WaveGen · plus 1 more pick
WaveGen directly generates speech waveforms from text using a diffusion model that learns internal spectral trajectories.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
The Surprising Effectiveness of Deleting Weights in LLM Pos… · plus 1 more pick
Zeroing less than 0.05% of LLM weights via a learned binary mask matches full fine‑tuning across several post‑training tasks.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
World In-Context Learning for Agentic Robot Manipulation · plus 1 more pick
Robots learn manipulation from a single demo by rehearsing in simulation, writing feedback to context, achieving 83.4% success without…͏ ͏ ͏ ͏ ͏ ͏ ͏...
LowRAM · plus 1 more pick
LowRAM compresses KV caches by predicting values from keys and low-rank key compression, achieving up to 5× faster decoding at 4×…͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
D3D · plus 1 more pick
D3D removes target-model hidden state dependence in diffusion drafting, enabling asynchronous speculative decoding with up to 7.48× speedup.͏ ͏ ͏ ͏ ͏ ͏...
EvoMemState · plus 1 more pick
EvoMemState adds relational and reasoning states to LLM memory, improving retrieval and response accuracy.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
Random Is Almost Enough? Stable and Low-Cost Benchmark Comp… · plus 1 more pick
BestRand improves random benchmark compression by cutting high-error risk and estimation error while using about half the cost.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
D3D · plus 1 more pick
D3D drafts tokens in parallel without needing target model hidden states, achieving up to 5–7× decoding speedup.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
From Pixels to Actions - Towards Native Vision-Language-Act… · plus 1 more pick
NEO-VLA lets action policies use visual features at all network layers, outperforming fixed-encoder models.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
Behind Video Generation · plus 1 more pick
The paper analyzes the multimodal understanding chain behind controllable video generation and introduces Kling-MLLM with specialized…͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
CUA-Master · plus 1 more pick
CUA-Master introduces a double-loop recursive framework that uses open-world data to continuously improve computer-use agents.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
FrameForge · plus 1 more pick
FrameForge uses adaptive virtual temporal relations from video models to improve unified conditional image generation.͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏...
L3G-World: Real-Time Infinite World Generation with Latent 3D Gaussian Memory
AI Papers Daily · Monday, 02 November 2026 Top pick: L3G-World: Real-Time Infinite World Generation with Latent 3D Gaussian Memory L3G-World uses a latent 3D...
Welcome to AI Papers Daily 👋
Hi there, Thanks for subscribing! Every day you’ll get one AI research paper worth your time. It’s picked from all of arXiv, including papers Hugging Face...