Sitemap
104 articlesA complete index of every article on this site.
AI
- Teaching AI to Reason Across Sound and Sight: Inside the OmniVideo-100K Dataset
- Diffusion Language Models Are Sold as 'Parallel.' This One Isn't — and Isn't Sequential Either
- A Famous Toy Model of Neural 'Superposition' May Not Show Superposition at All
- A Proof Checker Can Tell You What's Correct — Not What's Worth Proving
- Persona-Pruner: Carving a Single Character Out of a Big Language Model
- A Vision-Language Model Has 'Eyes' Inside It — and You Can Move Where They Look
- What 'Training' and 'Inference' Actually Mean — Read in NVIDIA's Own 10-K
- NVIDIA Has Described Its Chips the Same Way for Years — Here Are Three Filings
- By the Numbers: NVIDIA's Revenue Tripled, and So Did Its R&D Bill
- What a 'Data Center GPU' Is — Explained Through AMD's Instinct Disclosures
- Filing of the Week: The Boring Accounting Line That Reveals Alphabet's AI Buildout
- The Capex Question: Why Server Depreciation Is the Number Behind the AI Spend
- Hype Check: 'AI in Everything' vs. What Meta Actually Tells the SEC
- Glossary in Context: What 'Accelerated Computing' Means, per NVIDIA
- How Microsoft's 'Intelligent Cloud' Became an AI Story — in Its Own Filings
- Why AI Keeps Driving Companies' 'Computing Needs' Up — Meta Said It Plainly
- Filing of the Week: AMD's 10-K Quietly Makes the Case for a Second AI Chip
- Hype Check: NVIDIA's 10-K Describes the Workload, Never Promises the Demand
- Why AI Models Need 'Data Augmentation' — Read Through an NVIDIA Patent
- Patent of the Week: NVIDIA's Grant on Synthesizing Video From Audio
- Hype Check: Is 'Agentic AI' a New Capability or a New Word?
- Patent of the Week: Anthropic's Multimodal Agent That Drives a Computer Screen
- Why Everyone's Suddenly Talking About Inference Cost
- The Capex Question: Reading the AI Buildout Through Microsoft, Alphabet, and NVIDIA's Filings
- Filing of the Week: What NVIDIA's 2026 10-K Says Its Chips Are Actually For
- Three Ways to Tell the NVIDIA Data Center Story: A Workload, a 10-K, and a Patent
- Patent of the Week: Microsoft's Grant on Routing Models by Perplexity (G06N)
- Hype Check: When Microsoft Says 'Cloud and AI Infrastructure,' What Is It Actually Counting?
- Three Ways to Tell Google's AI-Efficiency Story: A 2020 Patent, a 2026 Patent, and a Capex Line
- What a Mixture-of-Experts Model Actually Does — Read Through Google's Own Patent
- Training vs. Inference, Explained Through NVIDIA's Own Workload List
- How Vision-Language Models Learn to Connect Pictures and Words
- RLHF in Plain English, Traced Through a Google Patent on Search-Feedback Fine-Tuning
- What 'Neural Architecture Search' Actually Does — Read Through a 2020 Patent
- Three Ways to Tell the 'AI-for-Science' Story: a Grant, a Platform, and a CPC Class
- Patent of the Week: The 2020 Publication Behind 'Smart' Document Linking
- Why Training Data Gets 'Transformed' Before a Model Sees It — a 2020 Patent
- Patent of the Week: How a Neural Network Reads DNA — a 2020 Illumina Publication
- How Speech Recognition 'Reconciles' What It Hears — a 2020 Grant
- Hype Check: When 'AI Platform' Patents Tag Every Class in Sight
- How AI Designs Itself Without Sharing Your Data — a 2021 Patent
- Patent of the Week: NVIDIA on Making Fake Data to Train Real Models (2021)
- What 'Quantizing' a Neural Network Means — a 2021 Patent in Plain English
- Why AI Chips Are Built Around 'Dataflow' — a 2021 Accelerator Patent
- Three Ways to Tell the 'Scale-Out' Story: a Patent, a Protocol, and a Bottleneck
- Patent of the Week: Watermarking an AI Model — and Why It's Harder Than It Sounds (2021)
- What 'Accelerated Training' Actually Means for a Neural Network — a 2021 Patent
- Why 'Batching' Decides What You Pay to Run a Transformer — a 2022 Grant
- Patent of the Week: NVIDIA on Swapping Backgrounds to Make Models Robust (2022)
- How an AI Learns to Move — DeepMind's 2022 Grant on Motor Primitives
- Three Ways to Tell the 'Old ML Meets New Hardware' Story — a 2022 Microsoft Patent
- What 'Optimizing' a Neural Network Means at the Hardware Level — a 2022 NVIDIA Patent
- Patent of the Week: A 2022 Hardware Circuit Just for Scheduling AI Work
- Hype Check: Does 'Explainable' AI in a 2022 Grant Mean You'll Understand It?
