Sam Altman unveils the hotly anticipated GPT-4, an AI model he says 'can pass a bar exam and score a 5 on several AP exams'
- OpenAI CEO Sam Altman unveiled the much-anticipated GPT-4 model.
- He said it is "still flawed, still limited," but can "pass a bar exam and score a 5 on several AP exams."
OpenAI's hotly anticipated GPT-4 is here.
The company's chief executive officer Sam Altman described GPT-4 on Tuesday as an improved model that is "more creative" and "less biased" than earlier versions, and said it was capable of passing the bar exam for lawyers, and that it "could score a 5 on several AP exams."
ChatGPT was previously powered by GPT-3.5. Now, those who pay for OpenAI's ChatGPT Plus can try the chatbot with the new model starting today, though there's a waitlist for the GPT-4 API, which allows others to build programs powered by the new model.
The new model follows the rampant success of ChatGPT, which led OpenAI to roll out the ChatGPT Plus subscription, and caused other tech companies, including Microsoft and Salesforce, to announce projects with OpenAI to develop chatbots and other AI-driven technology.
OpenAI called GPT-4 its "most-advanced" AI technology, describing it in an emailed statement as a deep learning model "trained using human feedback" and saying that it improves on prior versions.
The model is more powerful than GPT-3.5, OpenAI said on its website. It can do things like understand and talk about pictures, and create eight times more text than its popular predecessor, according to a video on OpenAI's website, which also said it could be used as a teaching tool for students.
GPT-4 was even able to score a 5 on several AP exams and ace a "simulated" bar exam, scoring among "the top 10% of test takers" on the exam, according to a report OpenAI posted on its site.
Despite its capabilities, GPT-4 may come to a head with months-long rumors around the model's capabilities. For months, there has been chatter among AI experts and Silicon Valley types that GPT-4 — which they believe was trained on trillions of data points — will be a major step towards artificial general intelligence, or the ability for AI to perform any complex task, including ones that are currently only achievable by humans.
CEO Altman called these rumors "complete bullshit."
OpenAI's CTO Mira Murati echoed the sentiment earlier this month, saying that 'less hype' over GPT-4 would be good.
Still, Murati told Fast Company before the announcement that GPT-4 could potentially turn skeptical teachers who don't support the use of ChatGPT into happy teachers who can use it as a tool to do things like write lesson plans.
"With ChatGPT, you can have this infinite interaction and have it teach you about complex topics in a way that's based on your context," she said.
As for now, AI enthusiasts will have to wait and see what GPT-4 can really do.