Hugging Face Introduces BM25 + Seb as Cost-Effective RAG Alternative
Hugging Face's new approach using BM25 and Seb outperforms 8B vector RAG in quality, speed, and cost for Bible QA and production search.
387 articles
In-depth coverage of new AI research — papers, benchmarks, and breakthroughs from leading labs and academia, summarized for fast reading and grounded in the methods that matter.
Hugging Face's new approach using BM25 and Seb outperforms 8B vector RAG in quality, speed, and cost for Bible QA and production search.
OpenAI has released a proof for part of the Yang-Mills existence and mass gap problem, a major challenge in theoretical physics. The proof, verified using Lean, has been assessed by a special group.
IBM outlines technical advancements toward its IBM Quantum Starling project, aiming for fault-tolerant quantum computing by the end of the decade.
AWS unveiled a semantic diffing tool that uses anti-unification to compare formal policies, helping identify meaningful differences in policy expressions.
Apple researchers introduced a new method called weighted state souping to enhance long-context retrieval performance in state space models.
Google Research unveils a machine learning framework that translates wearable data into actionable health insights for cardiovascular-kidney-metabolic risk profiling.
Hugging Face demonstrates how to add new domain experts to existing MoE models without retraining the base model, improving performance on specific tasks like Rust and Haskell programming.
A new book by mathematicians Steven Strogatz and Alex Townsend examines the role of mathematics in an era dominated by AI and large-scale computational challenges.
A startup leverages AI to tackle a long-standing scientific challenge, signaling a shift in how research is conducted.
Hugging Face launches a new benchmark to assess how well AI models can animate static vector graphics, with results showing only four models exceed 50% pass rate.
Google DeepMind researchers used agentic AI to mathematically prove key cryptographic guarantees of the Signal protocol, marking a milestone in AI-aided formal verification.
New research highlights trust as the key barrier to AI adoption, with Australia serving as a critical test case for global scalability.