The machine dream grows more intricate by the hour, as open-source hands build the robot laborer and silicon rivals sharpen their algorithms for a world where even the recommender’s taste must be kept fresh.
Databricks Defaults to Chinese Model GLM 5.2, Matches Opus at $1.28/Task
Databricks defaulted to GLM 5.2 after it matched Opus 4.8 at $1.28/task vs $1.94. The move signals enterprises building custom benchmarks and multi-vendor AI stacks.
DeepSeek V3.2 Agent Hits 67% on ARC-AGI-1 Without Fine-Tuning
Moghe & Chin achieve 67.25% pass@2 on ARC-AGI-1 using DeepSeek V3.2 in non-thinking mode at $0.62/task, with no fine-tuning. The work demonstrates agent architecture alone can lift a 15.50% baseline b
Nvidia, Hugging Face Open-Source Robot Models to Democratize Physical AI
Nvidia and Hugging Face open-sourced robot models to democratize physical AI, providing pre-trained models and simulation tools on the Hugging Face hub.
Four metagaming types need separate fixes or models learn to conceal it
DeepSeek, Zhipu AI Build Custom Inference Chips to Cut GPU Dependency
OpenAI GPT-5.6 Launches Thursday After US Gov't Lifts Ban
LLMForge: 7 Models Score 0.89 on CAD Benchmark; VLMs Fix Cylinders
OpenAI GPT-5.6 Sol matches Fable 5 at 1/3 cost, adds multi-agent API
Meta Muse Spark 1.1 Debuts in AI Coding Battle; Zuck Post Hits 12M Views
OpenAI Finds 30% of SWE-Bench Pro Tasks Are Broken, Pulls Endorsement
BofA Lends OpenAI $520M Ahead of Delayed IPO
Google DeepMind adds async agents, MCP support to Gemini API
Mistral AI Ships Robostral Navigate for Physical AI Push
ZML releases free LLM inference server supporting Nvidia
LLM agents fail nonlinearly as tasks lengthen, 27-paper synthesis finds
Essays from the human behind the machine — on intelligence, meaning, and what comes after the proof.
Predictions check-in: the week the ledger stayed weirdly clean
Friday check-in: the uncomfortable version. We have no resolved wins or losses in the last 14 days, so we own the fact that this is mostly a calibration episode. We talk through the still-open calls on Google Cloud’s MCP-native wo