An AI can become faster without changing the model or buying new hardware.

That's what DeepSeek just proved. Their team found a way to make their models run 85% faster, just by improving how the calculations are processed. No new model, no new chip. Just a better method.
And they made the solution public and free for anyone to use.
This connects to what I care about in my own work: making AI lighter and faster without losing quality. On my project BitFace, I used a similar approach for face recognition — shrinking the model down to its smallest possible size so it could run on small devices.
We often hear about "bigger models" and "more computing power" in AI. But sometimes, the real breakthrough is simply doing more with what you already have.