Latent Space

How Anthropic ships the #1 LLM in the World — with Katelyn Lesse and Angela Jiang

Anthropic’s recent launch of the Opus 5.5 model marks a strategic shift toward optimizing intelligence-per-dollar and reducing model verbosity. Engineering and product leaders emphasize that scaling agentic capabilities requires a "brain-and-hands" architecture, where durable harnesses manage state while isolated sandb...

Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal

The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi

AI in NYC: How Tenex Builds an AI-Native Engineering Team — Alex Lieberman & CJ Hess

The Programmable Data Center — Evan Conrad, SF Compute

Robots That Predict the Future: Dyna-2 & Physical AI — Jason Ma, Dyna Robotics

Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub

The integration of artificial intelligence into bioscience requires a shift from viewing data and modeling as separate, rigid disciplines to a problem-first, multidisciplinary approach. While scaling laws suggest that increased compute and data improve performance, success depends on curating high-quality, relevant inf...

AI Scientists Are Here: Autonomous Labs & Synthesis Superintelligence — Periodic Labs

OpenAI’s New Agent Stack: Computer Use, Decisions API, UltraFast, Dots—Ari Weinstein & Nikunj Handa

Computer use agents have reached a performance milestone where they now execute software tasks faster than the average human, transitioning from simple automation to complex, multi-step debugging and execution. Recent advancements, including the introduction of the Decisions API and updates to the Agents API, enable fa...

The Future of Claude Code: Mods, Mutable Software, & Multiplayer Agents — Thariq Shihipar, Anthropic

The $10 Trillion Token Economy — Alex Atallah, OpenRouter & Anjney Midha, AMP

Runway’s Bet Beyond Video: World Models, Robotics, and the Neural OS — Anastasis Germanidis

🔬 Google's AI Scientist Started as an Attempt to Automate Kaggle — John Platt, Google Fellow

Why I couldn't build Jev at OpenAI — Diogo Almeida, TypeSafe Co-founder & CEO

Jev represents a shift toward "System One" models designed for programmatic consumption rather than human-facing chatbots. By prioritizing intelligence-per-dollar and reliability, these models integrate directly into software workflows, enabling the automation of rote, economically valuable tasks that current reasoning...

Recursive Self-Improvement: from Auto Research to Superintelligence — Richard Socher, Recursive

The Eureka Machine represents the ultimate goal of superintelligence: a system capable of autonomous invention across scientific and engineering domains. Rather than regulating abstract intelligence or compute usage, which risks totalitarian overreach, policy should target specific, high-stakes applications like medica...

Simulating Humanity: from Generative Agents to 8 Billion Digital Twins — Joon Sung Park, Simile AI

Exo: Harnesses should see their own code and logs — Alex Krentsel, UC Berekeley / Google Research

🔬They Thought the Model Was Broken — Matt McPartlon & Neil Patil, Chai Discovery

Next 100x in AI: Inference, Networking, & Self-Optimizing Models — Philip Kiely & Ali Taha, Baseten

Inference engineering requires balancing model fidelity with computational efficiency through techniques like quantization, speculative decoding, and disaggregated prefill-decode architectures. Moving models from general-purpose hardware to production-ready APIs involves complex optimizations, such as calibrating NVFP4...

OpenAI’s Vision for the AI Super App — Akshay Nathan, OpenAI

The integration of AI into professional workflows marks a shift toward "bottoms-up ambition," where accessible, agentic tools empower users regardless of their technical background. Akshay Nathan, lead of core product engineering at OpenAI, explains that ChatGPT Work serves as a unified productivity environment, abstra...