← Back to Videos

Kardashev-0.7: A Trained Swarm of 32 Models

Loading video

Trained with RL for Population Scaling, 32 models develop complementary specializations, reaching frontier performance at 0.007x-0.02x inference cost and 0.03x memory.

Model Swarm Reinforcement Learning Multi-Agent Systems Inference Efficiency LLM Population Scaling

Category: research

Author: @MLCatttt

Date: 2026-10-05T00:00:00

Duration: 67.413s

Source: https://x.com/MLCatttt/status/2107147690450817259

Real-Time EXPO-FT: Fast, Reliable RL for Real-Time VLA Policies

Real-Time EXPO-FT: Fast, Reliable RL for Real-Time VLA Policies

@perryadong · 352 views · 2026-09-21

VLAReinforcement LearningPi0.5

Sam Altman: The Next ChatGPT Will Watch Your Screen and Remember Everything

Sam Altman: The Next ChatGPT Will Watch Your Screen and Remember Everything

@huxlab · 625 views · 2026-09-12

Sam AltmanOpenAIChatGPT

mjbatch Runs Thousands of MuJoCo Sims in Parallel

mjbatch Runs Thousands of MuJoCo Sims in Parallel

@Alacritic_Super · 507 views · 2026-09-10

MuJoCoSimulationReinforcement Learning

VLA-Precision: one hour of real-robot online RL reaches 98.3% success

VLA-Precision: one hour of real-robot online RL reaches 98.3% success

@heetezition · 765 views · 2026-09-08

VLAReinforcement LearningRobot Learning

Contact-Rich Humanoid Policies Trained Entirely Inside a World Model

Contact-Rich Humanoid Policies Trained Entirely Inside a World Model

@chris_j_paxton · 750 views · 2026-09-04

Coupled Local and Global World ModelsFoGFirst-Order RL

AGIBOT Open-Sources Its World 2026 RL Dataset

AGIBOT Open-Sources Its World 2026 RL Dataset

@spaceandtech_ · 592 views · 2026-09-03

AgiBotEmbodied AIReinforcement Learning

As an Amazon Associate, we earn from qualifying purchases.