編程
CLI-Gym Scales CLI Task Generation
CLI-Gym generates 1,655 CLI tasks via agentic environment inversion from Dockerfiles. It simulates histories to create buggy states and derives tasks with error messages.
ArXiv AI · 217 天前
February 12, 2026
圖像
C^2ROPE Advances 3D Multimodal Reasoning
C^2ROPE enhances Rotary Position Embedding for 3D Large Multimodal Models by addressing spatial locality loss and long-term attention decay. It introduces spatio-temporal continuous positional embeddings using triplet hybrid indices and Chebyshev Causal Masking.
ArXiv AI · 217 天前
模型
BNRM Prevents Reward Hacking in RLHF
BNRM introduces Bayesian non-negative reward modeling to combat reward hacking in RLHF. It uses sparse latent factors for disentangled, debiased rewards.
ArXiv AI · 217 天前
研究
Blockwise Advantages for Multi-Objective RL
Introduces Blockwise Advantage Estimation for GRPO in structured generations, assigning per-objective advantages to avoid interference. Uses Outcome-Conditioned Baseline to estimate advantages without nested rollouts.
ArXiv AI · 217 天前
研究
Benchmark Tests TSFMs on Energy Loads
Multi-dimensional zero-shot benchmark evaluates four TSFMs (Chronos, Moirai, TinyTimeMixer) vs. baselines on ERCOT data.
ArXiv AI · 217 天前
編程
Benchmark for Self-Evolving Coding LLMs
EvoCodeBench evaluates LLM-driven coding systems on self-evolution, efficiency, and human-comparable performance across languages. Tracks dynamics like solving time and improvements over iterations.
ArXiv AI · 217 天前
研究
Auto-Shaping Rewards for Robust Control
Proposes causal reward shaping from offline data for continuous RL under confounders. Derives tight value bounds via causal Bellman equation for PBRS.
ArXiv AI · 217 天前
基建
Authenticated Workflows Secure Agentic AI
Introduces authenticated workflows as a complete trust layer for enterprise agentic AI, protecting prompts, tools, data, and context. Enforces intent and integrity via cryptography and MAPL policy language.
ArXiv AI · 217 天前
應用
AugVLA-3D Boosts VLA with Depth Augmentation
AugVLA-3D integrates depth estimation from RGB inputs via VGGT to enrich 3D features in vision-language-action models. An action assistant module ensures consistency with control tasks.
ArXiv AI · 217 天前
音訊
AudioRouter Boosts LALMs via RL Tool Use
AudioRouter applies RL to teach large audio language models (LALMs) when to use external audio tools, improving fine-grained perception without heavy training. It optimizes a lightweight routing policy while freezing the base model.
ArXiv AI · 217 天前
研究
Aletheia Powers Autonomous Math Research
Aletheia is a math research agent that generates, verifies, and revises solutions using advanced Gemini Deep Think. It achieves milestones like fully AI-generated papers, human-AI collaborations, and solving four open Erdos problems.
ArXiv AI · 217 天前
應用
AI-PACE Framework Boosts Medical AI Education
AI-PACE synthesizes literature to propose a framework for integrating AI into medical education across the learning continuum. It identifies key competencies, curricular approaches, and strategies emphasizing longitudinal integration and interdisciplinary collaboration.
ArXiv AI · 217 天前
模型
AI Fails Basic Arithmetic Despite Advanced Math Wins
Frontier AI models excel in advanced math but consistently fail at multi-digit integer addition. Errors primarily stem from operand misalignment or carry failures, explaining most mistakes in top models like Claude, GPT, and Gemini.
ArXiv AI · 217 天前
基建
AgentTrace Enables AI Agent Observability
AgentTrace instruments LLM agents for structured logging across operational, cognitive, and contextual traces. Provides runtime transparency for security and monitoring in high-stakes settings.
ArXiv AI · 217 天前
模型
Affordances Build Partial LLM World Models
Proves LLMs possess predictive partial-world models via task-agnostic affordances for intents. Introduces distribution-robust affordances for multi-task efficiency.
