3. Fine-tuning LLM Agents without Fine-tuning LLMs A memory‑based learning framework that lets deep‑research agents adapt online without updating model weights.
@dair_ai
-
Jet-Nemotron: Hybrid LLM Architecture with Optimized Attention
By
–
4. Jet-Nemotron A hybrid-architecture LM family: starting from a frozen full-attention model, the authors search for where to keep full attention, which linear-attention block to use, and which hyperparameters match hardware limits.
-
Deep Think with Confidence: Pruning Weak Reasoning Paths
By
–
2. Deep Think with Confidence
— DAIR.AI (@dair_ai) 31 août 2025
A lightweight test-time method that uses model-intrinsic confidence to prune weak reasoning paths, improving both accuracy and token efficiency for self-consistency ensembles.https://t.co/X9M1n0mKok2. Deep Think with Confidence A lightweight test-time method that uses model-intrinsic confidence to prune weak reasoning paths, improving both accuracy and token efficiency for self-consistency ensembles.
-
Reinforcement Learning Techniques for LLM Reasoning Evaluated
By
–
10. A Deep Dive into RL for LLM Reasoning This paper reviews and rigorously re-evaluates reinforcement learning techniques for LLM reasoning, addressing inconsistencies caused by varied setups and unclear guidelines.
-
Efficient LLM Architectures Beyond Traditional Transformers Survey
By
–
9. A Survey on Efficient Architectures for LLMs Reviews advances in efficient LLM architectures beyond traditional transformers, including linear and sparse sequence models, efficient attention variants, sparse MoEs, hybrid designs, and diffusion-based.
-
GLM-4.5: Open Mixture-of-Experts for Agentic AI
By
–
8. GLM-4.5 An open Mixture‑of‑Experts family that targets a single model excelling across agentic tool use, complex reasoning, and real‑world coding.
-
QA Hallucination Detectors: Evaluation Metrics Limitations
By
–
7. Illusion of Progress The paper argues that common QA hallucination detectors look better than they are because evaluations lean on ROUGE.
-
ASearcher: Open-Source Framework for Expert-Level LLM Search Agents
By
–
6. Beyond Ten Turns This paper introduces ASearcher, an open-source framework for training LLM-based search agents capable of long-horizon, expert-level search.
-
DINOv3 Self-Supervised Vision Foundation Model Scales Data
By
–
1. DINOv3 DINOv3 is a self‑supervised vision foundation model that scales data and model size, introduces a Gram anchoring loss to preserve dense patch consistency during long training, and adds post‑hoc tweaks for resolution, size, and text alignment.
-
Top AI Papers of the Week: DINOv3, M3-Agent, and Medical Reasoning
By
–
Top AI Papers of The Week (August 11-17): – DINOv3
– M3-Agent
– Illusion of Progress
– TRImodal Brain Encoder
– Efficient Architectures for LLMs
– Deep Dive into RL for LLM Reasoning
– GPT-5 for Multimodal Medical Reasoning Read on for more: