AI at the Crossroads

The only safe AI is open source. Closed AIs are dangerous.

Suddenly, AI announcements are everywhere, and it’s clear that the rate of change going on has exceeded the capacity of people to understand it. First, let’s recap the announcements:

• OpenAI has released a significant upgrade to its popular large language model, GPT-4, to widespread acclaim.

• Google and Microsoft have folded AIs into their productivity suites.

• Meta’s LLaMA, with performance comparable to GPT-3, was released as open source in February but without the weights required to modify it. The weights were ~leaked shortly before these announcements, making it truly open source.

Let’s take a moment to put this news into perspective.