TheSequence

21 articles in the TruthFoundry index. Each links out to the original.

• Stripe Acquires OpenRouter to Control AI Token Economy · Stripe agreed to acquire OpenRouter to integrate model routing into its payments infrastructure.
• AI Scaling Laws Extend to Energy and Power Infrastructure · The next AI scaling law involves the efficiency of converting physical energy into intelligence.
• DeepSeek, Z.ai, NVIDIA Unveil New AI Models and Tools · Four major AI releases include DeepSeek V4-Pro, GLM-5.3, NVIDIA Nemotron 3.5 Lightning, and NeMo Switchyard.
• Distilling Test-Time Compute into Model Weights · Researchers propose compressing the intelligence gained from expensive test-time reasoning back into model weights.
• SpaceX Acquires Cursor; Anthropic Eyes Decart Deal Amid AI Model Shifts · SpaceX finalized its $60 billion acquisition of Cursor while Anthropic explores a $6 billion deal for Decart AI.
• AI Inference Systems Function as Complex Token Factories · Modern AI inference systems manage asynchronous traffic, memory, and GPU scheduling to serve diverse user requests efficiently.
• Meta, Prime Intellect, and OpenAI Release New AI Coding and Math Models · Three major AI releases include Meta's coding agent, Prime Intellect's harness, and OpenAI's Astra mathematical results.
• NVIDIA's Chris Alexiuk Explains Nemotron 3, Agentic AI, and Open Data Strategy · NVIDIA's Chris Alexiuk details the Nemotron 3 model family, its hardware-aligned tiers, and the role of open data in AI safety.
• Distilling Diffusion and Multimodal Models via Sequence Knowledge · New research explores distilling complex diffusion and multimodal model trajectories into smaller models.
• Google Restructures AI Division as Meta Deploys Coding Swarm · Google reorganizes DeepMind leadership while Meta launches autonomous coding agents.
• AI Agents Create New Elastic Workforce Measured in Tokens · AI agents allow parallel engineering tasks, creating a token-based workforce alongside human headcount.
• Gemini Robotics 2 Unifies Walking and Grasping in Single Policy · Gemini Robotics 2 integrates locomotion and manipulation into a single language-conditioned policy.
• Cross-Architecture Distillation Moves Transformer Knowledge to Non-Transformer Models · Researchers explore distilling transformer capabilities into fundamentally different neural architectures like state-space models.
• AI Race Shifts to Robotics, Open Models, and Economic Reality · The AI industry pivots from model demos to robotics, open weights, and the challenge of converting capital into revenue.
• The Sequence Robotics #905: Who Builds the Robot Brain? · The Sequence Robotics newsletter analyzes the divide between digital AI models and physical robotics startups.
• AI Shifts from Research Era to Industrial Engineering · Ilya Sutskever argues AI has moved from a research phase to an era of industrial-scale engineering.
• Laguna S 2.1 Outperforms 1.6T DeepSeek-V4-Pro-Max on Benchmarks · 118B parameter Laguna S 2.1 model scores higher than larger competitors on Terminal-Bench 2.1 and DeepSWE.
• Large Language Models Generate Synthetic Data to Train Smaller Models · Large language models can manufacture training datasets to teach smaller models, shifting data from static geology to dynamic generation.
• Anthropic's Opus 5 and Atoms' $1.7B Raise Signal AI's Physical Expansion · Anthropic releases Opus 5 for complex workflows while Atoms raises $1.7B for physical AI robotics.
• Google is NVIDIA's Closest Full-Stack AI Rival, Though Not a Universal Replacement · Google mirrors NVIDIA's full-stack AI control but faces challenges from AWS and AMD regarding universal replacement.
• Sequence AI's Inkling Model Uses Sparse Routing for Trillion-Parameter Efficiency · Sequence AI's Inkling model activates only 41 billion parameters per token from a 975-billion parameter total.