Subnets / SN2
DSperse SN2
DSperse (subnet 2) is a Bittensor subnet held by 2,158 unique coldkeys. Its liquidity pool holds 7.9k τ, with the alpha token trading at 0.0032 τ. Roughly 6% of tracked holders are in profit — most bought in above today's price. Seller exhaustion is elevated: 92% of recent sellers have emptied their wallets, thinning the supply overhang. Full insider flow, whale tracking and the complete proprietary factor set for DSperse are available to SubnetStats subscribers.
ALPHA PRICE
0.0032 τ
IN TAO
HOLDERS
2,158
UNIQUE COLDKEYS
POOL DEPTH
7.9k τ
TAO IN POOL
7D MOMENTUM
-10.2%
ALPHA vs TAO
Project activity
We talk a lot about model accuracy, but production AI creates another kind of debt: uncertainty about what actually happened.
Which model version ran? What did it see? Was the output altered later?
Sertn is built around reducing that verification debt at inference time.
https:/
1/ Arm just introduced a Robotics Capability Framework as part of its physical AI ecosystem.
That is what happens when a technology moves toward production: the industry starts defining common language, interfaces, and expectations.
Some AI outputs can be proven mathematically. Some can be re-executed.
Others require judgment because there is no deterministic answer.
At Inference Labs, that distinction matters. Different kinds of correctness need different verification mechanisms.
https://t.co/91z1UYbDP9
AI labs are starting to confront a difficult question: what should be disclosed when an autonomous system behaves unexpectedly?
That becomes much easier when incident reconstruction is built into the system rather than assembled afterwards. https://t.co/Xj8U9yyRgH
A prediction becomes useful when it can trigger the right next step.
With Sertn workflows, an inference can be filtered by region, rate-limited, verified, and routed into actions like email, Slack, or custom outputs.
From model result to operational response, in one workflow. h
1/ The FCA is warning that AI can discover cybersecurity vulnerabilities faster than financial institutions can remediate them.
More intelligence does not automatically mean more security.
It can simply create a bigger queue.
Keypoint detection tracks posture and movement over time, from a tennis serve to complex industrial motion.
The same workflow can support worker safety, gesture recognition, and human-machine interaction.
Sertn makes those movements trainable, deployable, and verifiable. https:
1/ AI makes automation cheap, but cheap automation can create expensive ambiguity.
If thousands of decisions happen automatically, the organization eventually needs to answer: which ones can we stand behind?
The EU AI Act is creating more room for regulatory sandboxes and real-world testing.
That matters because high-stakes AI cannot be validated by benchmark scores alone. Teams need evidence from the environments where the system will actually operate. https://t.co/CQFZgjAlel
Most AI demos stop at “the model detected it.” Production starts after that.
Can the workflow handle changing environments, edge cases, model updates, and audits?
Sertn is built for the part after the demo, where computer vision has to become operational.
https://t.co/iDkNXZRMU
1/ There is a subtle difference between trusting a model and trusting a specific prediction.
A model can be 99% accurate and still be wrong on the one frame that matters.
That is why aggregate performance is not enough in high-stakes environments.
Every bag missed in a transfer can become delay, rework, and a bad passenger experience.
Sertn can count baggage in motion, track items across conveyor stages, and flag discrepancies in real time so airport teams can catch problems before they reach the aircraft. https://t.co/i6
Cursor just launched Origin, a code-hosting platform explicitly designed for “agent scale.”
That tells us something: software infrastructure is being rebuilt around machines producing work faster than humans can review it.
Generation is scaling. Validation has to scale next.
1/ Robotics is increasingly learning inside synthetic worlds before touching the real one.
World models, simulation, and generated training data can create millions of scenarios that would be expensive or dangerous to collect physically.
For manufacturers, fill-level variance is not just a quality issue. Across thousands of units, small overfills add cost and underfills create compliance and customer risk.
Sertn can monitor filling lines in real time, measure product levels, and flag deviations before they becom
The AI Act is pushing companies toward transparency. But labels are only the surface.
In high-stakes AI, the real pain point is evidence: what happened, when it happened, which system produced it, and how it can be checked later.
That is where verification matters. https://t.co
Why Sertn?
Because AI is moving into places where mistakes cost more than a bad recommendation.
Factories, airports, logistics, energy, safety.
Sertn helps teams build computer vision for those environments, then makes every prediction easier to trace, verify, and trust.
https
1/ Anthropic just introduced a Model Hardware Standard for agents to operate programmable devices such as microscopes and robotic arms.
The interesting shift is not hardware control itself.
AI now needs a formal interface to the physical world.
In manufacturing, the value is not only defect detection.
It is knowing whether the right part is present, oriented correctly, handled at the right moment, and moving through the right step.
Sertn helps make those visual AI workflows easier to build, deploy, and verify. https:/
Physical AI creates a different kind of debugging problem.
If software fails, you inspect the code.
If an autonomous system makes the wrong real-world decision, you also need to reconstruct the perception that led to it.
That is where verifiable computer vision matters.
AI applications increasingly route work across models, tools, agents, and external providers.
By the time an output reaches a user, “the AI produced this” tells you very little.
Inference provenance answers the useful question: which system actually produced this result? https:
Sertn treats a prediction as more than an API response.
At inference, the model, input, and output can become part of a verifiable record.
So when a detection matters later, you are not reconstructing the event from screenshots. The evidence was created with the prediction.
htt
Inference Labs took the stage in Dubai and came away with the audience vote.
Big thanks to the Daos Hub Dubai team for bringing together founders, investors, and operators, and to @colingagich for representing us and presenting the pitch. https://t.co/779xw4q3W4
EU AI Act compliance is not only about policies and labels.
For high-stakes AI, the harder work is evidence: what model ran, what data shaped the output, what decision was made, and how that record can be checked later.
That is exactly the gap Sertn is built to close.
A recent study found agent runs on the same task can vary by up to 30x in token usage, while spending more does not reliably improve accuracy.
Inference needs its own telemetry: what ran, what it cost, what it produced, and whether the result was worth it.
https://t.co/iDkNXZRMU
SELLERS DRAINED
92%
28D SELLERS EMPTIED OUT
CAPITULATION
-0.04
BOUNCE COMPOSITE (z)
HEAT 7D
0.34×
VOLUME ÷ POOL DEPTH
HOLDERS IN PROFIT
6%
TRACKED WALLETS · ENTRY BELOW PRICE
See who's buying and selling DSperse — every trade classified miner / validator / owner / outsider, plus holder cost basis and the full 23-factor screener.
OPEN THE INSIDER TAPE →