Products & Models
10 articlesGrok, ChatGPT, Claude, Gemini: A Field Guide to the Frontier
Marketing pages all claim the same thing. Refusal rates, long-context behaviour, tool reliability and data terms are where the real differences show up.
Reasoning Models, Explained Without the Marketing
Spending more computation at answer time buys accuracy on maths and code, and costs you tokens. The open question is whether the visible trace is the work.
Why Every Assistant Suddenly Wants to Remember You
Persistence, not capability, is the consequential change in consumer AI. Memory cuts cost and raises stickiness, and every failure mode it has is invisible.
Multimodal AI: When Your Model Can Watch, Listen and Draw
Screenshots, audio and scanned invoices all arrive as inputs now. Each new sense adds a failure mode, and one of them is a route for instructions.
Open Weights, Closed Weights: What Developers Actually Choose
The ideological argument gets the headlines. Licence terms, serving costs and evaluation budgets decide what actually ships, and the answer is often both.
Enterprise AI Pilots: The Projects That Never Shipped
Executives approved budgets, teams built prototypes, then nothing shipped. The reasons repeat across industries, and the fix is process and budget ownership.
AI in the Clinic: The Slow Path to Adoption
Imaging and ambient scribing already run in clinics. Clinical decision-making does not, because a model used in clinical care is a medical device.