Скачать или смотреть Turning Text Into Images: The Clever Trick Saving AI Compute

Every extra token, every oversized batch, every “just scale it up” experiment quietly burns energy, time, and money. We rarely talk about it because bigger numbers sound like progress.

What DeepSeek did is refreshingly simple but powerful: instead of treating text as endless streams of tokens, they compress it into visual form - basically turning paragraphs into compact images that models can read.
Same meaning. Fewer tokens. Less compute.

That shift - from more power to smarter representation - matters.
Because beyond the technical trick, it’s a reminder that efficiency isn’t just optimization - it’s ethics. Every token saved is compute that doesn’t need to run, energy that doesn’t need to burn, and access that doesn’t need to be gated by cost.

It’s not just about being efficient; it’s about being thoughtful.
If we can teach models to understand more while processing less, maybe we can start building systems that don’t come with such a heavy environmental and financial cost.

This short video is unpacking the idea - not because it’s flashy, but because it feels important.

Would love to hear your thoughts: how do you think we can make AI more sustainable, not just smarter?