roistrdn

From the distributed dream to the digital feedlot

I'll keep testing models as I'm able to, my main laptop is an m2 but I never downloaded any llm on it, I'll venture to do it, with 16gb of ram I can test up to 13b

I have an i7 laptop where I test models with ollama, mistral 7b and phi have run well, but they fall short and the times get long, testing paid models I do have to agree with you that anthropic's have been the best, but they really go off the rails and get too lost, I hope to be enjoying a local llama 3 very soon

Same pattern, currently saving up for a server with local models, the only way to break the cycle


gaming

What is the best Nintendo handheld?

gamer000gaming

For retro gaming on Android, if I can do it on my smartphone I won't complicate things, if not, then I'll use the Switch
I've played a lot of Mario Maker 2

We thought we were participating and building. Bullshit. We were the fuel: cheap liquidity, telemetry, and prompts. Welcome to the post-utopian era where your wallet and your prompts are the biomass the new feudal lords grind down to press into financial products and Enterprise plans. The dream of digital sovereignty has been recycled into barns segmented by ARPU, rationed “access,” and calibrated extraction.
We were sold two parallel epics: the one about financial decentralization and the one about “AI for everyone.” Two stories of technological emancipation. In both cases the arc looks way too similar: romantic phase, chaotic explosion, financialization, infra‑regulatory capture, and finally the stage where you, the user, stop being the protagonist and become input. Granular liquidity in crypto. Traffic and prompts in AI. You are throughput. Not a sovereign customer—inventory optimized for yield extraction.
While memecoins tied to political or media figures reach absurd market caps without cash flow or utility, protocols actually building technical foundations get displaced because distribution of attention and speculative expectation weigh more than intrinsic value. Narrative > technical merit. Same thing you’ll see in AI with “alignment” branding versus model transparency.
The end-user layer is emptied of novelty. The model that scales is B2B2C: you build for institutional desks, custodians, or regulated funds; they bundle the experience and milk their retail herd. You operate as structural liquidity and compliance layer. And here comes the variable we ignored before: licenses, whitelists, KYC, permissions. Compliance went from nuisance to moat. You can fork code; you can’t fork a license or banking access. Base (the L2 leveraged by Coinbase’s regulated position) and the expansion of a stablecoin like USDC illustrate this: legal legitimacy turned into a TVL magnet.
A couple cycles ago the conversation revolved around consensus, ZK proofs, L2 scaling, privacy. Today serious liquidity concentrates where there’s a narrative digestible by an investment committee: asset tokenization (RWA), structured instruments, capture of yield flows. The TPS race and TVL inflated via incentives are already historical residue. Capital efficiency matters: how many times can I rehypothecate the same dollar of collateral, how to isolate and commercialize the temporal component (principal vs yield), how to internalize MEV for the protocol instead of letting it leak invisibly.
It’s not the magical glow of cryptography gaining traction, it’s financial engineering applying techniques familiar to traditional traders: implicit forward curves, basis trades, optimized carry.
Artificial intelligence repeats the pattern with different raw materials.
Same principle: value extraction through control of the orchestration layer.
The general user feeds metrics (DAU, tokens generated, use cases), produces an implicit dataset of prompts to calibrate safety heuristics, response ranking, and monetization strategies, and serves as a statistical shock absorber to justify more hardware investment that later gets reserved for high‑margin clients.
The user senses or suspects gradual degradations: less precise answers, variable latency, silent cuts in context length. Meanwhile Enterprise clients enjoy privileged lanes: separate queues, higher precision (less quantization), stabler sampling. Nobody announces it on a banner, but the economic incentive is trivial: prioritize expensive compute where the contract justifies cost; amortize the rest with compressed versions.
Both sectors subject users to novelties just attractive enough to maintain throughput, but calibrated not to hand over so much value that margins crack. You are efficient cattle: you consume interface, produce signal, generate spread.
Initial decentralization or openness functions as the bootstrap phase: attract critical mass under an ideal, identify monetization vectors, then close the gates gradually.
Sovereignty isn’t paranoia: it’s designing a gradient where the marginal cost of extracting value from you without consent goes up.
Your knowledge is your ammunition.
                                        
