GPT-4 was launched on Tuesday, with the update allowing users to provide both image and text inputs.
OpenAI has launched GPT-4, an Artificial Intelligence (AI) language model that updates the previous version of the technology GPT-3.5 and “exhibits human-level performance on various professional and academic benchmarks,” according to the developers.
GPT stands for Generative Pre-trained Transformer, an artificial intelligence algorithm programmed to write like a human.
Earlier executives from Microsoft Germany announced that they would introduce GPT-4 within a week and that it will offer a lot of new possibilities, such as the use of videos, however, Microsoft confirmed today that Bing Chat, a chatbot the company developed with OpenAI is already running on GPT-4.
Multi-modality: Using text, photos, and videos
While there had been speculation that the new version would be able to generate images in addition to text from the same interface, it turns out that is not the case. GPT-4 can handle image inputs but cannot output anything more than text. Models that make use of text, images, and video are called multimodal.
An example of how that could work would be to send an image of the inside of your fridge to the AI, which would then analyze the ingredients available before coming up with recipe ideas. At the moment this capability is only available through one of OpenAI’s partners, Be My Eyes. Its Virtual Volunteer can answer questions about images sent to it.
GPT-4: Making the grade
According to OpenAI, the upgrade to GPT has seen it improve massively on its performance on exams, for example passing a simulated bar exam with a score in the top 10%. GPT-3.5 was in the bottom 10%.
According to OpenAI the difference in casual conversation between the models can be hard to discern, however when dealing with more complex discussion GPT-4 is “more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5.”
This enables deeper discussions and facilitates solutions to more complex issues.
Technology that functions in any language
According to Microsoft Germany executive Andreas Braun, the technology that teaches machines to understand language in a way that only humans used to has become so advanced that it practically functions in all languages. Braun says that multimodality also makes the models even more comprehensive.
GPT-4 launched 14 March
OpenAI released GPT-4 on 14 March. It’s available to OpenAI’s paying customers who use ChatGPT Plus ($20 a month plus taxes where applicable) with usage limitations. According to OpenAI, the exact usage cap will adjust “depending on demand and system performance in practice.”
GPT-4 limitations
Despite the advances, the GPT-4 still has limitations. Like previous iterations it generally lacks knowledge of anything that happened after September 2021, and “it does not learn from its experience,” admits OpenAI.
What’s more, “It can sometimes make simple reasoning errors which do not seem to comport with competence across so many domains or be overly gullible in accepting obvious false statements from a user. And sometimes it can fail at hard problems the same way humans do, such as introducing security vulnerabilities into code it produces.”
0 Comments