Did OpenAI silently launch a new model? You're probably hallucinating – Business Insider
Jump to
Some ChatGPT users are getting hyped after the chatbot told them it was running on a new AI model.
OpenAI’s chatbot generated some speculation over the weekend after a number of users reported responses from ChatGPT that said it was being powered by “gpt-4.5-turbo.”
It would be a pretty big deal if that were the case. Except it probably isn’t.
GPT is OpenAI’s secret sauce. Specifically, it’s a large language model trained on troves of data that inform ChatGPT’s responses to questions. The latest version, GPT-4, was released in March to much fanfare.
OpenAI described GPT-4 as a “milestone” in its effort to scale up “deep learning.” It also suggested it had performance comparable with humans across a number of benchmarks, such as the bar exam.
It’s little wonder that ChatGPT’s suggestion of a bigger, better LLM powering it tantalized users.
Sully Omar, CEO of Canadian AI startup Cognosys, posted a screenshot to X on Sunday of a conversation with ChatGPT asking, “What is the precise name” of the model answering the query.
Its initial response said the model was “ChatGPT with GPT-4 architecture.” But after further prompting, it said, “This model is specifically named gpt-4.5-turbo.”
Matt Shumer, CEO of startup HyperWriteAI, wrote on X that he was “initially skeptical” of screenshots showing ChatGPT responses mentioning GPT-4.5. However, he claimed the responses were “replicable” on his attempt.
I was initially skeptical that these screenshots were real, but this does seem to be replicable on my end.
Whether this means it’s actually GPT-4.5 behind the scenes or not… who knows. https://t.co/FZrc4K4VqK pic.twitter.com/TrQYSmFhAn
So, did OpenAI shadow-drop an upgrade to the technology that has popularized AI this year? Given the fanfare OpenAI has generated around just about everything else it’s done, that seems unlikely.
One possibility is that ChatGPT simply hallucinated a new model for itself.
For example, OpenAI employees have claimed it’s a hallucination. Will DePue, a technical staff member, responded to a query on X about GPT-4.5 being legit with a clear rejection.
“No it’s a very weird and oddly consistent hallucination,” he wrote. Another employee said, “There’s no 4.5 and if there was it wouldn’t be released silently.”
you guys need to develop more resistance to crazy ai hype bros
there’s no 4.5 and if there was it wouldn’t be released silently and if it was released silently you wouldn’t have the api string self doxx as 4.5 🤣🤣
Nick Dobos, a former software engineer at Twitter, offered another reason as to why it might be a hallucination. All the examples of ChatGPT responding with “GPT-4.5” follow lots of prods for “extra details, he said.
“Super easy to see how that could get autocompleted after the internet is flooded with GPT-3.5-turbo and GPT-4,” Dobos wrote.
It’s, of course, possible that OpenAI is quietly testing a new version of its LLM. Google released its GPT rival Gemini earlier this month, putting pressure on OpenAI to respond.
Some users have also reported a sudden increase in the performance of ChatGPT after weeks of complaints that suggested it had been dumbed down or gotten lazy.
As Wharton professor Ethan Mollick noted over the weekend: “ChatGPT-4 suddenly got very good again for some reason, after being unreliable & a little ‘dull’ for weeks.”
Whether that’s because of a new GPT-4.5 or an improvement to GPT-4 is anyone’s guess. OpenAI boss Sam Altman dismissed a question asking if the GPT-4.5 leak was legit with a simple, “Nah.”
If OpenAI really wanted to give ChatGPT users a festive treat before Christmas, there would be little reason to keep it under wraps.
Read next