OpenAI announces GPT-4, says it outperforms 90% of humans on SAT

On Tuesday, OpenAI announced the latest version of its core large language model, GPT-4, which it says shows “human-level performance” on many professional tests.

ChatGPT-4 is “bigger” than previous versions, meaning it was trained on more data and has more weight in the model file, which also makes it more expensive to run.

Many researchers in the field now believe that many of the recent advances in artificial intelligence come from running increasingly large models on thousands of supercomputers in training processes that can cost tens of millions of dollars. GPT-4 is an example of a “scaling up” approach for better results.

OpenAI said it is being used Microsoft Azure for model training; Microsoft invested billions in the startup. OpenAI did not release details about the specific size of the model or the hardware used to train it, which could be used to reproduce the model, citing a “competitive landscape.”

OpenAI’s large GPT language model powers many of the AI ​​demonstrations that have excited people in the tech industry over the past six months, including Bing AI chat and ChatGPT, and the latest version is a preview of new advances that may begin to filter down to consumer products. such as chatbots, in the coming weeks. Bing’s AI chatbot uses GPT-4, Microsoft announced on Tuesday.

OpenAI says the new model will produce fewer wrong answers, be less likely to go off the rails and talk about forbidden topics, and even outperform humans on many standardized tests.

GPT-4 scored in the 90th percentile on the mock bar exam, the 93rd percentile on the SAT Reading exam, and the 89th percentile on the SAT Math exam, OpenAI claimed.

However, OpenAI warns that the new software is not yet perfect and in many situations it is less capable than humans. The company still has a serious problem with “hallucinations” or fabrication and is not factually reliable. He still tends to insist on being right when he is wrong.

“GPT-4 still has many known limitations that we are working on, such as social bias, hallucinations, and competitive cues,” the company said in a blog post.

“In casual conversation, the distinction between GPT-3.5 and GPT-4 can be subtle. The difference becomes apparent when the complexity of the task reaches a sufficient threshold—GPT-4 is more robust, creative, and able to handle much more nuanced instructions than GPT-3.5,” OpenAI wrote in its blog.

The new model will be available to paid ChatGPT subscribers and will also be available as part of an API that allows programmers to integrate artificial intelligence into their apps. OpenAI will charge about 3 cents for about 750 words of clues and 6 cents for about 750 words in response.