across like water through a broken dam.

I try to laugh; blood bubbles instead. —Told thee I was frightened.

—Et quand même tu as gravi. And still thou climbed’st. She wipes her gauntlet across my cheek, leaving a scarlet streak like war-paint. —God will remember.

She grips my shoulder. —Hold the banner. Even if thy hands slacken, let it not touch the ground.

The fight rolls past, boots drumming. Sounds blur—steel on steel, shouting, hollow thump of gates giving way. Through the crenel I see her white armour bobbing down the stair into the melee, soon swallowed.

I nod, taste iron, cold creeping from boots upward. I slide against the merlon until I sit, banner pole between breast and stone. Blood cools, tacky, fusing linen to wood, wrist to pole—coward, saint, and relic soldered by one slow heartbeat.

The sky narrows to a slit. I count rosary heartbeats, whispering my brother’s name. When the chroniclers come they will find me seated, standard upright, strip of her blood glued to mine, flag and coward become one splintered saint the soldiers swear they saw stand back up.
I haven't tried LLMs for creative writing in a while but I can't really identify major GPTisms. They might have changed since I last looked into that but for me this reads quite interesting. Using an dash to lead in dialog instead of quotation marks is something that's done in french writing, which is a really interesting touch IMO. If it has been done grammatically correctly I couldn't tell you since I don't remember my french.
 
Have you noticed its been over a year and the use cases for AI beyond shitposting hasn't really grown? supposedly they're using it for commercials and animated films but killing off the entertainment industry is not revolutionary enough to invest trillions into this tech.
I think cyberpsychosis shows that drive to grasp for any kind of voice out of the void.
its why humans need religion, or at least something to rely on, its why so many people felt miserable by the late 90s, self actualization is a bitch. its a big reason why fascism was invented by a ww1 veteran and lifelong socialist, once you actually see what people are going through, its obvious that humanity needs a voice to call out to.
 
AI beyond shitposting hasn't really grown?
You're not thinking of the mundane technical applications of AI. Things like HR (even more than resume reading), onboarding, sifting through big data, quick analysis, marketing, note taking etc. Its happening more behind the scenes than you realize. The great money maker these days is the invent some specific AI model for a niche but expensive industry.
 
You're not thinking of the mundane technical applications of AI. Things like HR (even more than resume reading), onboarding, sifting through big data, quick analysis, marketing, note taking etc. Its happening more behind the scenes than you realize. The great money maker these days is the invent some specific AI model for a niche but expensive industry.
I put allot of academic literature into ChatGPT. In one sitting I can put in thousands of pages of very dense information and get really good analysis. I'm just a retard on the Internet, imagine the knowledge banks a large institution can build using the big boy commercial grade models. I'vee also been able to do all kinds of shit i would have never been able to do. on my own in python without the robot. There a 100% chance there something you want to do but cant without AI, no matter how retarded you might be.
 
Hitting the limits of LLM itself, since LLMs are just matrix multiplication algorithms working similar to Markov chains.
I used to mess around with Markov babblebots. Most people could instantly identify them because they're primitive compared to current AI, but there were some people (let's call them retards) who would fall for them every time.

LLMs can generate fairly convincing output. I generally recognize them in any long enough output from the numerous glitches they have, but I have sometimes been infuriated to spend a minute reading something and then suddenly realize it's AI.

The most convincing seem to be those trained on some narrow field of interest if you are only asking within that field, but obviously if you know to stay within that field, you probably already know it's AI.
 
Last edited:
I used to mess around with Markov babblebots. Most people could instantly identify them because they're primitive compared to current AI, but there were some people (let's call them retards) who would fall for them every time.

LLMs can generate fairly convincing output. I generally recognize them in any long enough output from the numerous glitches they have, but I have sometimes been infuriating to spend a minute reading something and then suddenly realize it's AI.

The most convincing seem to be those trained on some narrow field of interest if you are only asking within that field, but obviously if you know to stay within that field, you probably already know it's AI.
My point is that these LLMs are Markov chains except with billions of dollars in budget to increase the likelihood of good text outputs
To insinuate that any intelligence or volition is behind this is just ridiculous to me
 
My point is that these LLMs are Markov chains except with billions of dollars in budget to increase the likelihood of good text outputs
To insinuate that any intelligence or volition is behind this is just ridiculous to me
I agree but nigga you know all the major companies in the US have deals to feed the government's quantum computers that they pretend don't exist. God knows what niggers like Altman have hidden in warehouses in the middle of nowhere. I genuinely believe the AI bubble exists almost exclusively to beat China in the race to quantum supremacy
 
AI Typing Detected.
Nah, too many typos. Honestly, AI writing is just what they taught you to for search engine optimization. Tons of leading questions, tons of stating what you already said and alike. Then make 5 sub-headlines and flesh it all out again.

The fact some "journos" still write this manually is crazy.
 
I used AI to check my grammar and compared it with my original text but lately it has been getting more and more awful at it (I write in Dutch, like God intended). The sentence structures and pacing become more unfriendly to read, it prefers to change every elegant word with its most dumbed down synonym, it likes to use vague descriptors and it doesn't understand ; . I have unironically been going back to the old tested method of reading my texts out loud multiple times, like olden times.

Has anyone else had this problem?
 
Back
Top Bottom