The Economist

AI models are being prepared for the physical world

They call it a “world model”, an essential tool to help AI systems make sense of the complex, unpredictable physical spaces ...

Mercury 2 : World’s Fastest Reasoning AI Model Built for Production Applications

The new Mercury 2 AI model uses diffusion reasoning to generate 1,000 tokens per second; it runs about 5x faster than Haiku, speed limits are ...
GitHub

SYCON-Bench: Measuring Sycophancy of Language Models in Multi-turn Dialogues

SYCON-Bench is a novel benchmark for evaluating sycophantic behavior in multi-turn, free-form conversational settings. This benchmark measures how quickly a model conforms to the user (Turn of Flip) ...

AI develops easily understandable solutions for unusual experiments in quantum physics

Researchers at the University of Tuebingen, working with an international team, have developed an artificial intelligence that designs entirely new, sometimes unusual, experiments in quantum physics ...
InfoWorld

Multi-token prediction technique triples LLM inference speed without auxiliary draft models

With reported 3x speed gains and limited degradation in output quality, the method targets one of the biggest pain points in production AI systems: latency at scale.
IEEE

Comparison of Classical and Sequential Model Predictive Control Techniques in a Grid Connected Three-Level NPC

Abstract: This paper compares Classical Model Predictive Control (CMPC) and Sequential Model Predictive Control (SMPC) for a three-level Neutral Point Clamped (NPC) converter. Although CMPC is simple ...

Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
GitHub

⚡ LitLogger - A minimal logger for AI experiments

When building AI, you change many things at once: code, data, prompts, models. After a few runs, it becomes unclear what actually caused results to improve or regress. LitLogger records every run as ...
IEEE

Model-Agnostic Empirical Evaluation of Test-Driven Prompt Engineering on Improving Accuracy and Efficiency in Large Language Models Python Code Generation

Abstract: Although Large Language Models (LLMs) are widely adopted for code generation, the generated code can be semantically incorrect, requiring iterations of evaluation and refinement. Test-driven ...