OpenAI has introduced GPT-4, the latest in their series of AI language models that power apps such as ChatGPT and the new Bing.
GPT-4 is now official!
The model can digest both text and visual input but can only answer via text. The company says the model is “more creative and collaborative than ever before” and “can solve tough issues with more accuracy”.
OpenAI warns that the systems still have many of the same issues as earlier language models, such as the tendency to “hallucinate” and the ability to produce offensive and violent material.
According to OpenAI, it has already worked with several companies to include GPT-4 into their products, including Khan Academy, Stripe, and Duolingo. The new model powers Microsoft’s Bing chatbot and is accessible to everyone through ChatGPT Plus, OpenAI’s USD 20 (around PHP 1.1K) per month ChatGPT membership.
OpenAI said in a blog post that the distinction between GPT-4 and its predecessor GPT-3.5 is “subtle”. GPT-4 “is still flawed, still limited,” according to OpenAI CEO Sam Altman’s tweet, who also noted that it “looks more remarkable on first use than it does after you spend more time with it.”
The performance of the system on several assessments and benchmarks, including the Uniform Bar Test, LSAT, SAT Math, and SAT Evidence-Based Reading & Writing examinations, is claimed to prove GPT-4’s improvements.
GPT-4 achieved results in the 88th percentile or above on the examinations mentioned, and a complete list of exams and the system’s results may be seen here.
Throughout the past year, there has been much speculation over GPT-4 and its capabilities, with many predicting that it would represent a significant improvement over earlier systems. OpenAI earlier issued a warning, and based on the announcement, the improvement is more incremental.
In its release of GPT-4, OpenAI stated that the system had undergone six months of safety training and that, in internal testing, it was “82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5.”
The new Bing runs on OpenAI’s GPT-4
Moreover, after OpenAI announced the GPT-4, Microsoft confirmed that the new Bing is running on GPT-4, which is customized for search.
“If you’ve used the new Bing in preview at any time in the last six weeks, you’ve already had an early look at the power of OpenAI’s latest model.” Said Yusuf Mehdi, Microsoft corporate vice president of Modern Life, search, and devices.
He added that Bing benefits from OpenAI’s updates to GPT-4 and beyond so that our users have access to the most comprehensive copilot features available.
However, it doesn’t mean the system doesn’t make some mistakes or produce harmful content, though. For instance, many users of Bing were able to get through the chatbot’s security measures in a variety of inventive ways, causing it to give risky advice, threaten users, and fabricate information.