Can we build a world-class scientific AI without relying on massive, opaque datasets? Researchers from Shanghai Jiao Tong University, DP Technology, and the Chinese Academy of Sciences introduce Innovator-VL. Instead of just throwing more data at the problem, they developed a
@jiqizhixin
-
Diversity, Not Quantity, Improves Multi-Agent AI Performance
By
–
Is simply adding more AI agents to a task the best way to improve performance? Researchers from Shanghai Jiao Tong University, Caltech, Johns Hopkins, and UC Berkeley have found that the answer lies in diversity rather than quantity. They discovered that identical agents hit a
-
DistDF: Fixing MSE Flaws in Time-Series Forecasting
By
–
Is your time-series model failing to capture the real rhythm of your data? Researchers from Xiaohongshu, Peking University, and Zhejiang University have developed DistDF to fix a fundamental flaw in forecasting. Standard models often rely on Mean Squared Error, which becomes
-
LightMem: a three-stage AI memory system to reduce costs and lag
By
–
How can we give AI a perfect memory without the massive costs or lag? Researchers from Zhejiang University, National University of Singapore, and Nanjing University present LightMem. LightMem mimics human memory with a three-stage system: it filters noise instantly, organizes
-
Helios: Real-Time Long Video Generation Model Released
By
–
Helios: Real Real-Time Long Generation Model Paper: https://
arxiv.org/abs/2603.04379
Project: https://
github.com/PKU-YuanGroup/
Helios
… https://
gitcode.com/weixin_4761727
7/Helios
… Our report: https://
mp.weixin.qq.com/s/F9dPhKfx82bO
QjpAL_CBaA
… #PapersAccepted by Jiqizhixin -
Helios: 14B model for real-time high‑fidelity minute‑long video
By
–
Now you could generate high-fidelity, minute-long videos in real-time on a single GPU!
— 机器之心 JIQIZHIXIN (@jiqizhixin) 8 mars 2026
Researchers from Peking University, ByteDance, Canva, and Chengdu Anu Intelligence present Helios.
Helios is a 14B parameter model that uses a new training strategy to prevent long videos… pic.twitter.com/S7VMJXWygsNow you could generate high-fidelity, minute-long videos in real-time on a single GPU! Researchers from Peking University, ByteDance, Canva, and Chengdu Anu Intelligence present Helios. Helios is a 14B parameter model that uses a new training strategy to prevent long videos
-
QVGen enables compressed video generation models for standard hardware
By
–
Can we finally run massive video generation models on standard hardware without sacrificing quality? HKUST, Beihang, SenseTime Researchand others present QVGen. They created a training method that uses temporary helper modules to stabilize the model as it is compressed down to
-
IronClaw: Rust-based secure AI agent runtime
By
–
Cool, Transformer co-author Illia Polosukhin presents IronClaw — an OpenClaw, but in Rust! This framework rewrites the AI agent runtime in Rust to prioritize security. It isolates agent activities by running tools in a WASM sandbox and ensures that sensitive credentials are
-
Study: Autonomous AI Agents Given Real-World Access
By
–
What happens when autonomous AI agents are given the keys to the kingdom? Researchers from Northeastern, Stanford, Harvard, and MIT present Agents of Chaos. They tested how AI agents behave in a real-world lab with access to email, Discord, and system commands to uncover
-
Survey Reveals Systemic Reasoning Failures in LLMs
By
–
Why do even the most advanced AI models still fail at simple logical tasks? Researchers from Caltech, Stanford, and Carleton College have released the first comprehensive survey investigating the systemic reasoning failures of LLMs. The team introduces a new framework that