GPT-3, the AI That Started It All, Is Switched Off Today

In 2020, GPT-3 was the first AI that could write in convincing sentences, answer questions and hold something like a conversation. It was the moment many people realised computers could actually "think" on paper. Nearly every early AI writing assistant, customer-service bot and homework helper was built on top of it. Its best-known version was called Davinci, and developers paid by the word — fractions of a penny each — to use it.

Today that service was shut down. The guidance is to move to much newer, much larger models. The problem is that these aren't plug-in replacements. A bigger model is like swapping a bicycle for a delivery van: more powerful, but slower to start, pricier to run, and it answers questions in a slightly different style. Code written for the old model often needs rewriting and retesting. Some developers say the suggested replacements are wildly oversized for what they actually need.

The practical effect for you, the user, is quieter but real. Some smaller apps and tools may get slower, change their voice, or start charging a subscription to cover higher costs. Others may simply stop working. Because GPT-3 was so cheap, many niche tools — hobby projects, classroom helpers, small business bots — ran on tiny budgets that no longer stretch as far.

The bigger lesson is about ownership. AI rented from a company can vanish overnight, on the company's schedule, with little say from the people who depend on it. That's why a growing group of enthusiasts run AI models locally, on their own laptops, where nothing can be switched off remotely. It's slower and less capable, but nobody can retire it for you.