engblogs

How the MIT-IBM Watson AI Lab accelerates early-career faculty impact by enabling cross-disciplinary collaboration and access to computational resources to advance NLP and LLM research from seed projects to scalable, real-world applications.

REA autonomously drives Meta's ads ranking end-to-end ML lifecycle, enabling multiweek, hypothesis-driven experimentation with a Hibernate-and-Wake mechanism and budget-aware resilience, delivering doubled model accuracy and fivefold engineering productivity gains.

TrajTok unveils an end-to-end video tokenizer that learns adaptive trajectory tokens through a unified space-time segmenter, enabling downstream task–driven token granularity and enhanced video understanding.

Public preview of Qwen3-Embedding-0.6B on Databricks enables retrieval-powered, multilingual enterprise AI agents grounded in governed data via Agent Bricks and Vector Search.

Clarifies the AI vs. machine learning distinction, contrasts rule-based and learning-based approaches, and maps real-world applications and practical considerations across data, models, and deployment.

A concise, technical overview of data science skills, career paths and education options, linking foundations in statistics and programming to ML engineering, data engineering and business analytics with practical learning pathways.

Concise introduction to GPT-5.4 mini and nano, highlighting their compact architectures and practical implications for developers.

AMES presents a backend-agnostic, two-stage late-interaction framework that unifies text, image, and video into a shared space for approximate multimodal enterprise search, employing ANN-based parallel token search with per-document Top-M MaxSim re-ranking in production Solr deployments.

Airbnb's evolution from vendor-driven observability to internal ownership, sharing hard-won lessons on reliability and platform visibility.

Case study of using DSPy to adapt and optimize Dash's relevance judge across models, yielding lower NMSE, improved reliability, and scalable, cost-efficient production.

Previewing multi-cluster GKE Inference Gateway enables globally scalable, fault-tolerant AI inference by intelligently routing across multiple clusters and regions with model-aware load balancing and shared accelerator resources.

Atos partners with AWS to scale AI education through the AWS AI League, combining hands-on, gamified learning with SageMaker JumpStart to fine-tune LLMs for real-world insurance underwriting and deliver measurable AI fluency at scale.