ArXiv AI · 217 天前
應用
Adversarial Threat Detection in Autonomous Driving
AD² analyzes vulnerabilities in end-to-end driving agents like Transfuser to physics, EMI, and digital attacks in CARLA. Driving scores drop up to 99% under threats.
ArXiv AI · 217 天前
模型
Adapters Unlock Reliable Self-Interpretation
Lightweight adapters trained on interpretability artifacts enable reliable self-interpretation in frozen LMs. A simple scalar affine adapter outperforms baselines in feature labeling, topic identification, and implicit reasoning decoding.
ArXiv AI · 217 天前
研究
ADAlign Auto-Adapts Graph Domains
ADAlign tackles graph domain adaptation by adaptively aligning discrepancies via Neural Spectral Discrepancy (NSD). Uses neural characteristic functions and minimax sampling without heuristics.
ArXiv AI · 217 天前
圖像
1% Params Beat Full Fine-Tuning
CoLin introduces a 1% parameter low-rank complex adapter for vision foundation models. It resolves convergence issues in composite matrices with tailored loss.
ArXiv AI · 217 天前
商業
AI Siri Before Cook Retires?
The article questions whether Apple's AI-upgraded Siri will launch before CEO Tim Cook retires. It emphasizes that while delays are tolerable, outright failure is unacceptable.
Ifanr (爱范儿) · 217 天前
商業
Samsung S26 End-Month Debut, 2nm Chip
Samsung Galaxy S26 is slated for reveal by month's end in a tech news roundup. It may introduce the first 2nm processor in smartphones.
Ifanr (爱范儿) · 217 天前
研究
Simpler Model Predicts 99% AI R&D Automation by 2032
Introduces a robust, 8-parameter model forecasting >99% AI R&D automation by late 2032. Based on conservative compute growth and algorithmic trends, it predicts 1000x-10M x efficiency gains and 300x-3000x research output by 2035.
AI Alignment Forum · 217 天前
研究
2032 AI R&D Automation Predicted
Simplified model forecasts 99% AI R&D automation by late 2032 via compute and algo trends. Uses 8 parameters, conservative assumptions like no full automation.
AI Alignment Forum · 217 天前
模型
Trace Length Signals LLM Uncertainty
Reasoning trace length serves as simple confidence estimator in LLMs to combat hallucinations. Performs comparably to verbalized confidence across models, datasets, prompts.
Apple Machine Learning · 217 天前
模型
Trace Length as LLM Uncertainty Signal
Apple researchers demonstrate that reasoning trace length serves as a simple, effective confidence estimator in large reasoning models. It performs comparably to verbalized confidence across models, datasets, and prompts, acting complementarily.
Apple Machine Learning · 217 天前
.png)
基建
Together AI Launches 2.6x Faster Inference
Together AI introduces Dedicated Container Inference, a production-grade orchestration for custom AI models. It delivers 1.4x–2.6x faster inference speeds.
Together AI Blog · 217 天前
研究
Real-World Tool Agent Evaluation
Hugging Face explores OpenEnv for evaluating tool-using AI agents in practical settings. The post details methodologies for real-world testing.
Hugging Face Blog · 217 天前
研究
OpenEnv Evaluated in Real-World Agent Environments
Hugging Face blog explores OpenEnv for evaluating tool-using AI agents in practical settings. It highlights real-world applications beyond simulated benchmarks.
Hugging Face Blog · 217 天前
研究
Mapping UX Design for Computer Agents
Study maps UX design space for LLM-based computer use agents via two-phase research. Phase 1 reviewed systems and interviewed eight UX/AI practitioners to create taxonomy.
Apple Machine Learning · 217 天前
基建
Dedicated Container Inference: 2.6x Faster AI
Together AI launches Dedicated Container Inference for production-grade orchestration of custom AI models. It delivers 1.4x–2.6x faster inference speeds compared to standard methods.
Together AI Blog · 217 天前