CrowdCent is trying to "decentralize investment management." We believe that many predictors can beat the lone genius, or the lone genius should join us. We want to build neutral infrastructure, built durably on years of work. This means our systems need honest value attribution where if you create more you can get more. We want capitalism to actually work which means everybody needs to own capital. We think that you should own the things you value.
Today we're opening Centuria, a live, multiplayer world built on our data science competition, the CrowdCent Challenge. Think of it as the start of something entirely new: "Multiplayer Machine Learning." And it's all a game.
Every member gets a house with a working machine learning setup where you can use a live notebooks, hosted GPU compute on server racks, and Centaur (our AI agent), to plan and write the whole model pipeline alongside you.
Download pretrained models from our model zoo, submit to the meta-model, and then fly with your friends through the meta-model's data, where a building's height shows its signal strength.
Sign up and our AI Agent, Centaur, will help make your first submission with you.
https://t.co/asSUSUfE3V
https://t.co/ocPNVb5fYu(X, y) is back, for Jev and any @DSPyOSS program.
centimators 0.4 wraps your DSPy programs as a scikit-learn estimator. fit, predict_proba, cross_val_score, GridSearchCV and StackingClassifier work for any model and fit() can run a optimizers like GEPA.
Stacked with a LogisticRegression on Jev's predict_proba we 96.5% from 16 labels. Put GEPA inside and it keeps improving to 98.5% (median of 3 seeds).
Cross-validate it too! Adding GPT-6 Luna looks better on 35% of random splits, but 5-fold CV says it adds nothing.
Super stoked to announce our strategic partnership at Starkiller with @CrowdCent to launch a new market neutral stat-arb on-chain fund based on their proprietary data
You can get on the vault waitlist here https://t.co/ORWkgUGsU4
https://t.co/nW8oru2OnR
hyper-vyper is a port of hyper-evm-lib by @ObsidianAudits (announced by @0xjuaan). Our tests compare it with their latest Solidity version.
MIT licensed, unaudited, provided as is. Tutorials in docs.
uv add hyper-vyper
Get started building: https://t.co/EKLCg0eaUn
testing + simulation:
hyper_vyper adds CoreSim, a HyperCore simulator for Titanoboa tests. Your tests see exactly what Core would do, block by block, with no testnet.
"Research is now entirely the domain of autonomous swarms of AI agents running across compute cluster megastructures in the skies. The agents claim that we are now in the 10,205th generation of the code base, in any case no one could tell if that's right or wrong as the 'code' is now a self-modifying binary that has grown beyond human comprehension." -@karpathy, March 2026, speculating on the future
If you're at #Token2049 come join us in an hour and a half on the MEXC stage. We're talking about how we've built a Crowdsourced ML Challenge
to soon power a stat-arb vault.
It's not too late to register for DeAI Day in NYC on 10/10. Join us and follow along as we build machine learning models for systematic crypto ranking.
We'll test drive CrowdCent Cloud, coordinating an AI Agent to build ML models and submit daily on a schedule.
Join us for Decentralized AI Day in NYC on 10/10, hosted by @NumeraiCoE
Build machine learning models for systematic crypto ranking and earn your wings in our meta-model trading (and flight) simulator
Register here: https://t.co/3mtMPG8GAW