- Patent of the Week: Salesforce's 2023 Grant on a Model That Sees and Talks
- How AI Generates an Image in Stages — a 2023 Google Patent
- Patent of the Week: DeepMind's 2023 Publication on Writing Code From a Task
- Three Ways to Tell the 'Data Wall' Story — a 2023 Microsoft Patent
- Patent of the Week: Adobe's 2023 Grant on a Language Model That Designs Documents
- What an 'In-Memory Compute Chiplet' Is — a 2023 Generative-AI Accelerator Patent
- Hype Check: When Every 2023 Patent Suddenly Says 'Generative AI'
- Patent of the Week: DeepMind's 2024 Publication on 'Look-Ahead' Sequence Models
- Why Attention Has to Be 'Hierarchical' to Scale — a 2024 NVIDIA Patent
- Patent of the Week: Amazon's 2024 Grant on AI Chips That Resize Themselves
- How a Diffusion Model Runs on Your Phone — a 2024 Google Patent
- Three Ways to Tell the 'Capacity Crunch' Story — a 2024 Microsoft Patent
- What 'Tuning a Model With Latent Variables' Means — a 2024 Google Patent
- Patent of the Week: NVIDIA's 2025 Publication on Scheduling Vision-Language Inference
- How AI Models Police Each Other — NVIDIA's 2025 Safeguard Patent
- Patent of the Week: Google's 2025 Grant on Language Models That Check Your Infrastructure Code
- What 'Model Deployment in Containers' Means — a 2025 NVIDIA Patent
- How AI Now Helps Compile the Code That Runs AI — an AMD 2025 Patent
- Patent of the Week: NVIDIA's 2025 Take on Letting a Language Model Fix Your Transcripts
- Filing of the Week: How NVIDIA's New 10-K Defines the AI Build-Out It's Betting On
- The Capex Question: Microsoft's 2020 10-K and the Cost of ‘Ubiquitous’ AI
- Filing of the Week: NVIDIA's 2021 10-K and a 53% Jump the Document Quietly Confirms
- The Capex Question: Microsoft's 2021 Capex Climbs to $20.6 Billion as AI ‘Advances’
- Filing of the Week: Alphabet's 10-K Says AI Is Now Behind ‘Many of Our Latest Innovations’
- What a GPU Actually Is for AI — Read Through AMD's Own 2021 10-K
- What a ‘Transformer Engine’ Is — Read in NVIDIA's 2023 10-K, the Year H100 Shipped
- Filing of the Week: Meta's 2022 10-K Names ‘Generative AI’ as a Product Tool
- Filing of the Week: NVIDIA's Revenue More Than Doubled to $60.9B — The 10-K Confirms It
- Hype Check: Meta's 2023 10-K Treats Generative AI as Both an Engine and a Risk
- Filing of the Week: NVIDIA Hits $130.5B and Launches Blackwell — By the Annual Report
- What ‘Multimodal’ Means — Read Through Alphabet's 2024 10-K and Gemini 2.0
Artificial Intelligence
- How NVIDIA's Invertible Neural Network Turns a Spectrogram Back Into Speech
- A Neural Network That Trains Itself in the Circuit: Reading an Analog-Hardware Filing From This Week's Pub Drop
- What the NIST AI Risk Management Framework Is — and What Its Four Functions Ask You to Do
- What Patent Class Are AI Inventions Filed Under? The Short Answer Is G06N
- What 'Data Center Capex' Actually Means — and Why It Is the Number the AI Story Now Turns On
- What HBM Is — and Why High-Bandwidth Memory, Not Raw Compute, Often Bottlenecks AI
- Training vs. Inference Compute: The Two AI Costs That Behave Nothing Alike
- What RLHF Is — How Human Feedback Turns a Raw Language Model Into One That Follows Instructions
- What a Transformer Model Is — and Why 'Attention' Replaced the Recurrence It Was Built On
- What a Mixture-of-Experts Model Actually Does — and Why It Lets a Model Grow Without Growing the Bill
- ScholarQuest Benchmarks LLM Search Agents on Academic Paper Retrieval, Reporting a Best Recall@100 of 0.314
- AutoPass Puts an LLM Inside the Compiler to Tune LLVM, Reporting 1.04x and 1.12x Geometric-Mean Speedups Over -O3
- Before It Writes a Rule, the GSA Is Asking the Obvious Question: What Happens to Federal Data Inside an LLM?
- To Catch a Multi-Stage Attack, Researchers Built the Dataset Nobody Had — and Fine-Tuned Small Models on It
- Hype Check: 'Distilled' Datasets Were Supposed to Beat Just Picking Good Examples. A New Study Says They Don't.
- The Real Bottleneck in Health AI Isn't the Model — It's Grading the Model
- A Transformer That Keeps Thinking Until It Stops Changing Its Mind
- How a 'Visual Verifier' Lets a Robot Grade Its Own Moves Before It Makes Them