News
Beyond a Single Extractor: Re-thinking HTML-to-Text Extraction for LLM Pretraining
2+ hour, 58+ min ago (310+ words) Beyond a Single Extractor: Re-thinking HTML-to-Text Extraction for LLM Pretraining'Apple Machine Learning Research Beyond a Single Extractor: Re-thinking HTML-to-Text Extraction for LLM Pretraining One of the first pre-processing steps for constructing web-scale LLM pretraining datasets involves extracting text from HTML....
The Potential of CoT for Reasoning: A Closer Look at Trace Dynamics
3+ hour ago (64+ words) AuthorsGregor Bachmann, Yichen Jiang, Seyed Mohsen Moosavi Dezfooli, Moin Nabi The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity June 5, 2025research area Computer Vision, research area Speech and Natural Language Processingconference ACL…...
AMUSE: Audio-Visual Benchmark and Alignment Framework for Agentic Multi-Speaker Understanding
3+ hour ago (348+ words) AMUSE: Audio-Visual Benchmark and Alignment Framework for Agentic Multi-Speaker Understanding'Apple Machine Learning Research AMUSE: Audio-Visual Benchmark and Alignment Framework for Agentic Multi-Speaker Understanding Recent multimodal large language models (MLLMs) such as GPT-4o and Qwen3-Omni show strong perception but struggle in…...
Apple Workshop on Reasoning and Planning 2025
1+ day, 3+ hour ago (914+ words) Apple ML researcher Iman Mirzadeh presenting at the workshop. Last year, Apple hosted the Workshop on Reasoning and Planning, bringing together Apple researchers and members of the broader research community for a two-day event focused on advancing the state of…...
Models That Prove Their Own Correctness
1+ week, 2+ hour ago (50+ words) AuthorsNoga Amit, Shafi Goldwasser, Orr Paradise, Guy N. Rothblum Fingerprinting Codes Meet Geometry: Improved Lower Bounds for Private Query Release and Adaptive Data Analysis January 10, 2025research area Methods and Algorithms, research area Privacy Our research in machine learning breaks new ground every…...
Asynchronous Verified Semantic Caching for Tiered LLM Architectures
1+ week, 1+ day ago (354+ words) Asynchronous Verified Semantic Caching for Tiered LLM Architectures'Apple Machine Learning Research Asynchronous Verified Semantic Caching for Tiered LLM Architectures AuthorsAsmit Kumar Singh, Haozhe Wang, Laxmi Naga Santosh Attaluri, Tak Chiam, Weihua Zhu Large language models (LLMs) now sit in the…...
Faster Rates For Federated Variational Inequalities
1+ week, 4+ day ago (288+ words) Faster Rates For Federated Variational Inequalities'Apple Machine Learning Research Faster Rates For Federated Variational Inequalities AuthorsGuanghui Wang, Satyen Kale In this paper, we study federated optimization for solving stochastic variational inequalities (VIs), a problem that has attracted growing attention in…...
Trace Length is a Simple Uncertainty Signal in Reasoning Models
1+ week, 4+ day ago (321+ words) Trace Length is a Simple Uncertainty Signal in Reasoning Models'Apple Machine Learning Research Trace Length is a Simple Uncertainty Signal in Reasoning Models AuthorsSiddhartha Devic, Charlotte Peale, Arwen Bradley, Sinead Williamson, Preetum Nakkiran, Aravind Gollakota Uncertainty quantification for LLMs is…...