Latent Space

Arca's AI head describes multi-day agentic workflows for financial work using persistent memory and custom infrastructure harnesses.

Arca's head of AI discusses agent architecture for long-horizon financial tasks in wealth management, including custom harness and persistence layers.

Google reports AI Mode in Search caused 15% revenue decline (~$500B impact), worst drop since 2004 IPO, raising questions about AI monetization strategy.

Latent Space covers AIE CODE conference, Mistral open-weight release, frontier monitoring developments, and social media AI news roundup.

Elorian's Glimpse model tackles visual reasoning gap via new decoy-counting benchmark testing frontier models' ability to filter distractors.

Google releases Gemini 4 Carbon, a frontier model leading independent benchmarks, initially available to cybersecurity researchers and defenders.

Supabase founders discuss why their PostgreSQL backend is becoming infrastructure choice for AI agent applications.

Anthropic eliminates coding interviews from hiring, citing AI capabilities in code generation as rationale for changed assessment approach.

Anthropic engineers discuss Claude Opus 5.5 architecture, separating model inference from agent execution via the Claude Platform.

Anthropic leaders detail Claude Opus 5.5 launch, API scaling, and managed agent architecture separating model brain from tool execution.

Tutorial on running offline coding agents with OpenCode, Ollama, and Qwen for remote/bandwidth-constrained environments.

Anthropic files for $2T IPO with 1400% YoY growth, 85% operating margins, $1B Q2 profit, and $134B cash reserves.

Supabase plans backend infrastructure changes to support AI agents building apps, including long-lived workloads and agent-friendly schema design.

Kimi 3.1 Pro 104BA-2.8T open-weights model claims parity with Fable-5.1; includes 2-3× faster inference API.

Analysis of emerging patterns in AI mathematical reasoning capabilities and predictability of model performance on unseen problems.

Latent Space reports on ongoing cybersecurity incidents affecting AI sector amid October threat activity.

Modal CEO Bernhardsson analyzes why AI compute infrastructure defaults to rectangular resource allocation, examining elastic GPUs and sandbox economics for agent deployment.

Moshi integrates market data into trader workflows via AI; Anekha Sokhal discusses product-workflow fit over raw capability.

OpenAI's Patch the Planet program discovers PHP2Shell vulnerabilities in core PHP using GPT-5.6 Sol; 2-month coordinated disclosure affecting 70% of web infrastructure.

Claude Mods reimagines agent interface design by shifting control from machine-facing harness to human-facing Attention Interface paradigm.

DeepSeek releases DeepInfer, open-source inference stack processing 155T tokens/day at ~80% margin, enabling $22M/day revenue with 81-day payback period.

Tenex discusses outcome-based pricing and agent-driven SDLC tooling for enterprise AI adoption at Fortune 2000 scale.

SF Compute's Evan Conrad outlines programmable data centers with GPU markets, physical settlement, and decentralized compute trading infrastructure.

DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash, and Mercury 3 released; limited immediate adoption expected.

Latent Space launches new podcast series; no technical content or AI developments announced.

Nick Cammarata claims agent research disproves polysemanticity and advances interpretability and alignment.

Modal raises $16B Series D at $1B ARR valuation, achieving decacorn status with 3x usage growth in 5 months.

OpenAI and Anthropic discuss autonomy-safety tradeoffs for financial agents at AI Engineer New York, balancing utility with control.

Leash Bio uses DEL assay data to validate AI model predictions in drug discovery and train chemistry agents for autonomous reasoning.

Gemini 4 Argon reaches GA; benchmark results (CursorBench) confirm previously disclosed performance claims.

Dyna Robotics unveils Dyna-2, a physical agent with hour-long autonomy using World Action Models for robotic control.

MSL’s emergence as a frontier lab is finally complete.

Anthropic presents dynamic workflows for coordinating Claude agents, applying Unix/Linux design principles to multi-agent orchestration.

Latent Space announces NYC subscriber meetups for tomorrow and Tuesday.

Standard Bots uses pretrained models with demonstration learning and deployment feedback loops to build reliable factory robotics systems.

Startup claims $100M+ ARR and $7.5B valuation 3 weeks post-launch; lacks technical details or verification.

Kohli (DeepMind) & Candido (Biohub) discuss limitations of AlphaFold and scaling approach to protein folding beyond structure prediction.

Latent Space reports minimal AI industry activity on this date.

Periodic Labs' Fedus and Cubuk discuss synthesis superintelligence across semiconductors and superconductors on Latent Space podcast.

yay small models

Kubernetes co-creators McLuckie and Beda develop cloud-native agent harness to scale autonomous systems beyond desktop environments.

Our head hurts.

Reflection Beam, a 501B open-weights US model, launches as domestic alternative to frontier proprietary systems.

Minimal substantive AI news reported; no specific announcements, models, or developments disclosed.

Ahmad Al-Dahle, ex-Meta Llama lead, applies AI to Airbnb's product development and guest experience.

Latent Space reports Pi 1.0 and Pi Durable releases with TypeScript support; coverage of AIE NYC event.

Interview with MIT PhD Alex Zhang on RLM research, PhD trajectories, and future research directions in AI.

Google announces Gemini 4 Argon with 1M context window, positioning against OpenAI Astra/Fable; limited access via Fairwind Program for government and cybersecurity professionals.

OpenAI's computer use API shipped in one week; podcast interviews CUA and platform leads on DevDay announcements and capabilities.

OpenAI DevDay 2026 announces GPT-6.1 Sol, Decisions API, Agents API, Spaces marketplace, and reaches 1.2B ChatGPT weekly active users.