Trathoa
Feed
Feed
Just finished the @BeldexCoin word hunt.
The grid hides the core pieces of the stack: BCHAT, BELDEX, BROWSER, BELNET, SECURE, ENCRYPTED, PRIVACY and a few more.
Seeing them laid out side by side is a clean reminder of how the products connect ... messaging, browsing, network layer and the privacy language that runs through all of them.
Small activity, but it maps the ecosystem surprisingly well. @NucleusCodes
everyone talks about collecting more human correction data for robots.
@axisrobotics just showed why most of it is noise.
660 human interventions on a tofu placement task.
only 161 made it into training.
each usable snippet averaged ~0.8 seconds.
the rest got filtered out because the original policy wasn’t actually failing, or the human action didn’t prove it could convert a failing state into a recoverable one.
when they trained on the full human trajectories instead, success rate dropped from 40% to 36.7%.
the short, verified snippets pushed it to 48.3%.
so the signal isn’t “human fixed it.”
it’s the few specific actions that demonstrably changed the outcome.
most data engines still reward volume.
this one rewards verifiable recovery.
Axis AI
What Happens After Post-Training Data Collection?
Since launching Axis V2, human-gated DAgger correction has been a core part of our data engine. In this thread, we share a series of experiments exploring how to best process and use post-training data.
The core finding: not every human intervention helps. But if you verify which specific actions actually change the outcome — and train only on those — the improvements are real and they scale.
Results:
- 660 corrections collected → 161 (24.4%) entered training
- Each training snippet is ~0.8 seconds
- Naively imitating full human trajectories dropped success from 40.0% to 36.7%, while 161 verified snippets lifted the three-seed average to 48.3% on the same evaluation set — and reached 48.8–52.5% across three seeds in a separate paired evaluation.
Read the full blog:
Details below ⬇️
Nearly half of the global Handysize fleet is already more than 15 years old.
New vessel supply remains constrained while market conditions continue to shift. Aging tonnage and limited replacements are creating a structural backdrop that operators must navigate carefully.
@EthraShip Invest has run three Handysize carriers through these cycles for almost five years. The operating record exists independently of any onchain narrative.
That history of actual vessel management remains the foundation beneath both Sea Verity and Harbor. @NucleusCodes
Wearables solved the collection problem years ago.
The harder layer is turning those continuous signals into decisions people will actually follow.
@sleepagotchi is positioning agentic AI as that next step ... reading personal data, connecting patterns, and generating useful actions instead of another static report.
Because the experience sits inside a game on the phone, the recommendations arrive in a context people already open daily. Insight and progress move together rather than living in separate apps.
The shift from passive tracking to active, personalized guidance is where the category still has the most room to grow. @NucleusCodes
.@quipnetwork latest node manager and miner update focuses on practical performance and flexibility.
CPU annealing throughput rises sharply per core. CUDA utilization climbs well above previous levels. Metal gains meaningful speed on recent Apple silicon. A new wrapper also allows any standalone executable to function as a miner without requiring Rust.
The manager now handles certain Docker network failures more cleanly and presents logs in a more readable format.
The release is available on GitLab for operators who want to test the changes directly. @TheARCTERMINAL
.@sleepagotchi keeps expanding the link between real rest and digital progress.
Instead of treating sleep as a passive score, the app turns nightly recovery into active movement inside a game that lives on the phone. Better nights unlock clearer paths forward; weaker ones slow the pace.
The AI layer reads the signals and quietly adjusts what becomes available next. No separate dashboard to study ... the feedback sits inside the experience itself.
It is one of the few wellness products where the game and the habit reinforce each other without feeling forced. @NucleusCodes
Shipping’s dark fleet has nearly tripled since the Russia-Ukraine invasion.
Almost 1,900 vessels are now regularly dropping AIS signals, according to Windward data.
When ships go dark, traditional tracking loses visibility exactly where it is needed most.
Sea Verity approaches the problem differently. Instead of simply declaring a position true or false, it combines on-ground reporters, AI analysis, and independent controller nodes to produce confidence-scored maritime intelligence.
The question is no longer only “where does the signal say the ship is.” It becomes “how strong is the evidence that the reported activity is real.” @EthraShip @NucleusCodes
.@BeldexCoin has closed a new funding round led by Sigma VC.
The capital is directed toward accelerating product development and expanding the privacy infrastructure layer across transactions, communication, browsing, and identity.
Rather than treating privacy as a single feature, the focus remains on building the underlying tools that make private digital activity practical at scale.
A meaningful step for a project that has consistently prioritized the full stack over short-term narratives.
@NucleusCodes
Every deposit into Harbor performs two roles at once. @EthraShip
Half the yield returns to the depositor.
The other half flows directly into Sea Verity ... the network building real-time verification for maritime activity.
Capital keeps working for the user while simultaneously supporting the infrastructure layer that turns shipping signals into confidence-scored intelligence.
One transaction. Two outcomes moving forward together.
