ADVERTISEMENT
Full Research Archive
Complete chronological index of all 263 research papers, technical analyses, and industry postmortems.
Browse Archives by Category
All Articles (263)
AI Hardware & Semiconductors 100
AI Agents & Systems 33
Foundation Models 27
Model Evaluation 15
AI Industry 9
Edge AI & Electronics 7
AI Safety 7
AI Research 7
Open Models 6
AI Infrastructure 5
Industry Analysis 2
Datacenters & Infrastructure 2
Writing & Information 1
Structural Biology 1
Small & Edge Models 1
Semiconductor Markets 1
Security & Technology 1
Security 1
Scientific Revolution 1
Robotics & Embodied AI 1
Quantum Information 1
Planetary Science 1
Physics 1
Particle Physics 1
Origins & Astrobiology 1
Open-Weight Models 1
Neuroscience 1
Metrology 1
Mathematics & Computation 1
Materials Science 1
Labour & Technology 1
Interpretability 1
Institutions & State Capacity 1
Institutions & Economy 1
Inference Economics 1
Industrial Revolutions 1
Humanity & Technology 1
History of Technology 1
History of Computing 1
Governance & Policy 1
Evolution & Ecology 1
Ethics of Technology 1
Energy & Civilization 1
Earth & Climate 1
Demography & Urbanization 1
Data & Training 1
Cumulative Culture 1
Cosmology 1
Condensed Matter 1
Chemistry & Catalysis 1
Alignment & Safety 1
Agriculture & Cities 1
Agent Evaluation 1
AI Security 1
AI & The Web 1
AI & The Web (1)
AI Agents & Systems (33)
- The Hardest Unsolved Problems in AI Agent Architecture
- The Economics and Physical Limits of Running AI Agents at Scale
- Ten Failure Modes That Define Production AI Agent Architectures
- Measuring AI Agent Reliability: What the Evidence Actually Supports
- How an AI Agent Architecture Actually Works, Step by Step
- From Scripted Bots to Autonomous Agents: A History of AI Agent Architecture
- Comparing the Main Approaches to AI Agent Architecture
- Building an AI Agent Architecture in Practice: An Advanced Technical Guide
- AI Agent Architecture in 2035: Four Scenarios, Their Signals, and What Would Falsify Them
- Ten Failure Modes That Define Production Claude Code Use
- How Verification Actually Works in a Claude Code Workflow
- How Claude Code's Memory Actually Persists Across Sessions
- How Claude Code Actually Uses the Model Context Protocol
- Claude Code, From First Principles: The Agentic Loop, Permissions, and What Actually Executes
- Agentic Coding Tools in 2035: Four Scenarios, Their Signals, and What Would Falsify Them
- The Hidden Infrastructure Bill Behind Every RAG Answer
- Ten Failure Modes That Define Production RAG
- RAG in 2035: Four Scenarios, Their Signals, and What Would Falsify Them
- Measuring What a RAG System Retrieves, Not Just What It Answers
- What Actually Happens Between a Query and an Answer in RAG
- Naive, Graph, and Agentic: A Systems Comparison of RAG Architectures
- Building Production RAG: An Advanced Technical Guide
- From BM25 to Agentic Retrieval: A History of Retrieval-Augmented Generation
- Typed Tools: Protocols as an Agent's Action Alphabet
- Retrieval Is an Evidence System, Not a Memory
- Embeddings and the Geometry of Similarity
- Why Coding Agents Fail: Long-Horizon Reliability in OpenAI Codex
- OpenAI Codex as a Software-Engineering System: Models, Harnesses, and Evidence
- Engineering Reliable Codex Workflows with Instructions, Tests, Sandboxes, and Human Gates
- Codex vs Claude Code: A Systems Comparison, Not a Benchmark Beauty Contest
- Codex in 2035: Scenarios for Software Teams, Verification, and Machine-Written Code
- Codex CLI, IDE, Cloud, and Code Review: Where Each Execution Surface Fits
- Reliable AI Agents Are Control Systems, Not Chatbots
AI Hardware & Semiconductors (100)
- Why Your New Laptop Costs $200 More, and It Has Nothing to Do With Laptops
- In 1987, "We Will Only Manufacture Other People's Chip Designs" Sounded Like a Bad Business Plan
- Who Actually Makes the Machines That Make the Chips
- What an Export Control Actually Restricts, Mechanically
- Moore's Law Hasn't Ended in 2026. The Thing That Actually Ends It Isn't Quantum Tunneling Alone.
- What a Foundry Actually Is, and Why the Idea Was Once Considered Strange
- UMC Isn't Trying to Beat TSMC. It's Not Even Playing the Same Game.
