This week, the update of the ChatGPT artificial intelligence, GPT-4, was unveiled, revealing an AI that excels in new areas compared to version 3.5. In particular, it is now capable of passing the bar exam with outstanding results.

OpenAI has reached a new milestone with ChatGPT. The company behind the popular artificial intelligence has just unveiled the GPT-4 update of its generative tool on Tuesday, March 14, which excels in new areas, starting with law.

The OpenAI team had their AI take the American bar exam using the GPT-4 version. As a result, ChatGPT passed with scores approaching those of the top candidates. The chatbot was even in the top 10% of best results in the exam. By comparison, the previous version GPT-3.5 had also passed the exam, but was among the bottom 10% of candidates, according to OpenAI.

Of course, this is not about pleading cases, but about stating the law in practical situations. It is already known that some law firms rely on tools based on algorithms to obtain various case law and texts. Here, ChatGPT demonstrates that it could become a relevant tool to support the work of lawyers, particularly for compiling numerous texts.

Inaccuracies and errors still present, according to OpenAI

However, while ChatGPT has evolved impressively, OpenAI acknowledges that its conversational robot still shares inaccuracies and even errors. "It is still not entirely reliable. It imagines facts and makes reasoning errors," says OpenAI.

Above all, ChatGPT relies on information that precedes September 2021. In the case of law, for example, it would not take into account new laws, regulations or case law that have occurred since then.