                                         Thanks for your time

Education


We thought we were participating and building. Bullshit. We were the fuel: cheap liquidity, telemetry, and prompts. Welcome to the post-utopian era where your wallet and your prompts are the biomass the new feudal lords grind down to press into financial products and Enterprise plans. 
The dream of digital sovereignty has been recycled into barns segmented by ARPU, rationed “access,” and calibrated extraction.
We were sold two parallel epics: the one about financial decentralization and the one about “AI for everyone.” Two stories of technological emancipation. In both cases the arc looks way too similar: romantic phase, chaotic explosion, financialization, infra‑regulatory capture, and finally the stage where you, the user, stop being the protagonist and become input. Granular liquidity in crypto. Traffic and prompts in AI. You are throughput. Not a sovereign customer—inventory optimized for yield extraction.
While memecoins tied to political or media figures reach absurd market caps without cash flow or utility, protocols actually building technical foundations get displaced because distribution of attention and speculative expectation weigh more than intrinsic value. Narrative > technical merit. Same thing you’ll see in AI with “alignment” branding versus model transparency.
The end-user layer is emptied of novelty. The model that scales is B2B2C: you build for institutional desks, custodians, or regulated funds; they bundle the experience and milk their retail herd. You operate as structural liquidity and compliance layer. And here comes the variable we ignored before: licenses, whitelists, KYC, permissions. Compliance went from nuisance to moat. You can fork code; you can’t fork a license or banking access. Base (the L2 leveraged by Coinbase’s regulated position) and the expansion of a stablecoin like USDC illustrate this: legal legitimacy turned into a TVL magnet.
A couple cycles ago the conversation revolved around consensus, ZK proofs, L2 scaling, privacy. Today serious liquidity concentrates where there’s a narrative digestible by an investment committee: asset tokenization (RWA), structured instruments, capture of yield flows. The TPS race and TVL inflated via incentives are already historical residue. Capital efficiency matters: how many times can I rehypothecate the same dollar of collateral, how to isolate and commercialize the temporal component (principal vs yield), how to internalize MEV for the protocol instead of letting it leak invisibly.
It’s not the magical glow of cryptography gaining traction, it’s financial engineering applying techniques familiar to traditional traders: implicit forward curves, basis trades, optimized carry.
Artificial intelligence repeats the pattern with different raw materials.
Same principle: value extraction through control of the orchestration layer.
The general user feeds metrics (DAU, tokens generated, use cases), produces an implicit dataset of prompts to calibrate safety heuristics, response ranking, and monetization strategies, and serves as a statistical shock absorber to justify more hardware investment that later gets reserved for high‑margin clients.
The user senses or suspects gradual degradations: less precise answers, variable latency, silent cuts in context length. Meanwhile Enterprise clients enjoy privileged lanes: separate queues, higher precision (less quantization), stabler sampling. Nobody announces it on a banner, but the economic incentive is trivial: prioritize expensive compute where the contract justifies cost; amortize the rest with compressed versions.
Both sectors subject users to novelties just attractive enough to maintain throughput, but calibrated not to hand over so much value that margins crack. You are efficient cattle: you consume interface, produce signal, generate spread.
Initial decentralization or openness functions as the bootstrap phase: attract critical mass under an ideal, identify monetization vectors, then close the gates gradually.
Sovereignty isn’t paranoia: it’s designing a gradient where the marginal cost of extracting value from you without consent goes up.

Your knowledge is your ammunition.

                                         Thanks for your time

movies

Which is your favorite Jurassic Park movie?

samhainsam

What were your favourite cartoons?

![](https://m.stacker.news/100770) I can't think of another one

Music

Daily Discussion Thread ~Music Saloon!

