Latent Space

What it is

Latent Space is a technical publication about the business and engineering of AI. It was founded by swyx (Shawn Wang), with rotating co-hosts including Alessio Fanelli and Vibhu Subramanian. According to its about page, the podcast ranks among the top US technology podcasts and the publication reaches more than 200,000 subscribers and over 10 million views across its channels.

It includes:

• A podcast (audio and video) with long-form interviews of researchers, engineers and founders.
• A newsletter with analysis and curated AI news.
• A Discord community with a weekly LLM Paper Club where members discuss research papers.
• Connections to the AI Engineer Summit and World's Fair, conferences associated with the same team.

The publication is closely tied to the idea of the "AI engineer" as a distinct role: software engineers who build products on top of models rather than training them.

Why it is worth the time

Most coverage of AI is written for people deciding whether to invest in it or worry about it. Latent Space is written for people who have to build with it. Conversations get into evaluation problems, inference costs, model training details, developer tools, what did not work and why, and the unglamorous engineering behind products.

How to use it

• Background listening, not study. Its value is keeping your map of the field current, so you are not surprised by a technique or tool that has become standard.
• Pick episodes by guest and topic. You do not need every episode; choose those related to what you are building.
• Join the Paper Club if you want to learn to read research papers with others.
• Take one action per episode: try a tool mentioned, read a referenced paper, or note an idea for a project.

Before you start

Episodes assume familiarity with language models, APIs and common terms such as RAG, fine-tuning, inference and agents. Beginners will find some conversations dense. If that is you, build a small AI project first; the episodes will make far more sense afterwards.

Where it falls short

It is fast-moving commentary, so older episodes date quickly. Guests are often founders with products to promote, so treat claims about their own tools with appropriate scepticism. Long episodes require time; the newsletter is a lighter way to follow.

Common questions

Is it free? The podcast and main newsletter are freely available; check the site for any paid options.

Is it for beginners? It suits working engineers and advanced students more than complete beginners.

Where do I listen? On the site and the usual podcast and video platforms.

Best for: engineers who build with AI and want to stay oriented without reading every launch announcement.