Kamal Ghorui
Companies like OpenAI (who made ChatGPT) and Anthropic (who made Claude) spend massive amounts of money—sometimes over $8 billion—to train their AI. They keep their code “closed,” meaning it’s locked away like a secret recipe.
However, recent drama has surfaced involving a Chinese company called DeepSeek. They’ve been accused of “cloning” these expensive models. But since they can’t just download the secret code like a movie file, they use a sneaky trick called Distillation.
How “Distillation” Works
Think of it like this: A “student” AI (the rookie) asks a “teacher” AI (the pro) millions of questions.
The Goal: The student doesn’t just want the answer; it wants to see the teacher’s “work.”
The Result: By watching how the pro AI thinks, the student AI learns to give almost the same answers.
The Savings: While the original company spent billions, the “cloner” can recreate a similar brain for as little as $50 or $500 in just a few hours.
Why Is This a Problem?
If everyone can just copy the best AI for free, the companies that spent the billions might stop inventing new things. But there are two even bigger worries:
Safety: When a company “distils” an AI, they might strip away the safety rules. Imagine a super-smart AI that doesn’t have a “don’t be evil” button anymore.
Spying: AI is being used for high-stakes government work. If a rival country copies that AI’s logic, they could use it to cause chaos or hack systems.
What’s Next?
We can’t really stop this with just laws or treaties, because it’s too easy to do in secret. Instead, AI companies need to build better “burglar alarms” to detect when another bot is trying to milk them for information.
As countries like India start building their own top-tier AI, they’ll need to make sure their “digital brains” are locked up tight from day one.
END OF ARTICLE
