GPT-4, OpenAI’s new nugget: between increased precision and technological competition

By Baptiste, on March 16, 2023 - 2 min read

is making waves again with the arrival of GPT-4, the new star of language models, which outperforms its predecessor and once again demonstrates the talent of the San Francisco-based start-up. With the resounding success of ChatGPT, based on GPT3.5, the company has captured the world’s attention with its ability to converse naturally with humans.

The public has had a major role in the evolution of ChatGPT, providing critical feedback that has helped improve the design of the new model. OpenAI worked to strengthen the security of GPT-4 by making it less vulnerable to manipulation by malicious users.

A more reliable and efficient language model

The new generation is also more reliable, with a 40% lower chance of producing incorrect responses compared to GPT3.5. Language models, like those of Google and Facebook, tend to invent facts when they don’t know how to answer a question, and GPT-4 intends to prove that it can avoid these pitfalls.

GPT-4 also shines in its accuracy and versatility. Available to subscribers for $20 a month, the model accurately answers questions that left GPT3.5 perplexing. For example, AI researcher Chris Nicholson saw a marked improvement in results when he asked GPT-4 to develop a basic method for learning Spanish.

Expanded capabilities and innovative applications

Among its new capabilities, GPT-4 is able to analyze images and extract relevant information. OpenAI plans to use artificial intelligence to help create menus based on the contents of a refrigerator. The start-up Be My Eyes also plans to use this capability to help visually impaired people perceive the content of images.

Impressive results in prestigious exams

OpenAI reports that GPT-4 has also achieved impressive results on prestigious exams, such as the LSAT and Ivy League college entrance exams. For example, on a mock bar exam, GPT-4 ranks in the top 10% of students, unlike GPT3.5 which was in the bottom 10%.

The era of competition and the quest for transparency

OpenAI’s transparency leaves much to be desired. The company admits that it does not disclose details about the architecture, hardware, computational power, database construction or training method of GPT-4 because of the competition and the implications of such a model.

Microsoft, which has invested heavily in OpenAI, is banking on AI to improve its services and plans to integrate GPT-4 into its search engine, the “new Bing“. For now, GPT-4 is available in ChatPlus.

The future of artificial intelligence and the rivalry between tech giants

Competition in the field of artificial intelligence is intensifying, with giants like Google revealing their own chatbot, Bard, which should be integrated into their search engine in the coming months, despite a rocky start. Start-ups in the sector are attracting more and more investment and could, one day, compete with the tech heavyweights.

The race for innovation in artificial intelligence is on, and with the launch of GPT-4, OpenAI is proving that it is determined to stay ahead of the pack. As language models continue to evolve and become more accurate, the potential applications are endless. Users, businesses, and researchers alike look forward to seeing how these advances will transform the technology landscape and shape our future.