The Alignment Revolution

The Race to Control What We Cannot Understand

Two weeks after ChatGPT's launch, engineers at OpenAI’s headquarters in San Francisco stare at the usage metrics of their brand new baby. They are stunned. ChatGPT has achieved 100 million users in two months, the fastest technology adoption in human history.

People were using it for everything. Students were using it to help with their homework. Office workers were using it to compose emails. People were asking ChatGPT medical questions, seeking legal advice, using it for therapy. These are significant life decisions, and people were trusting a chatbot barely two months old.

Then the reports start coming in. A user convinced ChatGPT they should leave their spouse. Another used it to plan a cyberattack (ChatGPT provided detailed steps before safety systems caught it). Students were cheating en masse. Lawyers were submitting AI-generated briefs with hallucinated case citations. Someone asked ChatGPT how to make a bomb. It initially refused, but then they tricked it into answering by rephrasing the question. ChatGPT complied.

An alarmed Sam Altman (OpenAI CEO) confessed in a meeting, "We built something incredibly powerful. Millions are depending on it. And we don't fully understand how it works. We can't predict what it will say. We can't guarantee it won't cause harm. And we just gave it to the world."

There was a long pause.

"We need to figure out alignment. Fast. Because this is just GPT-3.5. GPT-4 is coming. And GPT-5 after that. They'll be more powerful. More persuasive. More capable. If we can't control this..."

He doesn't finish the sentence. He doesn't have to. The unspoken question hangs in the room. What have we created? And can we control it?

The revolution is not in the technology. It's in the choice we're making to build the most powerful technology the world has ever seen.

 

Much more to come. Stay tuned!

 

Links

AI Revolutions home page.