HNS CAPITAL

Kripto lejupslīde tevi nenogalina ar sitienu.
Tā tevi nogalina lēnām: ar cerību, ar sviru, ar domu "tā drīz atgūsies."
Izdzīvot lejupslīdē nav par to, lai nopelnītu daudz naudas, bet gan par to, lai netiktu izslēgtam no spēles.
1. Pieņem patiesību: tirgus var būt slikts ilgāk, nekā tu domā.
Lielākā kļūda, ko pieļauj jauni tirgotāji, ir:
"Šis kritums ir par daudz, tas noteikti atgūsies."
Nē.
Kripto var tirgoties sāniski – krist – izsūkt tevi tukšu mēnešiem, pat gadiem.
👉 Pirmais, kas jādara, lai izdzīvotu, ir pārstāt paredzēt apakšējo robežu.

I’ve probably abandoned more AI ideas than I’ve actually shipped.
Not because the ideas were bad. Most of them still sound reasonable to me honestly. The problem was always everything wrapped around the idea itself.
You start thinking about APIs, deployment, model hosting, data pipelines, inference cost. Then suddenly the actual insight becomes maybe 10% of the work and the rest turns into infrastructure maintenance before you’ve even proven the thing should exist
After enough cycles like that, you stop opening the laptop when a new idea shows up because you already know where the process leads.
That’s mostly why ModelFactory caught my attention.
People keep describing it as “no-code AI tooling,” which is true, I guess. But I don’t think convenience is the interesting part.
I think the more important thing is that the gap between expertise and deployment is getting weirdly small.
Right now a lot of valuable domain knowledge still depends on engineers to become usable software. A cardiologist might understand diagnostic edge cases better than anyone else in the room, but that knowledge usually stays trapped in intuition, notes, pattern recognition built over years. Same with traders, lawyers, researchers, operators inside niche industries
Weirdly, the people who understand a domain best are usually the least equipped to turn it into software themselves.
And historically that meant a lot of the value leaked outward. The people packaging the knowledge technically often captured more upside than the people who actually understood the domain in the first place.
ModelFactory changes some of that dynamic.
You can kind of see where this is going already: connect a Datanet, fine-tune through a dashboard, deploy through OpenLoRA. No real infrastructure stack to manually assemble. No engineering team needed just to test whether your knowledge can become something usable.
The OpenLoRA side matters more than people think.
Normally fine-tuned models become expensive fast because every variation needs compute attached to it. You train a model, you host a model, you pay for that model whether anyone is using it or not. At small scale maybe that’s manageable. But once you start testing multiple variations against different datasets or adversarial inputs, most solo builders hit a wall pretty quickly.
OpenLoRA gets around part of that by sharing the base weights and only loading the specialized layers when needed. So instead of every model behaving like its own isolated system, thousands of specialized models can run on the same GPU much more efficiently.
Without something like that, most people probably never even reach the experimentation stage. They stop at compute cost.
And I think that changes the shape of AI development more than people realize.
Because once building becomes cheap enough, technical execution stops being the main bottleneck. The harder problem becomes judgment.
You can absolutely build a model on weak or noisy data and still end up with outputs that look convincing for a while. That’s probably the dangerous part, especially in markets. Models don’t need to be correct to gain trust initially. They just need to sound coherent long enough for someone to rely on them.
Then reality shows up.
So I doubt the winners in systems like this will simply be the fastest builders. More likely it’ll be the people who understand the limits of their own datasets, the people who know when their model is failing even when the outputs still sound confident.
That’s a very different skillset from pure engineering.
And honestly I think that’s where OpenLedger becomes more interesting than it first appears. ModelFactory, OpenLoRA, permissioned Datanets — separately they look like tooling. Together they start looking more like infrastructure for turning domain expertise itself into something deployable.
Maybe that works at scale, maybe it doesn’t. I’m not even fully sure yet.
But the direction feels real.
AI development keeps getting compressed into workflows that non-engineers can participate in directly, and if that trend continues, the important question probably stops being who can build models.
It becomes whether someone actually knows something valuable enough to model in the first place.
$OPEN
#OpenLedger @Openledger

Ir kaut kas, kas mani nepārtraukti traucē par to, kā tiek ietverts DeFi tirdzniecība.
Visi runā par decentralizāciju, it kā tā būtu produkts. Tā nav. Decentralizācija ir ierobežojums. Patiesais produkts ir izpilde — ātrums, cenas, privātums.
Gadu gaitā, on-chain tirgotāji ir maksājuši vairāk, lai saņemtu sliktāku izpildi nekā centralizētās biržas piedāvā mazumtirdzniecības lietotājiem bez maksas. Augstākas maksas. Publiska pasūtījumu redzamība. Sliktāki izpildes rezultāti. Mēs to pieņēmām, jo nebija reālas alternatīvas. Mēs to saucām par pašglabāšanas izmaksām.

