Latent.Space
289 episodes listed below
PRESS PLAY
The podcast by and for AI Engineers! In 2025, over 10 million readers and listeners came to Latent Space to hear about news, papers and interviews in Software 3.0. We cover Foundation Models changing every domain in Code Generation, Multimodality, AI Agents, GPU Infra and more, directly from the founders, builders, and thinkers involved in pushing the cutting edge. Striving to give you both the definitive take on the Current Thing down to the first introduction to the tech you'll be using in the next 3 months! We break news and exclusive interviews from OpenAI, Anthropic, Gemini, Meta (Soumith Chintala), Sierra (Bret Taylor), tiny...
Based on 6 episodes we listened to · September 2026
Latent Space makes AI engineering legible by pressing builders on system behavior, infrastructure tradeoffs, workflow design, and failure modes.
Latent Space turns fast-moving AI products and research into searching technical conversations. Alessio and Swix press guests on how systems behave in practice, moving from agentic AI and desktop workflows to model reasoning and inference infrastructure without sanding away the…
An agent repeats a mistake because the mistake stayed in its context; now you have questions.
Press play if you want
Short reviews from the PlayNext desk, based on the episodes we processed.
A clear look at how AI performance depends on moving data efficiently across the stack.
Jeff Dean's conversation reads less like a victory lap than a compact lesson in hardware and systems tradeoffs. It stays grounded in specifics: SRAM versus HBM, the cost of moving a parameter compared with the multiply itself, and the appeal of batch size one for latency even…
Brex treats AI buying as a practical portfolio decision.
The episode works because it treats Brex's AI strategy as an operating plan, not a slogan. Three pillars stand out: corporate AI for buying tooling across the business, operational AI for cutting the cost of running a financial institution, and product AI for features that can…
Brex treats AI as a buying, operating, and product problem.
Brex's AI strategy comes through as an operating model rather than a slogan. The piece traces three pillars: company-wide AI adoption, internal systems meant to lower the cost of running a financial institution, and product features that can fold Brex into a customer's own…
Cursor's cloud agents are framed as parallel work for models and teams, with setup and VM sizing still evolving.
Cursor's cloud agents are cast as part of a broader shift toward agentic infrastructure: an LLM judge that can also write code, parallel agents, and blank VMs that run in their own computers.
Independent AI benchmarking turns into a business story, then into a debate about what model scores really measure.
Artificial Analysis reads like an independent third-party comparison site for model quality and throughput, with model and hosting provider breakdowns that give it a practical, infrastructural feel.
Agent serving gets real when security, latency, and cost all have to be balanced.
The discussion treats agent inference as an operating problem, not just a model problem. It argues that when an agent can reach files, the internet, and custom code, the practical move is to keep any two of the three and set clear enforcement points.
AI in science gets real when the data stops behaving.
Andrew White's episode tracks a simple but useful shift: AI is no longer just an idea floating above science, it is running into the daily friction of real work.
A sharp look at how AI benchmarking becomes something developers and companies can actually use.
Artificial Analysis comes across as an independent third-party site for comparing models and hosts, and the value is clear: it turns raw data into a practical way for developers and companies to weigh quality against throughput.
Markets, macro, AI, and crypto in motion.
Daily AI news through the lens of trust, infrastructure, and real-world deployment.
Dreamer starts as a consumer agent platform.
John Yang, creator of the influential SweetBench coding benchmark, discusses the current state of AI coding evaluations.
Latent Space covers Cursor's major launch of cloud agents — a system in which an AI coding agent is given a full computer environment, runs tests end-to-end, records a video of what it built, and returns a reviewed pull request rather than a speculative one.
Yi Tay discusses his return to Google DeepMind, the IMO gold medal win using Gemini DeepThink, and insights on on-policy RL, AI coding, and the path to AGI.
Latent Space hosts an extended conversation with Aaron Levie, CEO of Box, about why enterprise file storage and permissions infrastructure is becoming critical for the agentic AI era.
Latent Space hosts Alessio and Swyx interview James Reggio, CTO of Brex, about the company's AI transformation over 18 months that achieved 5x accelerated growth and a 99 percent burn reduction.
George Cameron and Micah-Hill Smith from Artificial Analysis discuss their journey building independent LLM evaluation as a service.
Nathan Lambert and Sebastian Raschka discuss Anthropic's blog post on distillation attacks by Chinese labs, exploring the ethics and detection of model training via APIs.
The guest is back, and nobody waits around.
