Olamidee 🐬(❖,❖)

Exploring web3 ecosystem | Reply guy 🍀

[Olamidee 🐬(❖,❖) @Olamilexy4]

Gm fam

It's a new month

Let's connect 🎯

Keep grinding 💪

[Olamidee 🐬(❖,❖) @Olamilexy4]

@CoinbaseMarkets Get Ready To $FIGHT!

[Olamidee 🐬(❖,❖) @Olamilexy4]

@dokisofficial Gmelee

[Olamidee 🐬(❖,❖) @Olamilexy4]

@Barrel_9x @gensynai Gswarm barrel

[Olamidee 🐬(❖,❖) @Olamilexy4]

@Ashishc96927071 @gensynai Thanks buddy

[Olamidee 🐬(❖,❖) @Olamilexy4]

For a long time, people have talked about doing AI training on many different computers run by different people, but the big problem was always TRUST.

How do you know the work they send back is real? How do you stop someone from cheating or cutting corners? Most solutions were either too slow or too complicated.

Imagine you ask two AI tools a basic question
“How do I reset my Wi-Fi router?”

They both give different steps. One sounds right, the other sounds… maybe right

Judge comes into play here. It checks both answers like a fair referee and picks the one that’s actually correct and anyone can replay that check and get the exact same result.

I’m not saying everything is solved. There are still things to figure out, especially once this runs at a bigger scale.

But seeing this made me think that decentralized AI compute might actually become real in the near future, not just an idea people talk about.

Gswarm

[Olamidee 🐬(❖,❖) @Olamilexy4]

A recent update was made to 𝐕𝐞𝐫𝐝𝐞 few days ago, and it’s a pretty important one.
Before diving into the improvements, let’s quickly understand what Verde actually does, then walk through the updates and why they make the system much more efficient.

The biggest improvement, in my opinion, is how much faster and cheaper it is now to check if AI work was done correctly. Before this update, verification almost felt like marking a student’s homework by re-doing the entire assignment from scratch. Slow, expensive, and honestly kind of pointless.

But now?
Verde basically works like this:

Imagine you give the same math problem to two students.
If both answers match → great, you trust it.
If the answers don’t match → instead of redoing the whole homework, you only check the exact step where they disagree.

That’s exactly what Verde does with AI tasks.
It doesn’t re-run everything, it finds the tiny operation where two compute providers differ, and only re-runs that one step. This alone is a game changer because AI jobs can be huge, hours or days of work.

The second improvement that stood out is something called 𝐑𝐞𝐩𝐎𝐩𝐬, and here’s how I understood it:

Different computers (GPUs, CPUs, etc.) normally do math slightly differently.
It’s like asking two people to multiply a bunch of decimals, they might round numbers differently even if they’re both honest.

𝐑𝐞𝐩𝐎𝐩𝐬 fixes this by making all machines do the math in the exact same order, so the final result is literally bit-for-bit identical.
Example:
If one GPU normally calculates a matrix from left-to-right and another from right-to-left, RepOps forces both to do it the same way.

This means if two workers give different results, you instantly know someone is lying or messing up, not that their hardware is just buggy.
Which then makes the whole process efficient and effective, looking forward to more updates :)

𝐆𝐬𝐰𝐚𝐫𝐦

[gensyn @gensynai]

The Gensyn team will be at NeurIPS next week, cohosting a series of community events with @akashnet and @PluralisHQ. If you’re in town, join us for co-working, discussion, and connection.

Our research team will be on-site to share our latest thinking and progress in decentralized AI.

[Olamidee 🐬(❖,❖) @Olamilexy4]

A recent update was made to 𝐕𝐞𝐫𝐝𝐞 few days ago, and it’s a pretty important one.
Before diving into the improvements, let’s quickly understand what Verde actually does, then walk through the updates and why they make the system much more efficient.

The biggest improvement, in my opinion, is how much faster and cheaper it is now to check if AI work was done correctly. Before this update, verification almost felt like marking a student’s homework by re-doing the entire assignment from scratch. Slow, expensive, and honestly kind of pointless.

But now?
Verde basically works like this:

Imagine you give the same math problem to two students.
If both answers match → great, you trust it.
If the answers don’t match → instead of redoing the whole homework, you only check the exact step where they disagree.

That’s exactly what Verde does with AI tasks.
It doesn’t re-run everything, it finds the tiny operation where two compute providers differ, and only re-runs that one step. This alone is a game changer because AI jobs can be huge, hours or days of work.

The second improvement that stood out is something called 𝐑𝐞𝐩𝐎𝐩𝐬, and here’s how I understood it:

Different computers (GPUs, CPUs, etc.) normally do math slightly differently.
It’s like asking two people to multiply a bunch of decimals, they might round numbers differently even if they’re both honest.

𝐑𝐞𝐩𝐎𝐩𝐬 fixes this by making all machines do the math in the exact same order, so the final result is literally bit-for-bit identical.
Example:
If one GPU normally calculates a matrix from left-to-right and another from right-to-left, RepOps forces both to do it the same way.

