New ChatGPT-4 gets top scores on law and medicine exams, claims OpenAI

Machine learning software company OpenAI has just unveiled its latest AI model, GPT-4, and is already making loud claims about the system's "unprecedented" stability and capabilities.

The company says GPT-4 performs very well on standardized tests, according to mock exam results published by the company.

In a report published Tuesday, OpenAI claims that GPT-4 scored in the 90th percentile or higher on the bar exam, the GRE oral exam, and the SAT reading and writing exam - without the model specifically preparing for those tests. This means that the automated system took these exams without "preparation" and was able to score higher than most humans.

According to the company, he could score 1410 on the SAT, pass the Bar and GRE exams, and get all B's and A's on the AP Art History, Biology, Calculus BC and Chemistry exams - high enough scores to earn college credit.

It also, quite coincidentally, performed well on sommelier tests. Perhaps GPT-4 has a future in writing descriptions on the back of wine bottles.

GPT-4 is the latest generation of OpenAI's Generative Pre-trained Transformer models, which use machine learning to create text based on the material they have been trained on. GPT-3 was launched in 2020 and has since generated a lot of buzz around how powerful this model is, how AI can help solve problems like misinformation, hate speech, fraud, and whether robots are coming to all of our workplaces.

Since the model is in the public domain, OpenAI GPT-3 has been used by companies to create realistic AI companions, mental health counseling, and chatbots posing as dead dictators - all with questionably successful results.

The new GPT-4 will be able to write in multiple programming languages, generate narrative scenarios, answer complex questions with step-by-step instructions, and interact with images, the company claims.

Founded in 2015 as a nonprofit open source research group by Silicon Valley investors Sam Altman, Elon Musk, Peter Thiel and LinkedIn co-founder Reid Hoffman, OpenAI has since become a for-profit company leading the AI arms race - an industry that has led to uncertainty in industries such as design and news writing, where AI is increasingly being used to create content for the web.

When OpenAI announced GPT-2 in 2019, it said it would not publish the source code for the training model because of "concerns about malicious applications of the technology." Three months later, it published the code on Github.

OpenAI will sell access to GPT-4 through a token-based system.

"GPT-4 and subsequent models have the potential to significantly impact society in both beneficial and harmful ways," OpenAI said in a press release about GPT-4. "We are working with outside researchers to improve understanding and assessment of potential impacts, and to create estimates for hazardous capabilities that may appear in future systems."