Gadiem ilgi veiktās pētījumu uzskatīju par bezjēdzīgiem. Izrādās, ka tie nebija.
Manā datorā ir mape ar nosaukumu “vecie pētījumi.”
Iekšā ir simtiem failu: on-chain dati, tokenu sadalījuma analīze, vaļu maku uzvedības modeļi, piezīmes no naktīm, kas pavadītas, skatoties mempool līdz 2 no rīta. Lielākā daļa no tiem nekad vairs netika izmantota. Daži faili, ko es domāju, ka paliks tur uz visiem laikiem.
Es nekad nedomāju, ka tās kādam citam varētu būt vērtība, izņemot mani.
Līdz pagājušajā mēnesī izlasīju daļu no ES AI likuma.
Ir viens sīkums, ko, manuprāt, lielākā daļa tirgus novērtē par zemu: uzņēmumiem, kas izmanto AI, arvien vairāk būs jānodrošina datu izcelsme — kuri dati radīja kuru rezultātu, kurš tos ieguldīja, kurš piekrita to izmantošanai, auditable un pārbaudāmi katrā solī.

Es jau gadiem strādāju pie AI, neiegūstot nevienu centu.
Katrs meklējums, katrs klikšķis, katra papildus sekunde, kas pavadīta lapā, kļuva par apmācību datiem kāda cita modelim. Neviena paziņojuma. Neviena maksājuma. Tikai pakalpojumu noteikumi, kurus neviens nelasa.
Neviens to nepauž kā ekstrakciju, jo produkts joprojām jūtas bezmaksas.
Tirdzniecības darījumiem tas iet vēl dziļāk. Pasūtījumu plūsma, ieejas laiks, izmēru modeļi — lietas, kas prasīja gadus, lai izstrādātu — var kļūt par uzvedības mācību datiem, tiklīdz tās nonāk platformā ar AI slāni.

the problem OpenLedger is solving isn’t technical. it’s economic. and it’s been building for years.
I’ve been thinking about a researcher I read about once. spent a decade building domain expertise in rare neurological conditions. published papers, contributed to forums, wrote detailed case analyses that nobody outside the field would ever read. then one day noticed that a major AI health platform was giving advice that sounded exactly like her work — phrasing, frameworks, specific clinical distinctions she’d developed over years.
she received nothing. wasn’t credited. wasn’t asked.
nobody broke a law. that’s the disturbing part.
the way AI training works, her knowledge entered a model the same way water enters the ocean — absorbed, dissolved, indistinguishable from everything else. no receipt. no trail. no way to prove contribution, let alone claim compensation. the intelligence appears from a black box and the people who built that intelligence watch it get monetized by others.
I keep thinking about what this does to incentives over time. not dramatically. quietly. the researcher publishes a little less. keeps the most valuable insights private. decides the cost of open contribution outweighs the benefit. multiplied across thousands of domain experts making the same rational calculation, the commons that AI draws from starts to thin.
not because of a decision. because of a structure.
that’s the problem OpenLedger is actually solving. the current AI economy is built on a foundation that quietly destroys the incentive to produce the knowledge it depends on. and nobody notices until the damage is structural.
what makes this harder than it looks: you can’t fix it by asking AI companies to behave differently. the incentive to extract without compensating is too strong and the mechanism to track contribution too absent. the only real fix is infrastructure that makes compensation automatic — where every inference that reaches for your data triggers a payment, not once, but every time that knowledge gets used.
that’s what PoA is. not a feature. an attempt to make extraction economically equivalent to contribution — to close a loop that the entire AI economy has been running open since the beginning.
I don’t know if OpenLedger pulls this off. the timing is genuinely uncertain. six months since mainnet. September brings supply pressure before adoption is proven. the problem they’re solving is accelerating faster than any protocol can address alone.
but I find myself caring about whether this works in a way I rarely care about crypto projects. not because of the token. because the alternative — an AI economy that keeps consuming human knowledge without compensation until the incentive to produce that knowledge quietly collapses — is a problem that doesn’t announce itself until it’s already too late to fix.
that felt worth writing down.
do you think the incentive to contribute openly to AI systems is already being affected — or is that still years away?
it’s that we may slowly stop giving it anything worth learning from.
@OpenLedger $OPEN
#OpenLedger