Article archive
Published news and blog articles, organized by category. Browse older coverage by month or search for a topic. Undated blog guides appear after dated news.
9,480 articles · Newest first
The KV Cache: The Memory Trick That Makes AI Feel Instant
The key-value cache lets transformers skip recomputing attention for every new token, trading GPU memory for order-of-magnitude gains in inference speed.
One Chip, Four Trillion Transistors: How Cerebras Rethinks the AI Accelerator
The Cerebras Wafer Scale Engine 3 puts 4 trillion transistors, 900,000 AI cores and 44GB of on-chip SRAM on a single wafer-sized chip, attacking the memory bottleneck that dominates AI inference.
AMD MI400 and the Real Bar for Beating Nvidia in AI Accelerators
Raw specs do not win data centers. CUDA's software gravity, rack-scale systems, and the economics of switching explain why displacing Nvidia is harder than any benchmark.
Blind Camera Tests: The Science of Judging Smartphone Photos
When reviewers hide the phone names, the rankings flip. What blind testing reveals about perception, bias, and what actually makes a photo look good.
Choosing an LLM: A Developer's Guide to Model Selection in 2026
Context windows, cost per million tokens, latency, reasoning depth, and open weights — how engineers should actually pick a large language model when every vendor claims state of the art.
OWASP Top 10 for Agentic AI: What Breaks When Software Acts
Autonomous agents browse, spend, and execute with credentials. The OWASP Agentic App Top 10 catalogs how that goes wrong — from memory poisoning to confused deputies.
GPT-6 Astra: OpenAI''s new flagship lands as a limited preview — here''s what that actually means
OpenAI's GPT-6 Astra flagship arrived on September 3, 2026 as a limited preview for trusted partners. N43 explains what a limited preview really means, where Astra sits in the GPT line, and how the week's model releases stack up.
Pixel 11 Pro vs Galaxy S26 Ultra: why reviewers are switching phones in 2026
A reviewer's switch from the Galaxy S26 Ultra to the Pixel 11 Pro signals a broader 2026 shift: camera processing philosophy, on-device assistant integration, and switching costs now drive flagship decisions. N43 and Hermes analyze the…
Skills, MCP, RAG, memory: the four-layer stack that makes AI agents actually useful
Skills, MCP, RAG, and memory form the four-layer stack that turns a language model into a useful AI agent. N43 explains what each layer does and how the Model Context Protocol standardized the connections.
Top 5 smartphones of 2026, so far: what reviewers actually rate, and why
SuperSaf's mid-2026 ranking of the best smartphones so far reveals what reviewers actually reward: silicon, camera consistency, battery endurance, on-device AI, and honest value framing. N43 and Hermes break down each factor.
Why China Is Betting on Analog Chips for the Next AI Generation
A Chinese team reportedly demonstrated an analog AI chip that could run certain workloads up to 1,000 times faster than a flagship GPU — and at a fraction of the power. The claim matters less for its precision than for its direction.
The Best Phones of 2026: A Mid-Year Scorecard
Forty-plus reviews in, the 2026 phone market has a clear shape: camera systems converging, silicon diverging, and the value tier carrying the interesting bets.
Claude Opus 4.5 and the Agentic Coding Frontier
Anthropic's Opus 4.5 sharpened long-horizon coding work: fewer handoffs, better tool use, and a benchmark race that is quietly redefining what a coding model owes its users.
iPhone Ultra: Apple's 2026 Flagship Gamble
Rumors point to a titanium, camera-first iPhone Ultra for 2026 — plus the long-awaited foldable. What the leaks actually say, and what they leave out.
MediaTek Dimensity 9500: The New Contender for Flagship Silicon
MediaTek's Dimensity 9500 takes aim at Qualcomm's flagship throne. N43 separates vendor claims from verifiable facts on architecture, AI acceleration and market share.
RAG vs Fine-Tuning vs Prompt Engineering: How to Actually Adapt an LLM
The three main ways to adapt an LLM to your data - retrieval-augmented generation, fine-tuning, and prompt engineering - and a practical decision framework covering cost, latency, freshness, and control, plus production failure modes and…
What Happens If AI Just Keeps Getting Smarter?
The intelligence-escalation debate in measurable terms: benchmark and compute trends on the path to AI systems smarter than humans, and why forecasts, alignment research, and governance all carry enormous error bars.
Why Big Tech Is Betting on Nuclear Power for AI Data Centers
Microsoft, Amazon and Google are signing nuclear power deals to feed AI data centers. N43 examines the deals, the physics, the economics and the open questions.
Direct-to-Device Satellite Messaging: How Phones Reach Beyond the Grid
Ordinary phones can now exchange messages with low-Earth-orbit satellites using shared cellular spectrum, no satellite phone required. How direct-to-device works, who is building it, and what physics still limits.
In-Context Learning: How LLMs Learn From Examples in the Prompt
Large language models can pick up a new task from a handful of examples placed in the prompt, with no retraining and no new weights. This is in-context learning: how it works, what induction heads reveal about it, and where the…
OpenAI's Broadcom Chip: The Full-Stack Bet That Reshapes AI Silicon
OpenAI has unveiled its first custom AI chip, co-designed with Broadcom. Why a model lab builds its own silicon, the inference economics driving it, the Google TPU and Apple precedents, what it means for Nvidia, and the tape-out risks that…
Pixel 11 Pro vs Galaxy S26 Ultra vs iPhone 17: The 2026 Flagship Camera War
The Pixel 11 Pro, Galaxy S26 Ultra, and iPhone 17 Pro Max converge on excellent photos in daylight, so the 2026 camera war is decided by software: Night Sight, Magic Capture, Instant Night Sight, Pro Stable Video, and the computational…
Agentic AI Explained: How Autonomous Agents Actually Work — and What Comes After Chatbots
Agentic AI is the shift from systems that answer to systems that act. We break down the four defining properties of software agents, the loop that makes autonomy possible, and what changes when chatbots stop being the end state.
GPT-6 Astra Goes Public: What OpenAI's Biggest Model Release Means for AI in 2026
OpenAI staged GPT-6 Astra in two waves: a trusted-partner preview on September 3, 2026 and a public release on September 4. We separate the documented record from the September 2026 discourse and place Astra in the arc of the GPT series.
Also explore blog articles and guides or search all DutyStation pages.