- TSMC: The Foundry the Whole Industry Runs Through
- CoWoS, Not the Transistor: Why Packaging Capacity Caps the 2026 AI Accelerator Supply
- TSMC's $265 Billion Arizona Bet: What Reshoring the Frontier Node Actually Costs
- Tokyo Electron Owns the Step Between Every Other Step
- Three Futures for the Bifurcated Chip Industry: A Scenario Analysis
- Two Japanese Companies Make More Than Half the World's Silicon Wafers
- What Actually Triggered the July 2026 Chip Stock Selloff
- Two Different Inventions Get Credited as "the Integrated Circuit"
- The HBM Memory Supercycle: A Quantitative Look at Pricing, Yield, and the Consumer Spillover
- The AI Boom's Biggest Winners Might Be Turbine and Transformer Companies
- An AI Rack Needs 36 Times More Fiber Than an Ordinary Server Rack
- The Eight Men Who Quit and Built Silicon Valley
- How China Prints Near-Frontier Chips Without a Single EUV Machine
- The AI Boom's Real Bottleneck Is a Transformer — the Electrical Kind
- Mapping the Whole Insurgency: 43 Startups, $17.7 Billion, One Target
- The Line to Connect to the Grid Is Now Longer Than the Grid Itself
- The Trade Deal That Was Supposed to Save US Chips. It Didn't — But It Also Didn't Do What Anyone Expected.
- Texas Instruments Raised Analog Chip Prices Up to 85% in 2026. Almost Nobody Outside the Industry Noticed.
- Tenstorrent Is Betting AI Chips Don't Need Proprietary Instruction Sets
- AI Is Now Designing Meaningful Parts of the Chips That Will Run AI
- SUMCO Just Bet Its Future on AI Wafers, Not Ordinary Ones
- STMicroelectronics Doesn't Bet on One Market. In 2026, That Diversified Bet Is What's Actually Working.
- SOX, SMH, SOXX: A Reader's Guide to the Chip Indexes
- SMIC's Real Constraint Isn't the Node. It's the Yield.
- SK hynix: The Company Nvidia Cannot Ship Blackwell Without
- Not Every AI Chip Story Is About the Datacenter
- Samsung's Memory Business Is Raising Prices Faster Than Almost Anyone
- Samsung Foundry: Two Separate Races, Run by One Company
- SambaNova's Chips Rewire Themselves for Each Model
- Renesas Just Closed Its Last 6-Inch Fab. That One Sentence Explains an Entire Analog Supply Squeeze Hitting Automakers
- South Korea Has Its Own AI Chip Bet, and It Isn't SK hynix
- How to Read a Chipmaker's Backlog Like an Analyst, Not a Headline Writer
- Rapidus: Japan Is Trying to Skip Straight From No Advanced Fab to 2nm
- Rain AI Is Betting That Digital Chips Are the Wrong Idea Entirely
- Quantum Computing Has Its Own Chipmakers — and They Are Not Competing With Nvidia
- Qualcomm Wants Out of the Phone. The Datacenter Is the Test.
- Electricity Access Is Now a Semiconductor Industry Input
- Positron's Entire Bet: Inference-Only, No Exceptions
- Two Different Things Both Get Called "Photonic Computing." Only One of Them Does Math With Light.
- onsemi Bet on Growing Its Own Silicon Carbide Crystals in 2021. In 2026, That Decision Is the Whole Company's Advantage.
- NXP's Automotive Business Grew 17% in 2026. Analysts Still Call the Real Recovery a 2027 Story.
- Vera Rubin: Nvidia's Next Platform Is Six Chips, Not One
- Nvidia by the Numbers: What's Actually Behind a $4.9 Trillion Company
- Nvidia by the Multiples: A Valuation Profile, Not a Pitch
- Why Would Nvidia Buy the Chip Designed to Kill the GPU?
- Nvidia's Next Bottleneck Isn't the GPU. It's the Wire Between GPUs.
- No, Nuclear Reactors Are Not Powering Today's AI Datacenters
- Neuromorphic Chips Keep Almost Breaking Through. In 2026, "Almost" Is Still the Right Word.
- Moore's Law Was Never a Law. It Was a Business Plan.
- Micron Is the Only Volume American Memory Maker Left. That's the Whole Story.
- Marvell Designed Amazon's Trainium and Microsoft's Maia. Almost Nobody Notices.
- Liquid Cooling Just Stopped Being Optional
- Lightmatter Wants to Replace Copper Wires With Light Inside the Chip Too
- Lam Research's Whole Job Is Taking Atoms Away, Precisely
- KLA's Entire Business Is Finding the One Flaw That Ruins a Chip
- Is the Chip Rally a Bubble? The Numbers on Both Sides, Weighed Fairly
- Intel's 18A: A Great Process Node Still Waiting on Its First Big Outside Customer
- 14A Is the Node That Actually Decides Intel Foundry's Future
- Infineon Is the World's Largest Automotive Chipmaker, and Its Best 2026 Story Still Isn't About Cars
- Huawei's Next Ascend Chips Have Lower Peak Specs Than the One They Replace
- Japan Built the World's Dominant DRAM Industry. Then Lost It in About a Decade. Here's the Actual Mechanism.
