Latent.Space

63 articles in the Truth Foundry index. Each links out to the original.

• Moonshot AI Releases Kimi K3: 2.8T Parameter Open Model Challenges Closed Leaders · Moonshot AI launches Kimi K3, a 2.8T parameter open model that ranks #1 in frontend coding benchmarks.
• Lila Sciences Builds AI-Driven Lab to Achieve Scientific Superintelligence · Lila Sciences aims to create a scientific superintelligence by treating automated labs as infinite data generators.
• OpenAI Codex Usage Surges 1M Daily as GPT-5.6 Sol Demand Hits Infra Limits · OpenAI's Codex agent platform adds 1 million users daily, driven by GPT-5.6 Sol demand.
• Five Trends Defining AI Engineering at World's Fair 2026 · AI engineering has matured from autonomous agents to harness engineering and loop-based oversight systems.
• Codex Usage Surges 10x in 6 Months to 7 Million Users, Outpacing Claude Code? · Codex's user base grew over 10x in six months, reaching 7 million, raising questions about its dominance over Claude Code.
• SpaceXAI Launches Grok 4.5, First Opus-Class Model Post-Cursor Acquisition · xAI launches Grok 4.5, an Opus-class model optimized for coding and agents, priced at $2/$6 per million tokens.
• Modal CTO Akshat Bubna: Cloud Must Evolve for Agent Experience · Modal raises $355M Series C to build cloud infrastructure optimized for autonomous AI agents.
• Lilian Weng Summarizes Harness Engineering for AI Self-Improvement · Lilian Weng analyzes 35 papers on harness engineering as the future of recursive self-improvement in AI.
• Thariq's Field Guide to Fable 5: Unhobbling Models and AI Advancements · Thariq's keynote on Fable 5 explores model unhobbling, AI benchmarks, and interpretability research.
• AI Engineers Debate Human Agency vs. Automated Autoresearch Loops · AIEWF speakers clash on whether humans must retain agency or delegate to autoresearch loops.
• Introspection CEO Roland Gavrilescu Unveils Autoresearch Infrastructure at AIEWF · Introspection launches Pi framework to enable self-improving agent loops in production.
• Cursor VP Pauline Brunet Outlines Enterprise AI Agent Deployment Strategy · Cursor's VP of Forward Deployed Engineering details the company's plan to build an AI software factory for enterprises.
• Genesis Molecular AI's PEARL Model Achieves Breakthrough in Protein-Ligand Docking · Genesis Molecular AI's PEARL model outperforms benchmarks in predicting protein-ligand binding poses without fine-tuning.
• Warp CEO Zach Lloyd Pivots to Software Factories and Agent Orchestration · Warp CEO Zach Lloyd announces a strategic pivot from terminal tools to automated software factory platforms.
• AI Engineer World's Fair Highlights Loops, Software Factories, and Forward Deployed Engineers · AIEWF day 2 focused on agent loops, autonomous software factories, and the rise of forward deployed engineers.
• Anthropic Releases Claude Sonnet 5 Amid Fable 5 Delays and Mixed Efficiency Reviews · Anthropic launches Claude Sonnet 5 as its new default mid-tier model while Fable 5 remains restricted.
• Sierra's Natalie Meurer Defines Forward Deployed Engineering as Customer Accountability · Sierra's Head of Agent Engineering argues forward deployed engineering is defined by customer accountability rather than specific technical skills.
• Ahmad Osman Explains Why Local AI Is Catching Up to Cloud Models · Ahmad Osman argues local AI is closing the gap with frontier models through efficiency and open-source infrastructure.
• Meta Brain2Qwerty v2 Launches Real-Time Decoding; Cursor Ships iOS Agents · Meta releases Brain2Qwerty v2 for real-time brain-to-text decoding while Cursor launches iOS agents.
• OpenAI Launches Restricted GPT-5.6 Family Amid U.S. Government Mandates · OpenAI launches GPT-5.6 Sol, Terra, and Luna models with access limited to trusted partners due to U.S. government requests.
• OpenAI Internal AI Usage Explodes: Research Tokens Up 56x Since Late 2025 · OpenAI reports median internal AI token usage surged 56x in Research and 32x in Support since November 2025.
• OpenAI Unveils Jalapeño Chip as AI Agents Battle for Organizational Control · OpenAI launches custom Jalapeño chip while Anthropic and others redefine AI agents as embedded organizational coworkers.
• Databricks Cofounders Advocate for Open Frontier Ecosystem in AI Data Strategy · Databricks cofounders promote open-source initiatives and data infrastructure to advance AI agent ecosystems.
• SpaceX's $28 Billion Annual Revenue from GPU Rentals · SpaceX generates $28 billion annually from GPU rental deals with companies like Reflection AI and Anthropic.
• Gray Swan Experts Warn AI Security Risks Are 'Gray Swan' Events Everyone Sees Coming · Zico Kolter and Matt Fredrikson explain why AI vulnerabilities are inevitable 'gray swan' events requiring new security mindsets.
• LS Paying Subscribers Get $250 Off AI Engineer Conference Tickets · LS offers $250 discount on AI Engineer conference tickets to paying subscribers.
• AI Engineering World Festival 2026 Tickets Face Limited-Time Discount and Rapid Sell-Out · Regular tickets for the AIE WF 2026 are expected to sell out by Monday, prompting a limited-time discount for subscribers.
• Zhipu's GLM-5.2 Emerges as Top Open-Weight Frontier AI Model · Zhipu's GLM-5.2 outperforms GPT-5.5 on benchmarks and is hailed as a viable open frontier model.
• Midjourney Unveils Full-Body Ultrasound Scanner and Spa Concept · Midjourney Medical launches a prototype full-body ultrasound scanner and plans a San Francisco spa facility.
• Radical AI Deploys Self-Driving Labs to Accelerate Materials Discovery · Radical AI uses automated self-driving labs to discover and characterize new alloys at unprecedented speeds.