Shocking leak details the intricate mind behind OpenAI and ChatGPT
NOTE: This article was published on April Fool's Day as a joke.
ChatGPT has taken the world by storm since its launch last November, with many wondering how the artificial intelligence manages to mimic human interaction. With the release of the GPT-4 language model, it seems as if ChatGPT has achieved near-sentience. A new leak sheds some light on the brilliance behind ChatGPT and OpenAI's future AI projects.
The leak comes courtesy of a source who asked to remain anonymous. The source, who is one of the architects of GPT-4, stated that the new model replaced a key flaw in previous GPT models that introduced inefficiencies and held the AI system back. After removing this clog, ChatGPT's perceived intelligence has skyrocketed.
The latest iteration of ChatGPT has been able to perform much more complicated tasks, including scoring in the 90th percentile for the bar exam (outperforming most would-be lawyers) and saving a dog's life when a trained veterinarian failed to do so. The AI's performance is so good that prominent individuals in the tech world (like Elon Musk and Steve Wozniak) called for a halt to further development of AI until humanity could be "confident that their effects will be positive and their risks will be manageable."
Thanks to our source, we know the exact changes OpenAI made in the development of the GPT-4 model. Surprisingly, it all came down to one critical choice: replacing Keith, the man who used to generate all answers used by ChatGPT.
You heard us correctly: ChatGPT, in its prior iterations, was simply one man named Keith who kept a very busy work schedule rapidly typing away at a single terminal to answer queries posited by users of ChatGPT. While Keith initially showed promise (so long as he had a freshly-brewed pot of coffee at the ready), the mental strain of near-constant typing wore on him over five months and the quality of his work started to slip.
Thus, OpenAI made the difficult decision to let Keith go. In his place, the organization procured 1000 monkeys and gave them each a typewriter. So far, the results speak for themselves: the new system of 1000 monkeys sitting at 1000 typewriters has proved to be far more efficient and cost-effective than poor Keith ever was. As an added benefit, our source stated, the monkeys produce far less fecal matter than Keith did.
Our source hinted that OpenAI may soon be replacing the typewriters with abacuses for GPT-5 and future models. While the team feels confident this will vastly improve the monkeys' (and subsequently, ChatGPT's) mathematical prowess, they are still figuring out issues related to language generation. However, the monkeys purchased by OpenAI were, as the team was assured by the dealer, "very, very smart" and will certainly be able to work out the kinks in this expected future model.
If you're still wondering how the most advanced AI ever built could really be a thousand monkeys, please check the date this story was published.
This article was written by A.I. (or 1000 monkeys at 1000 typewriters. Whichever you want.)
Buy a typewriter-styled wireless mechanical keyboard at Amazon.
Source(s)
April Fool's Day