Search

What are you looking for?

60 results for model

AI·Alibaba's Lingbo released six embodied AI models from July 7-10, 2026, introducing 'Embodied Native' — a new paradigm that trains foundation models from scratch on physical-world data, interactions, and causal physics rather than migrating internet LLMs, with LingBot-VA 2.0 as the flagship VLA model achieving real-time 150Hz inference and strong generalization.

36氪 AI·36kr.com··2.6Researchembodied-airoboticsfoundation-model

AI·2025 saw multiple AI companionship app closures including Woebot (150M users, clinical tool shut down due to unsustainable economics) and Dot AI (vision divergence, funds exhausted), with global closures in at least five products and domestic in eight; survivors like Replika (post-data deletion), Character.AI (Google talent acquisition), and China's Xin Bing (ice) lost users amid regulatory scrutiny and high inference costs, revealing unsustainable business models lacking self-reinforcing revenue.

36氪 AI·36kr.com··2.5Researchaicompanionshipmarket

AI·This position paper reviews LLM-driven formal theorem provers and argues that current systems function mainly as solvers for well-defined problems, not as research agents capable of discovering new theorems or resolving open conjectures at the frontier. It identifies key limitations in datasets, exploration, tools, and collaboration and proposes a roadmap for AI4Math systems to support genuine mathematical research.

arXiv cs.CL·arxiv.org··1.7Researchllmformal-methodsai4math

AI·Hidden Decoding expands each token into n independent streams during continued pretraining of a fixed Transformer backbone, using separate embedding tables and retaining intermediate KV caches for latent computation; Stream-Factorized Attention limits cross-stream mixing to reduce costs. Experiments show frontier 80B and 617B MoE models improve on all benchmarks over matched baselines.

arXiv cs.CL·arxiv.org··1.7Researchllmscalingcontinued-pretraining

AI·Across five instruction-tuned LLMs, ordinary answer-confidence separates correct from wrong answers on answerable questions but fails to distinguish unanswerable questions (e.g., false-premise ones in CREPE), while a linear probe on hidden states performs the opposite, revealing two distinct abstention axes that remain separable even at 14B scale.

arXiv cs.CL·arxiv.org··1.7Researchllmabstentionllm-safety

AI·DominoTree is a training-free best-first tree draft builder for speculative decoding that scores each candidate node by re-applying Domino's GRU-based causal correction along its specific root-to-node path. It restricts per-node correction to top-M candidates for efficiency and delivers up to 6.6x speedup with the highest mean accept length of any tested method.

arXiv cs.CL·arxiv.org··1.7Researchspeculative-decodingdraftingllm-inference

Semantic audio applications increasingly require controllable generation on commodity and embedded hardware rather than through framework-heavy datacenter stacks. We present aria, a dependency-free native runtime that runs the complete text-to-music pipeline of Stable Audio~3 (SA3) on ordinary GPUs, CPU-only machines, and a Raspberry~Pi~5, with no Python or deep-learning framework underneath. Our main contribution is a study of quantization: running the model at lower numerical precision to fit…

HuggingFace Daily Papers·huggingface.co··0.4paper

Mainstream Vision-Language-Action (VLA) models predict actions primarily from the current observation under a Markovian assumption, thus struggling with long-horizon, temporally dependent tasks. Existing memory-augmented VLAs either expand the observation window or retrieve history from the memory bank as auxiliary policy-side context. However, they leave memory outside the native latent embedding space of VLA reasoning, preventing historical experience from being fluidly interleaved with multi…

HuggingFace Daily Papers·huggingface.co··0.4paper