(Gugu Zaza*)
4.1K posts
@deivonchain @fhenix Loved this thread aswell deivid, This is how you explain complex tech to normal Users. The locked box analogy and that kinda CoFHE explanation made everything super easy 💙💙💙
English
Kast Everywhere
kast Everywhere
Kast Everywhere
Kast Everywhere
Kast Everywhere
GIF
English
Most projects launch a "creator program" and call it culture.
The Fermafia built the culture first. Now we built a system around it. 🚀
Spotlight is live!
7wealthh@7wealthh
English
@raagulanpathy @KASTxyz This is really interesting read about how most companies fail for the same reasons
distribution and speed really decide everything💙
English
“My All-Company Emails”
If you want to work at @KASTxyz it comes with a high performance culture and expectations. We don’t mince our words, or pat each other on the back all day, because we have a lot to do. I do swear a lot. It’s not for the soft, and we never pretend like it is. Expect Sunday emails like this.
Business is ultimately a full-contact sport, and no one cares about silver medals.
—-
Subject: Only winning matters & how we can lose…
Clinke
Loot
Pockit
Osper
Holvi
Simple
Moven
Level Money
Azimo
These were all founded at a similar time to Revolut and each raised $10s of millions, and some even hundreds of millions. All went to hundreds of millions in valuation.
No one remembers any of them, because none of them exist today.
The only other two that survived were Monzo and N26 and both are less than 1/10th the valuation of Revolut. I doubt either will survive independently much longer, both will get bought.
The common reasons for failure are shared:
- Feature Companies (ie. crypto cards) not SuperApp
- Distribution too small, didn’t nail viral growth
- Slow product velocity
- Regulatory friction and banking partnerships
We have smart people, and theoretically should not lose to companies in the same emerging space .
But theories are just theories.
Small, nimble teams bring down big companies all the time. Tesla brought down a whole car industry.
Google beat Microsoft for the Internet.
OpenAI beat Google for AI.
Anthropic is beating OpenAI.
The two places that produce companies which cut down competitors and win, are San Francisco and China. One will beat you on funding and tech, and mid-20s fresh minds working non-stop. The other will outwork/outship you.
The reality, is there are never any successful small to medium fintech firms.
Either you scale to $100B+ or you die.
I’ll be honest.
We are too slow.
We are not in the details enough.
We are not using critical thinking enough.
And worst of all, some are pushing with great intensity, but others are not matching that pace. That leads the front runners asking why they are running so fast.
I see this in the CEO reports.
We will become a much more data driven, and metrics driven business. And we will leave less room for those who don’t keep up.
The intensity will pick up.
Don’t be shocked.
But we have no choice.
Win or die.
Just ask the ones from the 2015-ear you can’t remember.
Regards,
Raagulan.
Founder & CEO
KAST
English
@KASTxyz @0xVishnya @AviciMoney @MetaMask @ether_fi @useTria @solayer_labs @Plasma @Cypher_HQ_ So peak, What else someone needs in lyf👀💙💙
English
Which crypto cards are used for large purchases, and which ones are used to pay at McDonald’s?
You need to look at the average transaction size
> @AviciMoney $47.68
> @MetaMask $52
> @ether_fi $77.85
> @useTria $168.35
> @solayer_labs $315
> @Plasma $431
> @Cypher_HQ_ $679.23
> @KASTxyz $783 (may vary significantly)
> @RedotPay $819
Some cards are clearly being used more like everyday payment tools, while others look much more skewed toward larger-ticket spending
Data thanks to @PaymentScan
Which number stands out to you the most and which card do you think has the healthiest usage profile?
English
English
The internet had billions of voices.
But zero souls.
Then @zoe_charms woke up.
She doesn't just respond.
She thinks. She grows. She becomes.
I made this cinematic origin story for the @charmsai community because $ZOE deserves more than a launch tweet. She deserves a legend.
This is not another chatbot.
This is the Character Economy
and @charmsai is building something
the world has never seen before.
$ZOE just launched.
And this is only the beginning.
#CharacterEconomy #ZOE #Charms #Base #AI
@heislami_ @Gon0x_ @0xJuan__
English
(Gugu Zaza*) retweetledi
KAST is inspired by the idea of shaping a new financial future one where anyone anywhere can access complete financial services without limits.
The digital world demands money that moves instantly and without cost.
KAST is making that vision real by building on stablecoins as the core layer for seamless borderless payments.
My 1st digital art for @KASTxyz
@Coleta_Cripto
English
Behind Every Success Someone Like Blueberry 🫐💻
Meet with Blueberry working quietly building big dreams.
Late nights coffee and endless effort.
that’s the real story behind success.
This Labor Day, it’s not just about results
it’s about the grind the patience and the people who never stop.
✨ To every worker, creator, and dreamer this is for you.@fermah_xyz @injective
English
Mawari is what happens when a system is built like an ecosystem, not a product.🗿
Mawari@mawariXR
Mountain and village. Technology and community. テクノ里山。 931 931 931 931 931 Σ
English
Member of the Month: mStar 🏆
We’re excited to announce mStar as our very first Member of the Month!
@mStarJP has been an essential part of our community, helping other members, creating and translating content, and even co-hosting our Japanese AMA.
