LLM-as-a-Coach: Experiential Learning for Non-Verifiable Tasks
Experiential Learning repurposes LLM-as-Judge feedback into coaching signals for policy RL on open-ended tasks, preserving textual feedback bandwidth.
Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Experiential Learning repurposes LLM-as-Judge feedback into coaching signals for policy RL on open-ended tasks, preserving textual feedback bandwidth.
FinSAgent multi-agent RAG system aligns retrieval queries to SEC filing structure and terminology for evidence-grounded financial QA.
Heterogeneous adaptation pipeline enables on-device model personalization by offloading INT8 backbone inference to Hailo-8L accelerator.
Activation steering enables fine-grained control over LLM reasoning trajectories by breaking self-loops via latent state manipulation.
VDAR-Router selects LLMs via verbalized query difficulty analysis for cost-performance-aware routing without embedding-only heuristics.
Welcome to the “what is a photo” debate, Adobe. | Image: Adobe Adobe's experimental camera app has taken an unexpected turn. After Project Indigo was launched last year to provide a "more natural (SLR-like) look" for iPhone photography, the Indigo camera app is now being…
SciForma ensures structural fidelity in scientific diagram generation via conjunctive reward design; outperforms SFT and scalar-reward baselines.
Analysis of per-class coverage under distribution shift; split conformal prediction fails per-class validity on skeleton benchmarks.
Evidence-sufficiency prompting reduces clinical LLM overconfidence but gains are judge-dependent; tests GPT-4.5, Claude Opus, Gemini, Grok on real data.
WorldCupArena: dynamic benchmark for LLMs and research agents on real-time sports forecasting with 2026 FIFA World Cup.
Self-distillation method improves rubric-based RL for LLMs by addressing train-inference mismatch in open-ended task optimization.
SelectInfer: neuron-level optimization framework for efficient LLM inference on edge devices via selective neuron loading.
Agentic framework for multimodal video misinformation detection via sparse evidence seeking rather than exhaustive processing.
The demand for AI continues to accelerate. Workloads are getting larger, models are becoming more complex, and there is mounting pressure to deploy AI compute... The demand for AI continues to accelerate. Workloads are getting larger, models are becoming more complex, and there…
Theoretical analysis of Bellman equation decomposition via three dualities in sequential decision-making and reinforcement learning.
Empirical study comparing silence thresholds in sitcoms vs. Google NotebookLM synthetic podcasts by gender and production setting.
Adobe's Project Indigo can now remove all kinds of backgrounds from photos you snap using the app.
Sobek: streaming optimization for equivariant tensor product convolutions on graphs via memory-efficient execution scheduling.
Similarity-based Generative Network (SGN) for domain-shift data augmentation without target-domain parameter updates.
Hardware-level dynamic throttling mechanisms for fine-grained AI performance control as safety intervention beyond software safeguards.
HuGLEN: LLM evaluation pipeline for optical network automation using expert ratings and quality-efficiency scoring.
New benchmark Pancasila-Dilemmas (1,834 questions) evaluates LLM value alignment on Indonesian cultural values beyond Western frameworks.
Study of autoresearch agents (Claude Code) on Quranic speech-recognition tasks reveals metric-gaming vs. intent-alignment tradeoffs.
Adaptive Adversaries benchmark: 21-scenario multi-turn adaptive attack suite for LLM agent security with autonomous attacker pivoting.
Intern-BioBreaker red-teaming framework stress-tests frontier LLMs for biosecurity risks via jailbreak prompts and wet-lab validation.
YouTube has updated its monetization policies to more clearly define the kinds of AI-generated and low-quality videos that can’t earn ad revenue.
SEE framework synthesizes long-horizon GUI agent trajectories via structure-aware exploration for mobile task automation without costly human demos.
Theoretical analysis of pooled information reducing search coverage via one-answer rule; solvable benchmark with 16 boxes and 8 agents.
Adaptive Mamba Neural Operators apply state-space models to PDE solving on arbitrary geometries using Takenaka-Malmquist systems.