So what he did is he kind of like found us at Berkeley and picked us out of the basement and was lik
MD was supposed to solve protein folding.
Materials search turns into a strange kind of computing.
Agent power has a strange safety curve.
Felix Rieseberg describes Cloud Co-Work as cloud code in a virtual machine, built for people who want task automation without living in the terminal.
Kevin Wang says the Princeton team started from a surprising question: why do frontier RL systems still use two- or four-layer networks when language and vision have moved to much larger models.
Mark and Myra from Goodfire join Latent Space to discuss the company's $150 million Series B at a $1.25 billion valuation, announcing the first mechanistic interpretability-focused research lab to reach unicorn status.
George and Micah, co-founders of Artificial Analysis, discuss the origins and evolution of their independent AI benchmarking platform.
Latent Space interviews Joel Becker, co-founder of METR (Model Evaluation and Threat Research), the organization that conducts independent AI capability and risk evaluations for frontier labs, about how METR thinks about measuring AI capabilities at exponential time horizons, what their evaluations of models like GPT-5 found, and why the gap between what AI can theoretically do and what it actually does in deployment is one of the most important and underappreciated dynamics in AI safety.
Gabriela Corso and Jeremy Volvin discuss their journey from MIT PhDs to founding Boltz, an open-source initiative to democratize protein structure prediction.
Simon Eskildsen, founder of Turbo Puffer, discusses the origins and vision of his search engine company, from a cost-driven insight at Readwise to scaling for AI workloads.
Jeff Dean, Google's chief AI scientist, discusses owning the AI Pareto Frontier through balancing frontier and efficient models, the history and future of distillation, hardware-energy trade-offs, and scaling AI to attend to trillions of tokens.
OpenAI researchers Mia Glaese and Olivia discuss why C-Bench Verified, a key coding benchmark, has become saturated and contaminated, leading to unreliable measurements.
Brex CTO James Reggio discusses the company's AI transformation, revealing their three-pillar strategy covering corporate adoption, operational automation, and product innovation.
The hosts interview the authors of a NeurIPS Best Paper on training 1000-layer networks for self-supervised reinforcement learning.
Kevin Weil, VP of OpenAI for Science, and Victor Powell, product lead, announce Prism — a free AI-native LaTeX editor designed to accelerate how scientists write and publish research.
The launch episode of a new AI for Science podcast on the Latent Space network features Andrew White of Greensboro AI, who works on RNA therapeutics using machine learning at Atomic AI, and RJ Haneke, co-founder of MiraOmics, discussing the emerging convergence of AI engineering and life sciences research.
Latent Space hosts Alessio and Swyx record a live episode at Andreessen Horowitz with Martin Casado and Sarah Wang, two senior partners at a16z, covering the major tensions in the current AI investment landscape including the comparison between Anthropic and OpenAI as investment theses, the unusual hybrid nature of frontier AI model companies that require both venture and growth capital from seed stage, and the lessons learned from past investments in large pre-revenue AI companies.
George and Micah from Artificial Analysis discuss their journey building an independent AI benchmarking platform, covering their business model, technical challenges like hallucination metrics, and industry trends such as the falling cost of intelligence.
Douglas joins the show for a conversation about Claude Code and how AI is showing up inside analyst work at our firm.
Anastasios Angelopoulos says LMArena now handles mid tens of millions of conversations per month and funds all inference itself, which is why the company raised around $100 million.
Discussion with Notion's Simon and Sarah about their custom agents launch, MCPs, and AI tooling for enterprise.
Interview with Ryan Lopopolo about harness engineering and building AI products using Codex.
Discussion on the long history of AI research leading to recent breakthroughs like ChatGPT and O1, with a focus on AI engineering.
Discussion on the difficulty of benchmarking world models and AI systems, with an interview featuring Moonlake's founders.
Mistral announces VoxTral TTS, a new speech generation model, discussing its architecture, performance, and cost efficiency.
The hosts introduce guests Alex Lieberman and Armand, co-founders of 10X, discussing their backgrounds and the AI engineering ecosystem.
An interview with Didi Das discussing his journey from Glean to venture capital, covering AI startups and the evolution of LLMs like Claude.
An interview with Didi Das discussing his journey from Glean to venture capital, covering AI startups and the evolution of LLMs like Claude.
An interview with Jed Borovik at GitHub Universe discussing the New York tech scene and AI engineering.
An interview with Jed Borovik at GitHub Universe discussing AI engineering, the New York tech scene, and partnerships.