If you're someone who is still swooning over ChatGPT's capabilities, then get ready to be even more surprised with its latest version, GPT-4. OpenAI released GPT-4 on March 14, which according to the company is more reliable, creative, and will be able to handle much more nuanced instructions than GPT-3.5 and its previous versions.
According to OpenAI, "GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks."
Claiming to be a breakthrough in problem-solving capabilities, GPT-4 is capable of reading, analysing or generating up to 25,000 words of text. One of the biggest highlights of GPT-4 is its ability to understand images as input. Also, it can write code in all major programming languages and provide detailed reasoning and insights.
The latest version is available for use only on ChatGPT Plus and as an API for developers to build applications and services. While ChatGPT Plus requires a subscription of $20/month, to get API access, one needs to join the waitlist.
Here's a breakdown of how GPT-4 compares to ChatGPT and its versions.
Allows Image Inputs
Open AI shared several samples of visual inputs to demonstrate how it works. For example, when asked, "What is funny about this image? Describe it panel by panel," here's what GPT-4 replied.
GPT-4 's response: "The image shows a package for a "Lightning Cable" adapter with three panels.
Panel 1: A smartphone with a VGA connector (a large, blue, 15-pin connector typically used for computer monitors) plugged into its charging port.
Panel 2: The package for the "Lightning Cable" adapter with a picture of a VGA connector on it.
Panel 3: A close-up of the VGA connector with a small Lightning connector (used for charging iPhones and other Apple devices) at the end.
The humor in this image comes from the absurdity of plugging a large, outdated VGA connector into a small, modern smartphone charging port."
Other samples shared by OpenAI demonstrated GPT-4 capability to do calculations based on data in the image, detect elements in an image, provide a detailed summary of images containing text, and even help you understand memes.
Advanced Reasoning Capabilities
With its broader general knowledge and problem-solving abilities, GPT-4 is able to solve difficult problems with greater accuracy. To prove this with an example, OpenAI asked the same situation-based question to ChatGPT and GPT-4. While both answers were factually acceptable, GPT-4 came up with a better reasoning and solution.
Higher Percentile Score
On testing the two models i.e, GPT-4 and GPT-3.5 on a variety of benchmarks, GPT-4 passed a simulated bar exam with a score around the top 10% of test takers, while GPT-3.5’s score was around the bottom 10%.
Safer and More Aligned
According to OpenAI, in the internal evaluations, it was discovered that the newly launched GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5. GPT-4 is built by incorporating more human feedback including ChatGPT users and experts, for improved behaviour.
Limitations and Risks
Despite its advanced technologies and collaborative capabilities, GPT-4 still has many known and possibly unknown limitations that are being addressed simultaneously. The limitations include social biases, making up situations to generate responses or even giving out harmful advice. Just like its previous versions, GPT-4 is still not completely reliable and could make reasoning errors.
Adding another limitation, Open AI's research said, "GPT-4 generally lacks knowledge of events that have occurred after the vast majority of its data cut offs (September 2021), and does not learn from its experience." It further stated, "GPT-4 can also be confidently wrong in its predictions, not taking care to double-check work when it’s likely to make a mistake."
GPT-4 is gaining popularity exponentially with several apps including Duolingo, Be My Eyes, Khan Academy, Stripe and Morgan Stanley collaborating with the latest model. The government of Iceland is also partnering with OpenAI to use GPT-4 in the preservation effort of the Icelandic language. Its features and capabilities are expected to expand more as like ChatGPT, the company will continue "updating and improving GPT-4 at a regular cadence as more people use it."