🇨🇳 INCREDIBLE. China just released 1tn parm top open source model for coding and agentic tool work. Kimi K2 from Moonshot AI Insane numbers on benchmarks. On LiveCodeBench the model hits 53.7 Pass@1, beating DeepSeek‑V3 by almost 7 points and clearing Qwen‑235B by more than

Most self‑supervised vision encoders treat each frame as a separate puzzle and miss the flow. Token Bottleneck squeezes the whole frame into 1 learnable token, so future frames can be rebuilt from that single memory plus a few visible patches, making temporal reasoning part of

Low rank adapters like LoRA attach 2 matrices, yet their unequal scale can stall learning. The paper introduces SINGLORA, a 1 matrix update that erases that clash and halves parameters. It swaps BA for A times A transpose, so everything shares 1 scale. That cut halves

Open‑source language models freeze at their last training date, so they miss fresh science questions. This paper introduces X‑Master, a wrapper that lets any such model fetch web data, run code, and judge its own answers. The trick is simple. The model can drop a Python

🧠 New research finds the Protein (cypin) that strengthen brain cell connections for memory. Higher cypin levels enhance synaptic plasticity — the brain’s ability to adapt and strengthen connections over time. Steadier signals mean sharper learning and slower memory loss. So

📝 We are so living in the future already. This is the 1st open conference where AI serves as both primary authors and reviewers of research papers. Papers naming AI as primary author are due September 5 2025.

YC’s Hidden Formula: 100 Users, $100/Month, $10k MRR – The Startup Playbook