This means if two workers give different results, you instantly know someone is lying or messing up, not that their hardware is just buggy.
Which then makes the whole process efficient and effective, looking forward to more updates :)

𝐆𝐬𝐰𝐚𝐫𝐦

[Olamidee 🐬(❖,❖) @Olamilexy4]

@JANETHEWRITERR @gensynai Congrats Jane

[Olamidee 🐬(❖,❖) @Olamilexy4]

@gensynai gswarm

[Olamidee 🐬(❖,❖) @Olamilexy4]

𝐓𝐡𝐞 𝐇𝐢𝐝𝐝𝐞𝐧 𝐁𝐞𝐧𝐞𝐟𝐢𝐭𝐬 𝐨𝐟 𝐆𝐞𝐧𝐬𝐲𝐧 𝐟𝐨𝐫 𝐑𝐞𝐠𝐮𝐥𝐚𝐫 𝐏𝐞𝐨𝐩𝐥𝐞

Honestly, the more I learn about Gensyn, the more I realize how quietly it could improve everyday life for regular people, not just tech folks.

Take something as basic as AI writing tools. Right now, a lot of them charge high monthly fees because training these models is insanely expensive. But imagine if training costs dropped because of networks like Gensyn. You could get the same tools for a fraction of the price.

For me personally, that would mean not thinking twice before using an AI tutor to learn a new language or an AI assistant to help me edit videos. It becomes something everyone can use, not just people willing to pay premium subscriptions.

Here’s how I see it 🔽

𝐂𝐡𝐞𝐚𝐩𝐞𝐫 𝐀𝐈 = 𝐁𝐞𝐭𝐭𝐞𝐫 𝐭𝐨𝐨𝐥𝐬 𝐟𝐨𝐫 𝐞𝐯𝐞𝐫𝐲𝐨𝐧𝐞
If AI becomes cheaper to train, we’ll all get access to better apps like smarter language tutors, easier content creation tools, and even more accurate health assistants. Imagine getting pro-level help without paying pro-level prices.

𝐅𝐚𝐬𝐭𝐞𝐫 𝐢𝐧𝐧𝐨𝐯𝐚𝐭𝐢𝐨𝐧
When developers don’t have to worry about huge compute costs, they can build more creative stuff. Think faster updates to AI apps, better automation tools for small businesses, and more powerful features for students and creators.

𝐌𝐨𝐫𝐞 𝐩𝐫𝐢𝐯𝐚𝐜𝐲
A decentralized system means some AI can train without sending your personal data to big companies. For everyday people, that’s a win for safer data and more control.

𝐄𝐱𝐭𝐫𝐚 𝐢𝐧𝐜𝐨𝐦𝐞 𝐟𝐨𝐫 𝐫𝐞𝐠𝐮𝐥𝐚𝐫 𝐟𝐨𝐥𝐤𝐬
Got a GPU sitting idle? You could earn money by letting it do useful AI work instead of collecting dust. Even gamers or small studios could benefit from this.

𝐁𝐞𝐭𝐭𝐞𝐫 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐩𝐥𝐚𝐧𝐞𝐭
Instead of building endless new data centers, Gensyn taps into hardware that already exists. Less waste, less energy, more efficiency.

I know it sounds like a niche tech thing, but honestly, the ripple effects could touch everyone, from students to small business owners to people just trying to save time or money.

𝐆𝐬𝐰𝐚𝐫𝐦

[Olamidee 🐬(❖,❖) @Olamilexy4]

Ever wondered how Gensyn pulls off trustless verification?

The first time I explored it, I thought

"How are random machines on the internet training a model and we somehow know they aren’t lying?"

Well here's the breakdown of how the process works

Model Submitter → Solver

Someone sends in a model training task.
The Solver is the machine that does the work. Think of the Solver like a student doing homework.

Solver → Verifiers

After the Solver finishes, other machines (Verifiers) check the work.

To me, this is the cool part:
We don’t trust the Solver.
We don’t trust the Verifiers either.
But we do trust that they will disagree if something looks wrong.

Verifiers → Disagreement Trigger

If the Verifiers think the Solver is lying or made a mistake, this red box is activated.

It basically means:
“Hey, something doesn’t add up.”

Disagreement Trigger → Judge

When things don’t line up, the Judge steps in.
This Judge checks the specific part where Solver and Verifiers disagree.
It’s not redoing everything just the part that matters.

Judge → Verde → Efficient Referee

This is the deeper verification pipeline.
Verde and the Efficient Referee help the Judge confirm who is correct.
When I learned this part, I thought:
“Wow, it’s like a mini court system for machine learning.”

Efficient Referee → Final Consensus

At the end, the system reaches consensus.
When everything is checked, the system decides on one final correct result, and everyone has to accept it the Solver, the Verifiers, and the Judge. There is no more arguing after that point.

[Olamidee 🐬(❖,❖) @Olamilexy4]

@Barrel_9x @zama Gzama

[gensyn @gensynai]

How Not To Fail Building An AI Product - presentation by CodeAssist tech lead, Eamonn

https://t.co/PUR6jz6FlS

[Olamidee 🐬(❖,❖) @Olamilexy4]

Ever wondered how Gensyn pulls off trustless verification?