- Groq, Quickly: The Inference Chip That Argued GPUs Are Wrong
- GlobalFoundries Bet Its Whole Business on Never Reaching the Leading Edge
- The Chips Making Your EV Charge Faster and the Grid Run Cleaner Aren't the Ones Anyone Writes About
- PwC Says $1T. IDC Says $1.75T. McKinsey Says $1.6T. TSMC Says $1.5T. Here's Why the 2030 Forecasts All Disagree.
- Etched Built a Chip That Can Only Run One Kind of AI Model
- China's AI Chips Have a Second Bottleneck Nobody Mentions: Memory
- "We Didn't Make Money for 10 Years, and We Never Gave Up on It" — Nvidia's CUDA Bet, Explained
- The Yield Math That Explains Why Chips Stopped Getting Monolithically Bigger
- Wafer-Scale Computing: Why Cerebras Refuses to Cut Its Chips Apart
- Cerebras Just Had the Biggest US Tech IPO Since Uber
- What Marvell Actually Bought for $5.5 Billion
- Why Synopsys Paid $35 Billion to Add Physics Simulation to Chip Design
- Broadcom vs. Marvell: The Rivalry That Actually Decides Who Builds Custom AI Silicon
- Broadcom Designs the Chips Behind Six Hyperscalers' Own AI Silicon
- Inside OpenAI's 10-Gigawatt Broadcom Deal: A Quantitative Look
- Ayar Labs Bet Its Whole Company on One Specific Connector
- Europe Barely Registers in the AI Chip Race. Axelera AI Is Trying to Change That.
- One Dutch Company Makes the Only Machine That Can Print a Modern Chip
- High-NA EUV: The Upgrade That Buys the Chip Industry Another Decade
- ASE Technology Finishes the Chip TSMC Started
- Applied Materials Puts the Atoms Down. Nobody Else's Job Is Quite This One.
- Apple Is One of the World's Largest Chip Designers. It Has Never Sold a Single Chip.
- Amkor Is Building America's First Complete Chip Supply Chain
- AMD's OpenAI Deal Is Measured in Gigawatts, Not Chips
- AMD's Untether Acquisition: A Small Deal That Explains a Large Strategy
- 18% of the S&P 500 Is Now One Industry
- 128 Weeks: The Part Actually Holding Up Your AI Datacenter
- Patterning at the Limit: What Actually Happens When a Chip Is Manufactured
- When the Substrate Became the Product
- The Network Is the Computer Again
- Locality Is the Whole Game: The Memory Hierarchy and What a Kernel Does Not Read
- What an AI Accelerator Actually Is: Silicon, Packaging, and the Memory It Can Reach
AI Industry (9)
- Ten Documented Incidents in xAI and Grok's Public Record
- How xAI's Infrastructure Build Compares to the Rest of the Field
- How Grok and Colossus Actually Work, as Far as the Record Shows
- Actually Building with the Grok API
- A Timeline of xAI, Grok, and Colossus
- Extrapolate Today's Trend Lines to 2030 and the AI Datacenter Looks Nothing Like Today's GPU Racks
- The AI-Chip Market Is About $80B Today. One Forecast Puts It Past $1.1T by 2035 — the Actual Argument for "Still Early."
- Eight Companies Have a Real Shot at Taking AI Compute Share From Nvidia. None of Them Do It the Same Way.
- The Economics of Codex: Tokens, Sandboxes, Review Time, and Software Throughput
AI Infrastructure (5)
AI Research (7)
- Ten Ways Mechanistic Interpretability Research Can Mislead You
- Running an Interpretability Investigation That Holds Up
- Probing, Sparse Autoencoders, Patching, and Steering: The Main Interpretability Methods, Compared
- Mechanistic Interpretability in 2035: Scenarios and Falsifiers
- How Mechanistic Interpretability Research Is Actually Done
- How Do We Know an Interpretability Claim Is Actually Right?
- A History of Mechanistic Interpretability
AI Safety (7)
- The Hardest Unsolved Problems in AI Alignment
- Ten Failure Modes That Define Deployed AI Safety Systems
- Outer Alignment, Inner Alignment, and Why the Difference Matters
- How Do We Actually Know a Safety Measure Worked?
