OpenAI Announces GPT-4, Says Beats 90% of Humans on SAT

Sam Altman, CEO of OpenAI, walks from lunch during the Allen & Company Sun Valley Conference on July 6, 2022 in Sun Valley, Idaho.

Kevin Deitch | Getty Images News | Good pictures

Announced by OpenAI Latest version It said on Tuesday that its flagship large-language model, GPT-4, has demonstrated “human-level performance” in several professional tests.

ChatGPT-4 is “bigger” than previous versions, meaning it’s trained on more data and has more weights in its model file, and costs more to run.

Currently, many researchers believe that the latest advances in AI come from running ever-larger models on thousands of supercomputers in training processes that cost millions of dollars. GPT-4 is an example of a “scaling up” approach to achieve better results.

OpenAI said it used Microsoft Azure to train the model; Microsoft has invested billions in startups. OpenAI did not disclose the specific sample size or details on the hardware used to train it, citing the “competitive landscape” that could be used to reproduce the model.

OpenAI’s GPT big language model powers many of the artificial intelligence demos that have captivated the tech industry over the past six months, including Bing’s AI Chat and ChatGPT, and the latest version is a preview of new developments that are starting to filter through. Consumer products such as chatbots in the coming weeks. Bing’s AI chatbot Uses GPT-4Microsoft said on Tuesday.

OpenAI’s new model will produce fewer truly incorrect answers, go off the rails, chat about taboo topics more often, and outperform humans on many standardized tests.

See also  Baltimore Ravens fire defense coordinator Dan 'Wing' Martindale, says source

The GPT-4 performed in the 90th percentile on the simulated bar exam, the 93rd percentile on the SAT reading test, and the 89th percentile on the SAT math test, OpenAI said.

However, OpenAI cautions that the new software is not yet perfect and is less capable than humans in many situations. It still has a big problem with “hallucinating” or creating objects, and it’s not really reliable, the company said. There is still the insistence that if it is wrong it is right.

“GPT-4 still has several known limitations that we are working to address, such as social biases, hallucinations, and aversive stimuli,” the company said in a blog post.

“In a casual conversation, the difference between GPT-3.5 and GPT-4 can be subtle. The difference becomes apparent when the complexity of the task reaches a high enough limit – GPT-4 is more reliable, creative, and can handle many more fine-grained algorithms than GPT-3.5,” OpenAI blogged. wrote in the post.

The new model will be available to paying ChatGPT subscribers and will also be available as part of an API that allows programmers to integrate AI into their apps. OpenAI charges about 3 cents for prompts of up to 750 words and 6 cents for responses of up to 750 words.

Leave a Reply

Your email address will not be published. Required fields are marked *