OpenAI's got a new report card for AI progress²¹⁶, and they're giving themselves a single star (maybe two).

What's going on here?

OpenAI's cooked up a five-level scale to track how close we are to AI that can outthink humans.

What does this mean?

OpenAI's new AI report card goes from "can chat" to "can run the whole dang company":

• Level 1, Chatbots: Chatty AIs (we're here now)

• Level 2, Reasoners: Smartypants problem-solvers (getting there soon)

• Level 3, Agents: AI assistants that can work for days (investors go wild)

• Level 4, Innovators: AI inventors and innovators (sci-fi stuff)

• Level 5, Organizations: AI that can do the job of an entire organization (umm, what? How’s this level 5)

OpenAI thinks they're juuuust about to level up to tier 2. They're already showing off some fancy new GPT-4 tricks internally that might qualify.

This level system isn't set in stone—they'll tweak it based on feedback from employees, investors, and their board.

Why should I care?

If you're keeping score at home on the "are the robots taking over?" game, this gives you OpenAI's playbook. It's a peek into how one of the biggest AI companies sees the path to artificial general intelligence (AGI).

Anthropic has its own framework called AI Safety Levels (ASL) with three levels, but that one focuses on safety more (and less on general capabilities).