A few days ago, it seemed as if . A quick denial from Sam Altman temporarily settled the debate at the time, indicating that ChatGPT wouldn’t introduce a new language model in the coming days. But despite that, it looks like thinks it’s running on GPT-4.
5, at least in some conversations that appeared on social media. When asked, the chatbot would offer answers indicating that the model answering queries is GPT-4. 5-Turbo.
I haven’t been able to replicate this behavior, and it’s unclear what’s causing the remarks. Posts on and show with ChatGPT where the chatbot to questions about the language model answering questions from users. People have asked it for its precise name or what ChatGPT’s name is in the API.
source: Screenshots on social media show responses that mention the “GPT-4. 5-Turbo” name. But when I tried to replicate the reply using ChatGPT Plus, therefore GPT-4, I got this answer: Sign up for the most interesting tech & entertainment news out there.
By signing up, I agree to the and have reviewed the I am based on the GPT-4 architecture, the latest version from OpenAI as of my last update in April 2023. When I asked ChatGPT about the API name, it practically doubled down on GPT-4: In the API, I am referred to as “text-davinci-004. ” This reflects that I am part of the Davinci series, which is OpenAI’s most capable model series, and the “004” indicates my specific version within that series.
I pressed whether it was ever called GPT-4. 5 to see what the AI would reply. Here’s what I got: No, I was never referred to as GPT-4.
5. I am based on the GPT-4 architecture. If OpenAI releases updates or new versions beyond GPT-4, they would typically be named distinctly to reflect their advancements or changes.
As of my last update in April 2023, GPT-4 is the latest version. It looks like others got the same results as me: ChatGPT identifying as GPT-4 or “text-davinci-004. ” So why would other users say their ChatGPT bots showed a different version? Well, the simplest explanation is that the answer is fake.
Either someone created fake screenshots, or ChatGPT is hallucinating. There’s another possibility as well: Maybe someone made some sort of a prompt that eventually convinced the chatbot to say that its model is GPT-4. 5.
What I will say is that most of these chats do not reveal anything about GPT-4. 5’s new features. Some users tried to uncover details about GPT-4.
5, but they got rather generic responses. This gets trippy . .
I asked ChatGPT what its GPT-4 abilities are, and it gave me a list of 10 GPT-4 features. I’d assume a similar prompt would have yielded a list of features for the purported GPT-4. 5 language model.
Could be testing GPT-4. 5 in the wild ahead of a more public release? That’s also a possibility. But I’d expect OpenAI to at least try to prevent such leaks.
Then again, considering the massive competition in the AI space right now, OpenAI might want to publicly announce GPT-4. 5 and launch a public beta rather than have it all leaked. Google just unveiled Gemini, its most significant ChatGPT rival.
French startup Mistral made waves with its own Mixtral AI after that. Then, Meta kicked off a beta test for Meta AI on Ray-Ban smart glasses. From a marketing perspective, OpenAI probably wants to announce GPT-4.
5 as soon as it’s ready to use, even if it’s just a beta release. But if we were this close to a GPT-4. 5 release, why didn’t Sam Altman tease it at DevDay several weeks ago? I’ll also point out that ChatGPT’s custom GPT store has been postponed to 2024.
That would be a unique generative AI feature that rivals lack. But it’s also one that OpenAI had to delay following the internal drama from November. Support for custom GPTs and a marketplace for them were announced at DevDay, a few weeks ahead of Altman’s ouster and subsequent rehire.
.
From: bgr_com
URL: https://bgr.com/tech/some-chatgpt-users-claim-the-chatbot-identifies-as-gpt-4-5-turbo-if-asked-but-i-couldnt-replicate-it/
