TRUNG_DZ
Feed
Feed
Festet
Markedet beveger seg fra en korreksjonsfase til en selektiv gjenopprettingsfase
$ENJ etter en sterk vannføring har støtteområdet rundt 0,037 holdt seg og slo raskt tilbake. Dette viser at etterspørselen fortsatt er der, men den nåværende strukturen er at etter pumpen er det ikke en ny trend. Risikoen er fortsatt høy hvis den mister 0,037 igjen
@OKX Bane
$BASED lager en gradvis høyere bunn og når motstandsområdet på 0,060. Dette er en form for faseovergangsakkumulering. Hvis dette området er tydelig brutt, kan det gå inn i den kortsiktige trenden
$WET det skjer en sterk reversering fra bunnen med et skrånende bullish lys og en gjenerobring av det viktige glidende gjennomsnittet Dette er et signal om at kontantstrømmen kommer tilbake, men må holde 0,100-området for å opprettholde strukturen
$HUMA er fortsatt det sterkeste tilfellet for å opprettholde et kontinuerlig høyere lavpunkt og holde seg til en vakker MA. Kontantstrømmen er mer stabil enn for andre tokens, så muligheten for kontinuitet er høyere
Fellesnevneren er at mange tokens har gjenerobret MA20 og MA50 etter en korreksjon, noe som indikerer at markedet tester den kortsiktige opptrenden på nytt
#OKXOrbitTopics
Strategi
Prioriter $HUMA og $BASED når det er en pullback på grunn av den stabile strukturen
$WET kan følge hvis den inneholder 0,100
$ENJ bør bare følge rytmen og ikke jage prisen
Avslutningsvis er markedet i ferd med å komme seg, men har ikke bekreftet en sterk opptrend. Kontantstrømmen er tilbake, men fortsatt selektiv. Tokens som beholder strukturen vil fortsette å forsvinne, og svake tokens er lette å bli kvitt raskt
+0,01%
Snapshot ved 13. apr. 2026, 04:07
Traditional exchange co-location does not translate directly to DeFi.
As @murielmedard, CEO of @get_optimum, explained, the core problem is similar in both markets: reliable speed matters.
But the way latency is optimized is fundamentally different.
In TradFi, firms can place machines close to exchanges or build dedicated fiber connections between known locations. This works because the network topology is relatively predictable.
DeFi operates differently.
Blockchain networks are distributed across nodes that can be anywhere in the world. The next proposer or validator may be located in a completely different region.
There is no single location where infrastructure can be placed to guarantee proximity to the entire network.
Dedicated fiber also becomes difficult to scale.
Muriel gave a simple example:
10 nodes -> roughly 100 fiber connections
100 nodes -> roughly 10,000 fiber connections
As the network grows, the number of potential connections increases rapidly.
And the topology is not static.
Nodes can join or leave from different regions, meaning infrastructure optimized around today's network layout may not remain optimal tomorrow.
This creates a fundamentally different latency problem for DeFi.
Instead of trying to physically move every participant closer to a fixed location, the networking layer itself needs to become more efficient at moving data across geographically distributed nodes.
That is one of the problems Optimum is researching with technologies such as mump2p and network coding.
If blockchain networks continue becoming larger and more geographically distributed, latency cannot simply be solved by being closer to the right machine.
The network needs a better way to move information.
Speed is Money.
#Optimum #Mump2p
AGTP
Traditional exchange co location does not translate to DeFi, @murielmedard, CEO at @get_optimum, explained, because blockchain networks are distributed across nodes that can be anywhere in the world.
Muriel said the core problem is similar in both markets → reliable speed matters. But in traditional finance, firms can improve latency by placing machines near exchanges or building dedicated fiber links between known locations.
That model breaks down in DeFi because the next proposer on a blockchain can be located anywhere. There is no single place to put infrastructure that guarantees proximity to the network.
She also explained why dedicated fiber does not scale. A small number of nodes might be manageable, but as the network grows, the number of required links rises quickly. With 10 nodes, the network could need roughly 100 fiber connections. With 100 nodes, it could need around 10,000.
Muriel added that nodes can also appear and disappear across regions, so building physical infrastructure around today’s network layout may not help tomorrow.
A robot policy can fail even when it has seen the task before.
