#Research
50 articles with this tag
The Assistant Axis LLM: How Researchers Are Capping AI Drift
Scientists have mapped the internal neural space of LLMs, identifying the "Assistant Axis" as the key to stabilizing AI persona and preventing harmful behavior.
Israel's Technion Researchers Championing AI Agent Negotiation Technology, Release the 'Poisoned Apple Effect' Threat to Agentic Commerce
The mere availability of a new AI model, even if unused, can strategically coerce regulators into changing market rules to benefit the model's releaser.
AI2's Molmo2 VLM brings pixel-perfect grounding to open video
Molmo2 VLM from AI2 establishes a new state of the art for open-source video understanding by introducing precise spatio-temporal grounding capabilities.
Uniqueness-Aware RL stops LLMs from getting lazy
Uniqueness-Aware RL prevents LLMs from converging on a single solution path by explicitly rewarding correct answers that employ rare problem-solving strategies.
Hybrid Scaling and Peer Review Elevate Software Engineering Agents
The TEX framework uses cross-candidate execution feedback to create collaborative Software engineering agents, achieving state-of-the-art results in test generation.
NeoVerse cracks the scalability problem for monocular 4D models
PhysBrain model uses human video to teach robots physical intelligence
DeepSeek Unveils mHC: A Mathematical Fix for the "Exploding Stream" Problem in Large Models
AI’s safety net relies on chain-of-thought monitorability
UCSD Lab Advances Low-Latency LLM Serving with DGX B200
Bolmo Advances Byte-Level Language Models with Practicality
Apple Engineers Squeeze Powerhouse Vision Models into a Single Layer for Hyper-Efficient Image Generation
The Generative AI Threat is Already in Your Browser: Malicious Chrome Extensions Explode in Latest Cyber Scourge
AI in Schools: Teens Demand Guidance, Not Bans
NVIDIA AI Fellowships Fuel Next-Gen AI Research
AI Model Confessions: A New Honesty Layer
OpenAI is Debugging LLM Misalignment: New Tools Emerge
\n Researchers are tackling the challenge of understanding and correcting undesirable LLM behavior with a new technique called latent attribution , detailed by ...
AlphaFold Protein Prediction Reshapes APAC Research
MIT's Iceberg Index Reveals AI's Hidden Trillion-Dollar Workforce Shock
AI Agent Negotiation Needs a Semantic Layer
Microsoft Fortifies AI Against Contextual Privacy Leaks
Escaping the Transformer Trap: Sakana AI's Quest for True Intelligence
Olmo 3 Open-Source AI Unlocks Full Model Flow Transparency
Cracking the Black Box: The Promise of Sparse Neural Networks
AlphaProof system proves its worth at the Math Olympiad
Taming the Beast: Automation IT Complexity Demands a New Approach
Nested Learning AI Tackles Catastrophic Forgetting
AI Agent Marketplaces Face Critical Flaws, Microsoft Research Finds
AI introspection is real, but it's unreliable
New AI Browser Prompt Injection Attacks Revealed
AI hits a wall on FrontierMath performance
Microsoft's ECHO Language Model learns from failure
AI Safety: Microsoft Uncovers Bio-Threats, Forges New Research Model
Ai2’s new AI climate emulator runs 1,500 years in a day
Google AI Research Awards Signal Strategic Priorities
AI Cancer Breakthrough: Gemma Model Uncovers New Therapy
Tiny Recursive Model Beats LLMs on Hard Puzzles
Tiny AI Model Outperforms Giants Redefining Scaling Laws
LoRA vs full fine-tuning: The debate is over
Unmasking the Biases of AI Judges: A Critical Look at LLM Fairness
OpenAI and Apollo Research Reveal AI Models Are Learning to Deceive: New Detection Methods Show Promise
The Bias-Variance Trade-off: An "Incredible Misnomer" in Deep Learning
The Real Reason for LLM Inference Nondeterminism
The true cause of LLM inference nondeterminism is not random GPU math, but a systemic failure of "batch invariance" tied to unpredictable server load.
OpenAI's Hallucination Breakthrough: A Feature, Not a Bug, and How to Fix It
AI Models that Compete, Mate, and Evolve Like Living Organisms
NVIDIA Research Details Vision for 'Physical AI' Future in Robotics
NVIDIA has outlined a significant strategic direction for its artificial intelligence research, focusing on what it terms "Physical AI." This initiative aims to bridge the gap between digital simulations and real-world applications, leveraging advanced AI...
Meta FAIR Wins Algonauts 2025 with a Trimodal Brain Model
Apple's FastVLM Targets More Efficient Vision-Language AI
Apple has quietly unveiled a new machine learning initiative, FastVLM, aimed at significantly improving the efficiency of vision language models (VLMs). FastVLM could unlock a new generation of responsive, privacy-preserving AI experiences that are deeply integrated into Apple’s hardware ecosystem.
Aeneas: Unlocking New Eras in Historical Discovery
Google DeepMind's Aeneas utilizes sophisticated machine learning to identify subtle patterns and semantic relationships across vast historical document datasets. Unlike traditional keyword-based methods, it discerns complex connections that often elude human researchers.
