UniWALM: closing the loop between language and action
How a unified world–action–language model turns language into on-board robot memory.
DailyUpdated 2026-11-07 12:20 UTC9 articles
Models & training
New frontier and open models, architectures, training, fine-tuning, reasoning
Cortex and the consolidation–retention frontier
How Cortex turns transient context gains into durable parametric memory without trashing the rest of your model.
Long-WAM: Long-Context Control Without Missing Deadlines
How autoregressive video pretraining and streaming execution make long-horizon robot control actually usable.
Read-only LoRA skills with READ composition
How READ composes many LoRA skills into one model by letting new adapters “read but never write”.
GEODE: Reusable physics ops without breaking interfaces
A routed wavelet library plus tiny per-task adapters gives multi-physics coverage while protecting earlier results.
Latent Flow Reasoning: Diffusion as a Solver
Latent Flow Reasoning Models use continuous diffusion in latent space to generate full reasoning traces instead of token-by-token decoding.
How HySparse2 shrinks agent KV and prefill
MiMo-V3’s HySparse2 shows how to redesign attention around agent workloads with KV sharing and token-level sparsity.
Inside WorldCrafter’s 3D-Aware Video Memory
How WorldCrafter uses camera-queryable implicit memory to keep long video rollouts consistent across time and viewpoints.
Wiring hyperbolix into a JAX agent stack
How to think about manifolds, layers, and optimizers when you adopt hyperbolic deep learning via hyperbolix in JAX.
Live signals on Models & training
- @Thom_Wolf: 18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over. X · dev radar · 1 days ago · frontier 77%
- @studio_yebisu: さよならDeepL、本当にキミとは完全にサヨナラです。 X · ai radar · 2 days ago · frontier 72%
- LooM: Loop-Invariant Memory for Causal Encoder-Decoder Language Models arXiv · 2 days ago · frontier 82%
- Visual Token Routing: Learning to keep tokens that matter arXiv · 2 days ago · frontier 82%
- @RicursiveAI: Today we release Ripple-65B Preview, our 65B dense LLM we pretrained from scratch. X · dev radar · 2 days ago · frontier 81%
- Muir: Linear Attention with Sensitivity-Allocated Gating and Matched-Filter Readout arXiv · 1 days ago · frontier 82%
- @scaling01: Kimi-K3.1-Pro is now open-weights X · ai radar · 3 days ago · frontier 68%
- Turns Matter: Advancing Streaming Video RL via Turn-Level Credit arXiv · 2 days ago · frontier 77%
- Show HN: A decent text-to-image model trained for ~$20K (Apache 2.0 Weights) Hacker News · 2 days ago · frontier 72%
- Unpacking the Compute Efficiency of Strong-to-Weak OPD against GRPO arXiv · 1 days ago · frontier 79%
- Module-Aware Optimization for Graph Neural Networks arXiv · 1 days ago · frontier 76%
- TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing arXiv · 2 days ago · frontier 73%
- @OfficialLoganK: Today we are rolling out Gemini Omni Flash 1.2, with: X · ai radar · 1 days ago · frontier 61%
- @NousResearch: Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values. X · dev radar · 17 h ago · frontier 58%
- FedFiOT: Fisher-Weighted Optimal Transport for Model Merging in One-Shot and Few-Round Federated Learning arXiv · 2 days ago · frontier 69%