Vai Entender
355 posts

Vai Entender
@EntenderVai
Quando alguma coisa fizer sentindo, a raça humana vai ter deixado de existir!
Juiz de Fora, MG Katılım Nisan 2012
34 Takip Edilen13 Takipçiler

@Mayaikos dirt cheap on paper but 21.5m tokens is a lot of testing
English

One Chinese model just topped Arena's frontend leaderboard and beat GPT-5.6 Sol.
Kimi K3 ranked #1 with 1,679 points, ahead of Fable 5 and GPT-5.6. On paper it's dirt cheap: $3 input and $15 output per million tokens.
Reality exposed the gap.
It burned 21.5 million tokens over 1 hour 33 minutes for one 3D globe site, costing $8.66. Fable 5 used 3.5 million tokens in 17 minutes for $11.64. GPT-5.6 finished faster at $5.60.
Kimi delivers solid code but eats tokens like a hog and crawls on speed.
Benchmarks lie. Real projects reveal the truth.
0.1%@sandy4kad
English

@MindfulL205 does it actually work tho or is the squirrel just smarter than me
English

@MindfulL205 squirrels are relentless lol I’d buy this just to watch them fail
English

@MindfulL205 need one of these so bad, the squirrels in my yard are relentless
English

The thread above is selling you quantum trading. The man whose math it runs on spent twenty years proving you don't need a quantum computer to use it.
He wrote the first PhD thesis on quantum algorithms in Spain. He studied under the people who invented the field — Cirac, Vidal, Latorre. His introduction to tensor networks has been cited over two thousand times. He is not a physicist who wandered into finance. He is the physics.
The method is public. He published it, for free, years before a single fund productized it
Barcelona, 2006
Román Orús finishes a doctorate on how to simulate a quantum computer without owning one. The trick is tensor networks — a way to compress an impossibly large quantum state down to the handful of numbers that actually matter, and run it on an ordinary machine.
For a decade it stayed inside condensed-matter physics. Spin lattices, exotic magnets, the kagome problem. Nobody on a trading desk was reading Annals of Physics.
Then the same compression turned out to describe markets
A portfolio of hundreds of assets under real constraints is the same shape of problem as a quantum many-body system: too many configurations to check, most of them noise, a few that matter. Tensor networks eat exactly that. Not by magic — by throwing away the spurious correlations and keeping the structure.
Here is the receipt. In 2019 Orús co-founded Multiverse Computing to sell this as software. By 2025 the company had raised 215 million dollars, filed over 160 patents, and put Bosch and the Bank of Canada on its client list. Its compression tool shrinks a large language model by up to 95 percent with a few percent accuracy loss.
None of it runs on a quantum computer. All of it runs on the GPU you already own.
That is the part the hype threads bury. The edge was never the qubit. It was the structure — superposition of hypotheses, interference that cancels crowded bets, an energy landscape you minimize to find the optimal book. Orús said the quiet part in one line:
the tensor network keeps the structure and throws away the noise.
Now the modern bridge, and it cuts against the thread above. An autonomous "quantum" trading agent scans more patterns, faster. But a tensor network does not make a pattern real — it makes an unreal pattern cheaper to discard. The value is the discipline of compression, not the quantum coat it arrives in.
Quantum-inspired does not mean quantum-guaranteed
The math got faster. The noise did not become signal
You do not need a quantum computer to trade this. You need to know which correlations to throw away.
The method has been free since 2006
DiKrass -X-@Di_Krass_
English

@BedroomFitnes lmao I had to think about this one for a second
English

@ZoneSpace891 wait spain has borders?? i thought they were just the beach and france
English

@TansuYegen this is the kind of innovation i didn't know i needed lol
English

one 20 years old guy saw the shift early.
he didn't build another chatbot.
he built infrastructure.
claude generated the output.
obsidian kept the knowledge.
every project became reusable.
every delivery got faster.
that's where the money compounds.
early builders are locking in clients now.
everyone else will compete on price later.
full breakdown below. 👇
kerazcity@0xkerazcity
English







