Back to blog home

AI and MLOps Roundup: November 2023

Multimodal models are a game changer, but do you actually understand how they work? What about embeddings and RAG? Check out our roundup of the top AI and MLOps articles for November 2023!

Machine Learning Engineering Guides and Tools

Stas Bekman has extensive experience training LLMs from his time at Hugging Face. His repo of ML engineering guides and tools can help you get started on your LLM journey:

ML Engineering guides and tools

Multimodality and Large Multimodal Models (LMMs)

Multimodal models are step change in AI. Chip Huyen explains the fundamentals of a multimodal system and current research advancing LMMs:

Both encoders and projection matrices are jointly trained together from scratch. The training goal is to maximize the similarity scores of the right (image, text) pairings while minimizing the similarity scores of the wrong pairings (contrastive learning).

What is Retrieval-Augmented Generation (RAG)?

Retrieval-augmented generation is all the rage, but do you actually understand RAG architecture and its benefits? Valeriia Kuka breaks it down for the rest of us:

Do you really understand RAG?

How to Monitor LLMOps Performance with Drift

Like predictive ML models, LLM performance also degrades over time. Learn the two types of performance problems and how to identify drift in prompts and responses:

Identify clusters of outlier prompts that caused drift and collate insights to improve LLM performance with fine-tuning or RAG

State of AI Report 2023

AI is much more than LLMs. From weather prediction to self-driving cars to music generation, the amazing State of AI report covers it all — and yes, that includes a whole lot of LLM research too:

2023 State of AI Report

How LinkedIn Is Using Embeddings to Up Its Match Game for Job Seekers

LinkedIn uses embedding-based retrieval to power search and recommendations across the platform, from the newsfeed to notifications to job openings. Their ML team shares how they built infrastructure to incorporate embeddings at scale:

An embeddings retrieval model

Embeddings: What they are and why they matter

Embeddings. You've probably heard of them. You may have worked with them. But how do they work and why do they matter?

3 dimensional space helps us understand more about embeddings

Retrieval Augmented Generation at scale — Building a distributed system for synchronizing and ingesting billions of text embeddings

Retrieval-augmented generation (RAG) is relatively straight forward, but the difficulty lies in scaling and becoming production-ready. And what if you're trying to ingest a billion records? Here's a great guide on how to get it done:

Distributed RAG pipeline

Stay up to date on everything AI and MLOps by signing up for our newsletter below.