Did OpenAI simply by accident leak the subsequent huge ChatGPT improve?

Social media went abuzz final night time with a number of posts speaking a few potential new AI mannequin from OpenAI, the corporate behind ChatGPT. It seems the corporate inadvertently revealed a weblog publish on the mannequin, which was then listed by engines like google Bing and DuckDuckGo.

Sharp-eyed customers on Reddit and X (previously Twitter) seen a briefly listed weblog publish mentioning the GPT-4.5 Turbo mannequin. Whereas the web page has since been taken down and now throws a 404 error, the cached description hints on the mannequin’s superior pace, accuracy, and scalability in comparison with its predecessor, GPT-4 Turbo.

You could discover the leaked snippet above mentions a “information cutoff” of June 2024. By “information cutoff,” the outline is referring to the date when the AI will cease being educated on info. This has led some to consider it’s both a typo or an indication of a possible July/August launch for GPT-4.5 Turbo. For context, the present GPT-4 Turbo mannequin had a information cutoff of April 2023.

One other vital piece of data is the point out of a 256k token context window, doubling the 128k capability of GPT-4 Turbo. This transfer could possibly be OpenAI’s response to the rising development of enormous context home windows, notably after Google’s latest developments with their AI mannequin Gemini. Again in February, Google introduced Gemini had a context window of as much as 1 million tokens. The arrival of GPT-4.5 Turbo may assist stage the enjoying discipline a bit.

OpenAI has remained tight-lipped in regards to the leak, leaving the standing of GPT-4.5 Turbo and the June 2024 date shrouded in thriller for now. The timing, nevertheless, is intriguing, as March 14th additionally marks the primary anniversary of GPT-4’s launch.