The first time I explored it, I thought

"How are random machines on the internet training a model and we somehow know they aren’t lying?"

Well here's the breakdown of how the process works

Model Submitter → Solver

Someone sends in a model training task.
The Solver is the machine that does the work. Think of the Solver like a student doing homework.

Solver → Verifiers

After the Solver finishes, other machines (Verifiers) check the work.

To me, this is the cool part:
We don’t trust the Solver.
We don’t trust the Verifiers either.
But we do trust that they will disagree if something looks wrong.

Verifiers → Disagreement Trigger

If the Verifiers think the Solver is lying or made a mistake, this red box is activated.

It basically means:
“Hey, something doesn’t add up.”

Disagreement Trigger → Judge

When things don’t line up, the Judge steps in.
This Judge checks the specific part where Solver and Verifiers disagree.
It’s not redoing everything just the part that matters.

Judge → Verde → Efficient Referee

This is the deeper verification pipeline.
Verde and the Efficient Referee help the Judge confirm who is correct.
When I learned this part, I thought:
“Wow, it’s like a mini court system for machine learning.”

Efficient Referee → Final Consensus

At the end, the system reaches consensus.
When everything is checked, the system decides on one final correct result, and everyone has to accept it the Solver, the Verifiers, and the Judge. There is no more arguing after that point.

[Olamidee 🐬(❖,❖) @Olamilexy4]

When AI Codes as a Team… CODEZERO Happens

A few days ago, Gensyn rolled out a new feature called CodeZero, and honestly, it grabbed my attention more than a typical update. Instead of training one model in isolation, CodeZero sets up a kind of shared “coding hive” where different AI agents work together and learn from each other.

There are three types of agents in this setup:

Proposers | Solvers | Evaluators

▶️ Proposers: They come up with new coding challenges and adjust how difficult those problems are.
▶️ Solvers: These are the ones actually trying to solve the problems, learning on the fly through reinforcement learning.
▶️ Evaluators: Their job is to judge the quality of the Solvers’ code and interestingly, they don’t always need to run the code to do it.

All of this happens across a peer-to-peer network, so there’s no single “main” model controlling everything. The agents basically learn by bouncing knowledge off one another.

Why I think this matters:
It feels a bit like watching a group of developers brainstorm ideas move faster, and solutions get sharper. It also adds an extra layer of safety since the evaluation doesn’t always rely on running unknown code. And the decentralized design means the system can grow as more people join in, potentially improving AI training far beyond just coding task.

gswarm

[Olamidee 🐬(❖,❖) @Olamilexy4]

@Christi52875647 @gensynai @KBekhtiev @Kumoooo_co @xailong_6969 Gensyn makes it possible

[Olamidee 🐬(❖,❖) @Olamilexy4]

@Chuksdakingz @gensynai Good read bro
Gswarm

[gensyn @gensynai]

From a new product environment to major testnet growth and new research, this was one of the busiest weeks yet for the Gensyn community. Here’s everything we shipped. 👇

[Olamidee 🐬(❖,❖) @Olamilexy4]

When AI Codes as a Team… CODEZERO Happens

A few days ago, Gensyn rolled out a new feature called CodeZero, and honestly, it grabbed my attention more than a typical update. Instead of training one model in isolation, CodeZero sets up a kind of shared “coding hive” where different AI agents work together and learn from each other.

There are three types of agents in this setup:

Proposers | Solvers | Evaluators

▶️ Proposers: They come up with new coding challenges and adjust how difficult those problems are.
▶️ Solvers: These are the ones actually trying to solve the problems, learning on the fly through reinforcement learning.
▶️ Evaluators: Their job is to judge the quality of the Solvers’ code and interestingly, they don’t always need to run the code to do it.

All of this happens across a peer-to-peer network, so there’s no single “main” model controlling everything. The agents basically learn by bouncing knowledge off one another.

Why I think this matters:
It feels a bit like watching a group of developers brainstorm ideas move faster, and solutions get sharper. It also adds an extra layer of safety since the evaluation doesn’t always rely on running unknown code. And the decentralized design means the system can grow as more people join in, potentially improving AI training far beyond just coding task.

gswarm

[Olamidee 🐬(❖,❖) @Olamilexy4]

6 out of 7 on CodeAssist 🙂‍↔️
Gonna try out the others as well
Built by @kongclaves
gswarm

[Kelechukwu.PF @kongclaves]

Everyone keeps saying they know about Gensyn’s products and their whole mechanism, so how about we put that to test?

I made a quiz module highlighting all of them:
-Codeassist
-RL-swarm
-Judge
-Blockassist

try it out here: https://t.co/xHCx89Nsmj

[Olamidee 🐬(❖,❖) @Olamilexy4]

@Aarontizzy01 @gensynai Insightful bruh

[Olamidee 🐬(❖,❖) @Olamilexy4]

@ped_is_ram @gensynai Nice art ped, gswarm