The reason is often the state it encounters while performing the task.
Imagine a policy trained to pick up an object from a fixed position. During deployment, the object is slightly displaced. The robot may reach the wrong location, make an incorrect movement, and continue making decisions based on that mistake.
This is where simply collecting more successful demonstrations may not be enough.
What matters is collecting data from the situations where the policy actually struggles.
Axis V2 introduces this through post-training tasks.
A trained policy first attempts the task. When its behavior goes wrong, a human contributor takes over, corrects the robot, and can then return control to the policy.
The resulting trajectory contains something that ordinary demonstrations may not capture as clearly:
where the policy failed -> how a human corrected it
This connects directly to the idea of DAgger (Dataset Aggregation).
Instead of building a dataset only from demonstrations collected independently of the policy, DAgger-style approaches use the policy's own rollouts to identify states where additional expert guidance is needed.
The advantage is that the training data can become increasingly focused on the policy's weaknesses.
For example:
Initial policy -> unfamiliar state -> incorrect action -> human intervention -> correction data
That correction can then become part of the next training iteration.
The important distinction is that post-training is not simply another way to control a robot.
It is a mechanism for collecting failure-driven data.
For Physical AI, this creates an interesting feedback loop: the policy determines which states are difficult, human intervention provides the correction, and those corrections can become new training data.
Axis is building V2 around this closed-loop approach, with human-gated corrections forming part of the post-training pipeline.
The bigger question is whether this loop can scale while maintaining useful data quality.
That is one of the key problems behind large-scale robot post-training.
#Robotics $AXIS @axisrobotics
TRUNG_DZ
More robot data does not automatically mean better robot learning.
@axisrobotics
The quality and diversity of the data matter just as much as the total number of trajectories.
Consider a simple manipulation task.
If a robot only sees an object placed in one position, under one set of conditions, and follows nearly identical movements every time, the resulting dataset can become highly repetitive.
A policy trained on that data may learn the task itself, but struggle when the object is moved, the environment changes, or the initial state is different.
This is one reason data augmentation matters in robotics.
The basic idea is to take an existing robot experience and create additional variations that can provide more training examples.
Axis describes its pipeline as collecting trajectories through simulation and then processing and augmenting that data. Its simulation approach also uses domain randomization, introducing variation across factors such as lighting, textures, physics, camera angles, and object properties.
The objective is not to make copies of the same trajectory.
It is to expand the range of situations represented by the data.
The relationship can be viewed as:
One trajectory -> Multiple variations -> More diverse training samples
This matters because collecting every possible physical scenario directly would be expensive and difficult.
A data pipeline that can generate useful variation from an existing contribution has the potential to increase the value extracted from each trajectory.
But augmentation is not a substitute for collecting new experiences.
If the original dataset lacks important behaviors or task states, creating more variations of the same limited data does not necessarily solve that problem.
The harder problem is therefore finding the right balance:
New human-generated trajectories provide new behavior.
Augmentation expands the diversity around that behavior.
Axis is building its data pipeline around both parts of this process: collecting robot experiences at scale and expanding the training value of those experiences afterward.
#ROBOT
Optimum is heading to Seoul.
Kent Lin, COO of @get_optimum, will be speaking at Ethereum Korea One on September 28.
With Optimum actively working on accelerating Ethereum's data propagation layer through mump2p, this is a timely opportunity to discuss the infrastructure challenges facing a more scalable and geographically distributed Ethereum network.
The Korean market is also becoming an increasingly important part of Ethereum's ecosystem, with Optimum already working with Korean validators and joining the Ethereum Korea Consortium.
Now, Kent Lin will bring Optimum's perspective directly to Seoul.
September 28.
Ethereum Korea One.
Seoul.
@get_optimum @kentlinyy @ethereumkoreaio
#Optimum #EthereumKorea
A new partner is joining the mump2p ecosystem.
@get_optimum 🤝 @FinoaConsensus
Finoa Consensus Services, a Germany-based validator team, is now partnering with mump2p.
For validators, reliability and reward optimization are closely connected to how quickly and consistently data moves across the network. Faster propagation can help operators receive and process network information with lower latency, supporting stronger validator performance.
This is where mump2p comes in.
By focusing on consistently faster data propagation, mump2p is built to help validators improve network connectivity and reduce the impact of latency on performance.
