AI Shorts

Baidu AI Research crew has simply launched ERNIE-4.5-21B-A3B-Thinking, a brand new reasoning-focused giant language mannequin designed round effectivity, long-context reasoning, and gear integration. Being a part of the ERNIE-4.5 household, this mannequin is a Mixture-of-Experts (MoE) structure with 21B whole parameters however solely 3B energetic parameters per token, making it computationally environment friendly whereas sustaining…

A workforce of researchers from MBZUAI’s Institute of Foundation Models and G42 launched K2 Think, is a 32B-parameter open reasoning system for superior AI reasoning. It pairs lengthy chain-of-thought supervised fine-tuning with reinforcement studying from verifiable rewards, agentic planning, test-time scaling, and inference optimizations (speculative decoding + wafer-scale {hardware}). The result’s frontier-level math efficiency with…

Table of contents Why is long context such a bottleneck for LLMs? How does REFRAG compress and shorten context? How is acceleration achieved? How does REFRAG preserve accuracy? What do the experiments reveal? Summary FAQs A staff of researchers from Meta Superintelligence Labs, National University of Singapore and Rice University has unveiled REFRAG (REpresentation For…

Large language fashions (LLMs) fairly often generate “hallucinations”—assured but incorrect outputs that seem believable. Despite enhancements in coaching strategies and architectures, hallucinations persist. A brand new analysis from OpenAI offers a rigorous rationalization: hallucinations stem from statistical properties of supervised versus self-supervised studying, and their persistence is strengthened by misaligned analysis benchmarks. What Makes Hallucinations…