OpenAI’s new GPT-4 can perceive text and images

OpenAI has found that GPT-4 considerably outperforms existing large language models

OpenAI today introduced GPT-4, its advanced and sophisticated model that can take in and generate up to 25,000 words of text, which is roughly eight times more than GPT-3.5, which powers ChatGPT.

According to OpenAI, GPT-4 can “solve difficult problems with greater accuracy, thanks to its broader general knowledge and problem solving abilities.” The current GPT-3.5 powering ChatGPT can only take text prompts as input, whereas GPT-4 can accept images as inputs and generate captions, classifications, and analyses. “While less capable than humans in many real-world scenarios, [GPT-4] exhibits human-level performance on various professional and academic benchmarks.”

In an example on the company’s website, we see an image with eggs, flour and milk alongside text that reads “What can I make with these ingredients?” The language model replied with:

There are many options for what you can make with these ingredients! Some possibilities include:

  • Pancakes or waffles
  • Crepes
  • French toast
  • Omelette or frittata
  • Quiche
  • Custard or pudding
  • Cake or cupcakes
  • Muffins or bread
  • Cookies or biscuits

These are just a few examples, but the possibilities are endless!

Another example of the model being able to take images as input can be seen below:

The new model is also reportedly smarter than GPT-3.5. The model exhibits human-level performance in professional and academic benchmarks. For example, “it passes a simulated bar exam with a score around the top 10% of test takers; in contrast, GPT-3.5’s score was around the bottom 10%,” wrote OpenAI.

OpenAI also evaluated GPT-4 on traditional models designed for benchmarking machine learning models. It found that GPT-4 considerably outperformed existing large language models.

Further, unlike ChatGPT, which has a classic and fixed verbosity, tone and style, developers would be able to prescribe the GPT-4 AI’s style and task by describing those directions in the system message. “System messages allow API users to significantly customize their users’ experience within bounds,” wrote OpenAI.

For now, ChatGPT Plus subscribers will get GPT-4 access on chat.openai.com with a usage cap. The company says that it will adjust the exact usage cap depending on demand and system performance in practice, but expects to be severely capacity constrained, with the capacity issues easing off over the coming months.

Read more about the advanced large language model here.

Image credit: OpenAI

Source: OpenAI