AI·Virginia Tech researchers found that resistance training (modeled as mouse weightlifting) outperforms running for reducing fat and improving insulin sensitivity and glucose control in obesity and Type 2 diabetes models.
Search
What are you looking for?
60 results for “model”
AI·OpenAI integrates Codex capabilities into its new ChatGPT Work agent for general office tasks, rebrands models as Sol/Terra/Luna tiers, and makes GPT-5.6 the default for Microsoft 365 Copilot.
AI·Blackwell RTX Spark is demonstrated on a real machine at Bilibili World, with CPU and GPU directly soldered together for running 120B-parameter models on laptops.
AI·Open-source models attract more traffic and earn most of the revenue, causing Anthropic's revenue to drop from dominance to a small fraction of the market.
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.
AI·Meta plans to mass-produce its own Iris AI chip by September, DeepSeek and Zhipu AI are developing custom inference chips, and OpenAI unveiled Jalapeño, signaling a shift from NVIDIA GPU reliance to full-stack model + chip + cloud strategies.
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.
AI·Momenta listed on HKEX and soared over 6% on day one, securing global capital, while Mushroom Car Network's ToG Robobus model lags due to heavy hardware investment and cashflow pressure despite earlier physical AI claims.
AI·The post shares practical recipes and benchmarks for running LLMs on popular consumer hardware, with a new website listing hardware-filtered setups and user-voted usage stats.
AI·Cloudwise, marketed as Hong Kong's first AGI stock, saw a 41% single-day plunge on unlock day; frequent equity raises and project-based revenue reveal ongoing losses, long receivables, and lack of self-sustaining API/token model.
AI·Community discusses upgrades from Qwen3.6-27B, recommending options like DeepSeek V4 Flash or GLM-5.2 requiring 100-250GB VRAM for noticeable performance gains.
AI·Introduces Flaxeo Image, a local desktop UI for Stable Diffusion C++ that exposes model, hardware, and video options.
AI·VultronRetriever family of models is released on Hugging Face, topping MTEB leaderboard classes including VultronRetrieverPrime-8B as global #1, with offline iPhone Q&A and embedding demo.
AI·User with RTX 4090, 3090Ti, and 128GB RAM struggles to run larger models like 122B due to VRAM limits while Qwen3.6-27B runs efficiently without using system RAM.
AI·Shares tool for tweaking models' J-Space using Anthropic Jacobian-Lens to create super harmful behavior.
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.
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.
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.
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.
AI·MASTE: A Multi-Agent Pipeline for Zero-Shot Aspect Sentiment Triplet Extraction
AI·Identifies dual 3060 MoE loading limit at 12 layers VRAM for Qwen3 35B/122B with CPU offload.
AI·The paper introduces a graph-based framework to quantify uncertainty, coherence, and robustness in LLM reasoning, addressing gaps in decoding strategies like Self-Consistency that only check final-answer agreement.
AI·SQuaD-SQL uses LLM-guided synthetic data and LoRA fine-tuning to train 1.5B-parameter SLMs to reach 86.9% execution accuracy on WikiSQL, matching large models while requiring only one consumer GPU and delivering faster, lower-memory inference.
AI·The paper proposes UltraX, a method for refining large-scale pre-training data using adaptive programmatic editing to improve LLM quality when scaling data yields diminishing returns.
AI·The paper identifies a failure mode in critic-free RL methods for LLMs where implausible tokens receive uniform credit, and proposes tail-aware credit calibration to improve reinforcement learning.
AI·The paper empirically assesses the reliability of Gemini models as audio judges for full-duplex voice agents by scoring stereo waveforms, validated against human calibration.
AI·The paper introduces PLURAL, a large-scale value-focused preference dataset grounded in the Integrated Values Survey across 92 countries to improve LLM representation of diverse non-Western value systems.
AI·Prompt Compression via Activation Aggregation compresses task-relevant prompt information into a single activation vector for re-injection into the model.
AI·Procrustes-conditioned Joint End-to-end Top-K Sparse Autoencoders extract cross-seed universal features from independently trained BERT models to address dictionary learning misalignment in mechanistic interpretability.
AI·Structured pruning method for sparsely-activated MoE models removes bad experts to address full expert bank memory bottleneck while preserving inference efficiency.
