The next big thing in AI might be thinking smaller
A new generation of compact models is changing what we expect from on-device intelligence.
A new generation of compact models is changing what we expect from on-device intelligence.
A benchmark is a proxy. It stands in for a capability that is hard to measure directly, and like all proxies it can drift away from the thing it was meant to represent once people start optimising for the number.
For a generation of software companies, marginal cost was effectively zero. Products built on large models do not have that property: every request has a price, and that price shows up in the gross margin.
Building generation is now comparatively fast. Connecting it is not. In this sample analysis, interconnection queues rather than construction timelines are the binding constraint on new capacity.
Keep the useful automation. Lose the constant context switching.
Illustrative media-industry reporting for the demo.
The case for focused assistants that fit the workflow you already have.
A practical look at notifications, defaults, and the tools that earn our attention.
Sample research coverage on the gap between lab and production.
Demo science reporting with an emphasis on uncertainty.
Every article carries a three-line brief, extracted from its own sentences. Nothing is rewritten, so nothing can be invented.
Articles play through the browser's own speech engine, highlighting each sentence as it is spoken. No audio files to host.
Stories are matched on shared vocabulary rather than on category, and each card shows the words the two have in common.
The feed reorders itself around what you actually open. The history stays in your browser and never leaves the device.