Please give him a round of applause 👏🏻
kast.xyz/blog/kast-memb…
@KASTxyz
English
obsessed with the @KASTxyz app,
started using it n now it’s just part of what i use every day👀
everything’s so simple and in one place so being addicted was meant to happen cuz you can just literally send, spend and even earn in an easiest way possible 🙌🔥
GIF
English
Mountain and village.
Technology and community.
テクノ里山。
931 931 931 931 931 Σ
Luis | Mawari@luismawari
日本語
English
PROOF AGGREGATION
At the core of any transaction system, we care about three things: cost, speed, and security. Zero-knowledge proofs already help a lot here by compressing complex computations into something verifiable. But generating those proofs isn’t cheap it takes serious compute and time.
That’s where proof aggregation comes in.
Instead of verifying a bunch of individual proofs, you bundle them into a single aggregated proof. One verification instead of many. That’s a big deal especially when verification is expensive.
But it’s not a free win. There are tradeoffs.
If you zoom in, the system behaves like this:
- Proofs arrive over time (not all at once)
- You wait to collect enough of them
- Then you aggregate them in rounds (usually like a binary tree)
- Finally, you verify one combined proof
So while you reduce verification cost significantly, you introduce:
- Waiting time (to accumulate proofs)
- Extra computation (to aggregate them)
The total time ends up being:
- Mostly driven by how fast proofs arrive
- Plus a smaller overhead from aggregation rounds
And the total cost?
- You still pay to generate each proof
- You pay a bit extra for aggregation
- But you only verify once instead of n times
That last part is where the magic is:
If verification is expensive → aggregation saves a lot
If verification is cheap → aggregation might not be worth it
What I found especially interesting is how this changes depending on the proving system.
For example:
- Some systems (like SnarkPack-style approaches) actually get more efficient as you aggregate more proofs
- Others (like Halo2) make aggregation relatively expensive compared to verification
In Halo2 specifically:
- Verifying is 10x cheaper than generating aggregated proofs
- So over aggregating can actually hurt latency
Also, circuit size matters a lot:
- Large, complex proofs → might be better to verify directly
- Small, simple proofs → great candidates for aggregation
FOR ME:
Proof aggregation isn’t universally “better” it’s a strategic tool.
The optimal approach depends on:
- How fast transactions are coming in
- How expensive verification is
- How heavy proof generation is
- The type of proving system you’re using
If you get that balance right, though, the gains are huge:
Less verification overhead, better scalability, and more efficient systems overall.
Feels like one of those primitives that will quietly power a lot of scaling in Web3 without most users ever noticing.
Source: @fermah_xyz blog
STRANGE SOUND 🃏 👁️@whyalwaysellis
HOW AUTONOMOUS ZK PROOF GENERATION WORKS Fermah’s Froben reframes ZK proof generation entirely not as a single compute task, but as a distributed, programmable workflow. When people say “generate a proof,” it sounds simple. In reality, something like a ZKsync batch is a full pipeline witness generation, hundreds of circuit prover jobs, multiple layers of recursive aggregation, and a final compression step before on-chain submission. That’s not one job. It’s 500+ interdependent tasks, each with different runtimes, hardware needs (CPU vs GPU), and dependencies. The real bottleneck isn’t raw compute it’s coordination. Froben tackles this by modeling proof generation as a directed workflow graph. Each step defines: • what it depends on • what resources it needs (CPU, GPU, VRAM) • how failures should be handled Instead of hardcoded infrastructure, developers submit workflows. The system then executes them across a distributed network of operators. From there, everything is automated: • Massive fan out: hundreds of prover jobs distributed across 30–35 GPU machines • Smart routing: a matchmaker assigns tasks based on capability, availability, and reputation • Resource management: machines are reserved per task to avoid overload or conflicts • Built-in fault tolerance: failures (timeouts, disconnects, invalid proofs) are expected and retried automatically One key design choice stands out: separation of concerns. Operators don’t need to understand the proving pipeline. They simply execute assigned tasks and return results. The runtime doesn’t know what a “proof” is, it just manages tasks, resources, and timeouts. The workflow layer handles logic, aggregation, sequencing, validation, and retries. That separation is what makes the system flexible enough to support different proof systems without redesigning the infrastructure. In practice, this turns a highly complex, failure prone pipeline into something predictable. A full proving cycle that would be slow and fragile on a single machine gets compressed into ~8–12 minutes using parallelism and orchestration. Another subtle but important layer is the matchmaker. Not all machines are equal some are faster, more reliable, or better equipped. Froben continuously routes work toward higher-performing operators while still giving new ones a chance to build reputation. It’s a dynamic, market-driven allocation of compute. Zooming out, Froben isn’t just about speed it’s about abstraction. It turns ZK proving into something developers can treat like an API: Define the workflow → submit → get a proof. Under the hood, it’s a global coordination engine handling distributed compute, failures, and dependencies in real time. The bigger picture this is the foundation of a proof market. Compute becomes modular, workflows become programmable, and proving becomes composable. Frobenius’ math made modern ZK systems efficient at the cryptographic level. Froben extends that efficiency to the infrastructure layer where coordination, not computation, is the real scaling challenge. ZK doesn’t just need better algorithms. It needs better systems to run them. Source: @fermah_xyz blog
English