A few hundred people decide what AI tells the planet. Open models keep their errors local

A few hundred people decide what AI tells the rest of the planet.Even the noblest of us lose their judgment.

Sometimes gradually, at times abruptly and for good.

Some join cults and find their way out years later.Some never wake up.

We see something others don't. It makes us feel important, special.

And then we stop checking whether it's true.

Most of the time, the damage stays small. Ourselves. A person. A family. A company.

But judgment failure scales with power.

When one person with fogged judgment runs a household, a family suffers.When a small group with fogged judgment runs the systems everyone uses, everyone suffers.

History knows a lot of these stories. Every generation carries scars from the last time a small group was certain they were right.

Now back to AI.

Most of us use models from Anthropic, OpenAI and Google. Their AI hands the worldview of those few hundred people to all of us. Their sense of what's good and bad reaches into every chat, every answer, every person, every day.

Our kids grow up inside this. They do their homework with these systems. They ask them the questions they don't dare ask us. Whatever worldview sits in the model seeps into theirs. Silently. Nobody asked us.

We would never accept one company writing every schoolbook on the planet. With AI we are doing exactly that right now, for something far more intimate than schoolbooks.

No bad intent required. Fallibility is enough.

That's why diversity in open source LLMs and sovereign AI matters so much. When one group's judgment fogs up, and at some point it will, the fog stays local. And a school, a company, a country can choose and adjust its models the way it chooses its textbooks, and replace them when they fail.

The timing helps. Open models have caught up. On several benchmarks they sit right next to the closed labs, sometimes above them. Give it a year and the real question may be why anyone still pays for a black box.

This gets concrete the moment AI moves beyond the chat window into your systems. Document extraction, internal search, classification, drafting: the use cases with real ROI run well on open models today, hosted where you decide.

That option exists. Take it.