OpenAI announces GPT-4, the next generation of artificial intelligence language model – artificial intelligence

company He claims that the model is “more creative and collaborative than ever before” and “can solve difficult problems with greater precision”. It can analyze both text and images, though it can only respond via text. OpenAI also warns that the systems retain many of the problems seen in earlier language models, including a tendency to fabricate information and the ability to generate violent and malicious text.

OpenAI says it has already worked with several companies to integrate GPT-4 into their products, including Duolingo, Stripe, and Khan Academy. The new model is available to the general public through ChatGPT Plus, a $20 monthly ChatGPT subscription from OpenAI, and today we learned that’s what has powered Microsoft’s Bing chatbot since day one. It will also be available as an API for developers to build on, with a queue It is here.

In the Detailed blog postOpenAI has stated that the distinction between GPT-4 and its predecessor GPT-3.5 is “narrow” in casual conversation (GPT-3.5 is the model that supports ChatGPT). OpenAI CEO Sam Altman tweeted that GPT-4 is “still flawed, still limited,” but that it also “still looks more impressive on first use than when you spend more time with it.”

The company reports that GPT-4 improvements are evident in the system’s performance on a number of tests and standards, including the Standardized Tape Test, LSAT, SAT Math, and SAT Evidence-Based Reading & Writing. In the reported tests GPT-4 scored at the 88th percentile and above, a full list of system tests and results can be seen here.

Speculation about GPT-4 and its capabilities has been rife over the past year, with many suggesting it would represent a huge leap over previous systems. However, based on OpenAI’s announcement, this is not the case as the company previously warned. “People want to be disappointed and they will be disappointed,” Altman said in an interview with GPT-4 in January.

The rumor was raised again last week when a Microsoft executive leaked that the system would be released this week in an interview with the German press. The executive also suggested that the system should be multimodal, meaning that it would be able to produce not only text but other content as well. Many AI researchers believe that multimedia systems that integrate text, audio, and video offer the best path toward creating more capable AI systems.

GPT-4 is one of them, but it supports fewer media than some expected. OpenAI indicates that the system can accept text and images while producing only text. The company also says that the model’s ability to analyze text and an image simultaneously allows it to interpret more complex queries. In the following samples, you can see the system explaining the unusual memes and images:

It’s been a long journey to get to GPT-4, as OpenAI – and AI language paradigms in general – slowly gained traction for several years before skyrocketing in popularity in recent months.

The initial paper describing GPT was published in 2018, with GPT-2 announced in 2019 and GPT-3 in 2020. These models are trained on huge text datasets, most of which come from the Internet, which are analyzed to find statistical patterns. . These patterns are then used to predict the next word. It’s a relatively simple mechanism to describe, but the end result is flexible systems that can produce, abstract, and paraphrase type, as well as perform other text-based tasks such as translation or code generation.

OpenAI initially delayed releasing its GPT models because it feared they could be used for malicious purposes, such as generating spam and disinformation. But in late 2022, the company launched ChatGPT – a chatbot based on GPT-3.5 that anyone can access. The launch of ChatGPT sent the tech world into a frenzy, and Microsoft quickly followed suit with its own Bing AI chatbot (part of the Bing search engine) and Google is tracking with its own offering.

As expected, the wider availability of artificial intelligence (AI) language models has created problems and challenges. The education system is still adjusting to having programs that write great college essays. In addition, online sites such as Stack Overflow and science fiction magazine Clarkesworld have been forced to shut down submissions due to streaming AI-generated content while early uses of AI writing tools in journalism were problematic at best. However, some experts are of the opinion that the adverse effects were less than expected.

In its announcement about GPT-4, OpenAI highlighted that the system had undergone six months of security training, and that in internal testing, it was “82% less likely to respond to requests for unauthorized content and 40% more likely to provide feedback.” from GPT-3.5.

However, this does not mean that the system does not make errors or produce malicious content. For example, Microsoft revealed that the Bing chatbot has been powered by GPT-4 since its inception, and many users have managed to break Bing’s firewalls in all kinds of creative ways, making the bot give dangerous advice, threaten users, and fabricate information. GPT-4 still has no knowledge of events “that occurred after the vast majority of its data was turned off” in September 2021.