- Building a Deployment Safety System That Actually Holds
- AI Alignment and Safety in 2035: Two Axes, Four Scenarios, and What Would Falsify Them
- A History of AI Alignment as a Research Field
AI Security (1)
Agent Evaluation (1)
Agriculture & Cities (1)
Alignment & Safety (1)
Chemistry & Catalysis (1)
Condensed Matter (1)
Cumulative Culture (1)
Data & Training (1)
Datacenters & Infrastructure (2)
Demography & Urbanization (1)
Earth & Climate (1)
Edge AI & Electronics (7)
- Ten Failure Modes That Define Small and On-Device AI Deployments
- Small and On-Device AI in 2035: Scenarios and Falsifiers
- Shrink It, Train It Small, or Search for It: The Main Strategies for Small Models, Compared
- Shipping an On-Device AI Feature That Actually Works in the Field
- How Do We Know a Compressed Model Still Works?
- How a Model Actually Gets Small Enough to Run on a Phone
- A History of Small and On-Device AI
Energy & Civilization (1)
Ethics of Technology (1)
Evolution & Ecology (1)
Foundation Models (27)
- Ten Failure Modes That Define Multimodal AI Systems
- Multimodal AI in 2035: Scenarios and Falsifiers
- How Multimodal Models Actually Handle Video, Audio, and Space
- Building a Multimodal AI Application That Actually Uses Its Inputs
- Adapters, Native Pretraining, and Unified Tokens: The Main Multimodal Architectures, Compared
- A History of Multimodal AI
- What "Enterprise-Ready" Actually Means for a Claude Deployment
- What Claude's Capability Evaluations Actually Measure
- What Actually Happens Inside a Very Long Claude Context Window
- The Safeguards Behind a Claude Refusal, and What They're Actually Trading Off
- The Real Cost of Running Claude in Production, Beyond the Per-Token Price
- Ten Failure Modes That Define Production Claude Deployments
- How Constitutional AI Actually Constrains a Model's Behavior
- Claude, From First Principles: Training, Constitutional Methods, and What Actually Shapes a Response
- Anthropic in 2035: Four Scenarios, Their Signals, and What Would Falsify Them
- A History of Anthropic and the Claude Model Line
- One Model, Many Modalities: What Multimodal Systems Actually Share
- Model Systems in 2035: Four Scenarios, Their Signals, and What Would Falsify Them
- From n-Grams to Reasoning Models: A Technical History of the Language Model
- The Physical Economics of Inference: Bandwidth, Energy, and the Limits of a Datacenter
- Building on Frontier Model APIs: Effort, Context, Caching, and Failure Handling
- Ten Failure Modes of Deployed Language Models, from Silent Truncation to Version Drift
- Dense, Sparse, and Distilled: Comparing Approaches to Frontier Model Capacity
- Serving a Frontier Model: The KV Cache, Batching, and What a Token Actually Costs
- What We Still Cannot Do: Open Problems in Frontier Model Systems
- Measuring Frontier Models: Contamination, Variance, and What a Score Can Support
- OpenAI Model Systems from First Principles: Weights, Post-Training, and Inference Compute
Governance & Policy (1)
History of Computing (1)
History of Technology (1)
Humanity & Technology (1)
Industrial Revolutions (1)
Industry Analysis (2)
Inference Economics (1)
Institutions & Economy (1)
Institutions & State Capacity (1)
Interpretability (1)
Labour & Technology (1)
Materials Science (1)
Mathematics & Computation (1)
Metrology (1)
Model Evaluation (15)
- Why Cost and Latency Belong in the Evaluation Score, Not a Footnote
- Ten Ways an Agent Evaluation Can Mislead You Even When It's Working Correctly
- How Benchmark Contamination Actually Works in Agentic Evaluation
- Building a Custom Evaluation Suite for a Production Agent
- Agent Evaluation in 2035: Two Axes, Four Scenarios, and What Would Falsify Them
- A Practitioner's Map of Agent Evaluation Frameworks
- A History of How We Learned to Evaluate AI Agents
- The Frontier Model Landscape in 2035: Four Scenarios, Their Signals, and What Would Falsify Them
- Ten Failure Modes That Show Up Across Every Frontier Model, Not Just One
- OpenAI and Claude on Formal Reasoning: What the Benchmarks Show, and Where They Mislead
- OpenAI and Claude on Agentic Coding: What the Independent Evidence Actually Shows
- How to Actually Compare Frontier AI Models Without Building a Misleading Leaderboard
- Context Window Size Versus What a Frontier Model Can Actually Recall From It
- Comparing Frontier Model Pricing Without Comparing Apples to Oranges
- How to Evaluate Codex Beyond SWE-bench and Vendor Scores
Neuroscience (1)
Open Models (6)
- Ten Failure Modes That Define the Open-Weight AI Ecosystem
- Open-Weight AI in 2035: Scenarios and Falsifiers
- Llama, Mistral, DeepSeek, Qwen, Gemma, and OLMo: What "Open" Actually Means for Each
- How Llama's Architecture Actually Works, Generation by Generation
- Actually Deploying an Open-Weight Model in Production
- A History of Llama and the Open-Weight AI Movement