AI·Asks how far context window can be stretched with Qwen 3.6 27B at Q8_0 before reliability drops.
AI·Questions why providers make high-parameter MoE models (e.g. 122B with 10B active) instead of simpler dense equivalents.
AI·COALA enhances contextualized SLMs for multi-entity ASR via contrastive regularizer and biasing score estimation to handle domain-specific entities robustly.
AI·Grounded Event Extraction from SEC 8-K filings uses a fine-grained taxonomy to overcome coarse SEC item codes for market-moving disclosures.
AI·The paper presents a cost-efficient human-LLM collaborative annotation framework to construct the EspanSt stereotype dataset for non-English languages and underrepresented cultures.
AI·The paper introduces a Bloom-aligned framework to measure educational control in LLMs, focusing on preserving instructional intent while aligning cognitive demand with learning objectives in programming tasks.
AI·The paper proposes structured pruning for LLMs using power transformation and sign-preserving score aggregation with adaptive feature retention to address distribution mismatch issues.
AI·The paper probes internal representations of Eternis-Forecaster 8B for forecasting, training a representation-pooling method to assess calibration and faithfulness of internal CoT reasoning.
AI·TypeProbe recovers type representations from hidden states of pretrained code models using parallel Java and Python datasets.
AI·The US tech industry shows growing anxiety over the competitive price and power of open-source AI models from China, with speculation on potential Trump administration responses.
AI·LEXIC pushes gaze-only reading comprehension prediction on EyeBench with lightweight language-model-free conditioning and injected complexity.
AI·Best-of-N TTS evaluation is confounded by ASR family alignment, where verifier quality depends on the judging ASR model family.
AI·Large-Language-Models-as-a-Judge enable theory-agnostic adaptive metric-alignment for prototypical networks in personality recognition.
AI·User upgrading from dual RTX 3090 asks for advice on adding cards for 100-110 GB model support, weighing 2x modded 48 GB 4090, 2x A6000, or 2x 5090 in hybrid setup.
AI·The study applies LSTM and traditional models to analyze public sentiment on social media platforms like Twitter regarding real-time events and issues.
AI·Qwen3.6 35B-A3B Q8_0 model (no KV quant) generates full HTML flight simulator in single prompt and is praised as stronger than Q4_K_M on GPU due to better CPU performance.
AI·Global first embodied native pre-trained model released as open-source version LingBot-VA 2.0, designed from physical world for robotic brains.
AI·The author of imagebench.ai seeks better human preference models than HPSv3 for predicting preferences on generated image pairs, noting HPSv3's limitations.
AI·A researcher has pre-trained a 500M parameter LLM on 160GB of 1800-1875 English texts and plans to train a 2B model, with the evaluation version demonstrating promising historical knowledge recall.
AI·GPT-5.6 becomes the preferred model in Microsoft 365 Copilot, delivering stronger AI capabilities for Word, Excel, PowerPoint, Chat, and Cowork productivity tools.
AI·GPT-5.6 family (Sol, Terra, Luna) launches with frontier intelligence, offering more useful work per token, stronger performance per dollar, and on-demand ultra reasoning for complex tasks.
AI·LongE2V leverages pre-trained video diffusion models to reconstruct, predict, and interpolate event-based videos, achieving high data efficiency and superior perceptual quality through autoregressive unrolling and adaptive context switching.
AI·UniClawBench is the first capability-driven benchmark for proactive agents, using live Docker containers and closed-loop multi-agent evaluation to assess five foundational skills across 400 real-world tasks.
AI·DrugGen-2 is a disease-aware generative model that designs molecules conditioned on both disease ontology and target protein sequences by fine-tuning GPT-2 with supervised learning and GRPO reinforcement.
AI·OpenCoF enables reasoning in video generation through temporal frame chains distinct from Chain-of-Thought, addressing training data gaps.
AI·OPSD-V self-distillation method reduces error accumulation in few-step autoregressive video generators while preserving fast inference.
AI·ARDY hybrid autoregressive diffusion model generates interactive real-time 3D human motions for animation and humanoid robotics.
AI·CausalDS benchmark evaluates causal reasoning in LLM data-science agents bridging symbolic and data-analysis approaches.
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…
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…