OpenAI's GPT-4.5 Turbo Accidentally Exposed, Likely to Launch in June
-
OpenAI's GPT-4.5 Turbo seems to have been leaked, as search engines like Bing and DuckDuckGo have indexed the GPT-4.5 Turbo product page before the official announcement.
However, the indexed links lead to a 404 page. But some preview text in the search engines claims that GPT-4.5 Turbo is OpenAI's fastest, most accurate, and most scalable model to date. The implications behind these terms remain to be seen.
It is predicted that this new model will likely be officially unveiled this Thursday, coinciding with the one-year anniversary of GPT-4's release. OpenAI's CEO, Sam Altman, is also expected to appear on the Lex Fridman podcast, following their in-depth conversation after the GPT-4 launch. GPT-4.5 Turbo: Expanded processing capacity and the latest information. The preview text mentions a 'context window,' referring to the amount of data the model can process at once. GPT-4.5 Turbo can handle up to 256,000 tokens, doubling the previous GPT-4 Turbo's 128K processing capability, roughly equivalent to 200,000 words. However, more important than the volume of input data is how reliably the model can process all this data. Current models with large context windows often ignore information, making them less valuable for analyzing large documents. If GPT-4.5Turbo performs better in this regard, even if its performance is similar to current models in other aspects, it would be a significant win. Google is currently beta-testing Gemini1.5, a model with strong large-context capabilities.
The preview description also indicates that GPT-4.5Turbo will be updated until June 2024, referred to as the "knowledge cutoff date." This suggests that the GPT-4.5Turbo model may be released in June. This would be quite unusual for OpenAI: to date, newly launched models have been immediately available upon release.
Rumors about the upcoming GPT-4.5Turbo date back to December 2023. Speculations suggest that the new model might include video or 3D capabilities in addition to text and image processing. However, the leaked preview does not mention multimodal capabilities. A few days ago, OpenAI CEO Altman hinted at a new AI model that is "worth waiting for." Additionally, a recent announcement from the University of California, Irvine (UCI) revealed that they are testing GPT-4.5 with their own chatbot, ZotGPT. Based on Azure AI, ZotGPT provides free access to GPT-4.5, supporting chat history, file uploads, and large context windows. This may also hint that GPT-4.5 Turbo could be coming soon.