Public_N_M_E

In my youth I rejected everything that didn't have heavy guitars, during my journey of opening up to other styles, albums like this served as a bridge and I still listen to them today 

https://youtu.be/_KLfvO7O_Es?si=EzomVfGAfMIAZUi1

Spectrum 128K

I think now there are more learning opportunities, it's more within reach

For years, I had a Sinclair ZX Spectrum 128K box staring at me from the top of the closet. 
I was small and curious, so I always asked about that box. The answer was always the same: "It's your brother's, but it doesn't work." Turns out my brother had bought it back in 1987, when I wasn't even born yet. But not knowing how to connect it to the TV, he sent it to the closet where it stayed for over a decade.

Years later, I took it down from the closet. I grabbed the manual and managed to connect it to the TV; it wasn't that hard. The machine took a while to boot up, but I waited watching those colored stripes and listening to the system beeps, until the RoboCop game loading screen appeared. But the most interesting thing was something else: reading the manual, I discovered stuff I had no fucking clue about: commands, syntax... I didn't know what the hell a programming language was.

The thing is, my brother didn't bother to read the manual. A machine that cost a fortune in the late 80s, with real programming capabilities, ended up in the closet. The Spectrum 128K wasn't just a console; it was over ten years of lost learning. That machine could have been teaching programming logic, problem-solving, algorithmic thinking...

Curiously, the same brother who abandoned the Spectrum for not reading the manual was the one who bought me both the Game Boy Color and, later, the PlayStation 2 at launch. The Game Boy Color became my usual companion with Pokémon Blue and Wario Land. The PlayStation 2 arrived on launch day with a brutal technological leap, but it was already a machine for consuming content (as a kid, I was delighted).
Playing Pokémon Blue, like any kid of that era, I heard rumors about Mew. That Pokémon that supposedly could be caught with "weird tricks": go to Cerulean City, position yourself so a trainer detects you, press START just when he's about to talk to you, use Teleport to escape, and then fight the Slowpoke of another specific trainer.
As a kid, I would have loved to know that trick was a glitch so I could show off to my friends. What really happened was that by pressing START at the exact moment, you interrupted the trainer's loading process in memory, and by using Teleport you kept that corrupted data. Then, when fighting the Slowpoke, the game read that corrupted data as if it were a valid Pokémon, and since the Slowpoke had a specific ID that matched Mew, it appeared. It was a brilliant memory exploit.

The reflection is this: how much knowledge is lost through intellectual laziness, and even more, how many opportunities are lost when a child asks with curiosity and that curiosity isn't exploited. My brother had in his hands a machine that could have taught him programming fundamentals in 1987, when that knowledge was gold. Instead, he preferred to declare it "broken" out of laziness.

But beyond that, when I was little and asked about that box in the closet, the answer was always "it doesn't work" instead of "let's see what this is." When a child shows curiosity about something, exploiting that curiosity can change their life. Instead of closing doors with vague answers, entire worlds can be opened.

That abandoned Spectrum represents something deeper: the difference between consuming technology and understanding it. Bitcoin also represents that dynamic: the difference between being a passive user and being someone who can create with the tools they have. The real cost of not reading manuals isn't just a machine in the closet; it's a decade of lost advantage in the programming world.

The lesson isn't just about technology, but about intellectual curiosity. And above all, about the responsibility we have when a child asks what is that? because that question can be the beginning of a completely different life.

No resentment, just learning.

This guy has composed a lot of interesting stuff, I always come back to listening to him

Wario Land 3 for Game Boy Color

k00b

Wario Land 3 and Pokemon Yellow were my favorite games   on my Game Boy Color which I still have 🕹️

Aphex Twin completely blew my mind and opened me up to genres I never knew existed. He changed how I listen to music forever.

https://youtu.be/6yy-Es6rp3U?si=LX4CqF8QmB9HoXl7