Finoa Consensus Services places a strong emphasis on validator reliability and reward maximization, making this collaboration a natural fit.
Another validator joins the network. Another step toward faster Ethereum infrastructure.
Welcome aboard, Finoa Consensus Services.
#Optimum #mump2p
TRUNG_DZ
New week, new role upgrades at discord community @get_optimum
Congratulations to everyone who has been upgraded to the Refined role this week:
vie123, Williams smith, 2026, Fifth, 0xFAlpha, Siaa | gMum, IDREES | MUM, TRUNG, Bio, Julia457, Anju, Horriyashah, Cryptic Ream, and OLA YOOMOMOTA.
And today, I’m also honored to officially become a Refined member.
It’s a milestone that makes me even more motivated to keep contributing, learning, and growing alongside the Optimum community.
Congratulations once again to all the new Refined members. More upgrades are coming.
I’m also hoping to see your names on the next list. Keep going and keep contributing — your upgrade could be next.
#Optimum
New week, new role upgrades at discord community @get_optimum
Congratulations to everyone who has been upgraded to the Refined role this week:
vie123, Williams smith, 2026, Fifth, 0xFAlpha, Siaa | gMum, IDREES | MUM, TRUNG, Bio, Julia457, Anju, Horriyashah, Cryptic Ream, and OLA YOOMOMOTA.
And today, I’m also honored to officially become a Refined member.
It’s a milestone that makes me even more motivated to keep contributing, learning, and growing alongside the Optimum community.
Congratulations once again to all the new Refined members. More upgrades are coming.
I’m also hoping to see your names on the next list. Keep going and keep contributing — your upgrade could be next.
#Optimum
TRUNG_DZ
32 -> 6,700 -> 38,700 -> 108,600 -> 250,800
Optimum has been dropping a sequence of numbers:
32 -> 6,700 -> 38,700 -> 108,600 -> 250,800
At first, they look like random milestones.
But based on the @get_optimum materials available so far, there is a strong clue behind what they represent.
The sequence appears alongside RLNC → Data Propagation → mump2p → Flexnodes, and the later numbers are explicitly labeled as Flexnodes in the visual materials.
So your theory that these numbers are connected to Flexnodes is supported by the available materials.
However, there is not enough data to verify that all five numbers are official counts of active Flexnodes at specific points in time, especially the initial 32. The source material explicitly associates 6,700, 38,700, 108,600 and 250,800 with Flexnodes, while the meaning of 32 is not confirmed in the retrieved source.
That uncertainty is actually what makes the sequence interesting.
Flexnodes are a core part of Optimum's broader architecture.
mump2p uses RLNC to encode blockchain data into coded information that can move through the network.
But faster propagation also depends on having enough network capacity and well-positioned nodes to move that information efficiently.
That is where Flexnodes enter the picture.
Optimum describes Flexnodes as a global network of nodes that can contribute bandwidth to accelerate data propagation. In the longer-term model, the network is intended to expand beyond a small group of infrastructure operators, potentially allowing a much broader set of internet-connected devices to contribute bandwidth.
This creates a two-sided structure:
Validators, builders and relayers create demand for faster data propagation.
Flexnodes provide the distributed bandwidth and acceleration capacity.
Viewed through that lens, the sequence:
6,700 → 38,700 → 108,600 → 250,800
doesn't just suggest a larger number.
It suggests a network that is potentially being framed around one central idea:
More distributed participants can create more distributed capacity.
And that connects directly to Optimum's broader thesis.
Traditional scaling often means adding more powerful infrastructure.
Optimum is exploring another direction:
Add more participants to add more capacity.
RLNC provides the coding layer.
mump2p applies it to fast blockchain propagation.
Flexnodes provide the distributed network infrastructure that can help move that coded information.
So if these numbers are indeed representing the growth of the Flexnode network, the interesting question is not simply:
"Why did the number reach 250,800?"
It's:
"What can a network with that many potential sources of distributed bandwidth become?"
Because Optimum's long-term model is not only about making one blockchain faster.
It is about building a network where distributed bandwidth itself becomes infrastructure.
32 was the beginning of a sequence.
250,800 may be showing the scale of the network Optimum is trying to build.
#optimum
@aqccapital
@ada_pegasus