INDIA FAKE NEWS

What Makes Open AI’s Newest Release, GPT-4, Distinct from Its Predecessor, ChatGPT?

Facebook
Twitter
LinkedIn
Pinterest
Pocket
WhatsApp

If you’re still in awe of what ChatGPT can do, you’re in for an even bigger shock with the release of GPT-4. On March 14, OpenAI released GPT-4, which the company claims is more stable, imaginative, and equipped to deal with much more nuanced instructions than GPT-3.5 and its predecessors.

OpenAI claims that GPT-4, “a large multimodal model (accepting image and text inputs, emitting text outputs),” exhibits human-level performance on various professional and academic benchmarks despite being less capable than humans in many real-world scenarios.

GPT-4 can read, analyse, or generate up to 25,000 words of text, making it a potentially game-changing advancement in problem solving. The ability to process images as input is a major strength of GPT-4. In addition, it can generate source code in all the major programming languages and provide deep analysis and reasoning.

The most recent version can be used exclusively on ChatGPT Plus, and it also serves as an API for programmers to create new software and services. ChatGPT Plus has a $20/month subscription fee; however, API access requires a spot on a waiting list. What follows is a comparison of GPT-4 to earlier versions of ChatGPT.

Accepts Image Data

GPT-4’s ability to accept and generate image inputs and outputs is what sets it apart from older versions like ChatGPT and GPT-3.5. It’s not just an image description tool; it can also answer questions about images with inference.

To show how it functions, Open AI provided several examples of visual inputs. To give just one example, when questioned as to “Why is this picture so amusing? Please give a panel-by-panel breakdown of the event “Here is GPT-4’s response.

An answer from GPT-4:

“An adapter for a “Lightning Cable” is shown in a three-paneled package.

Panel 1 A VGA connector (a bulky, blue, 15-pin connector typically used with computer monitors) is plugged into the smartphone’s charging port.

Panel 2 shows the “Lightning Cable” adapter’s packaging, which features an image of a VGA connector.

Panel 3 shows a detailed view of the VGA cable’s tiny Lightning connector, which is used to power and charge iPhones and other Apple products. The image is humorous because of the incongruity of a bulky, antiquated VGA connector being inserted into a compact, cutting-edge smartphone charging port.”

OpenAI also provided examples showing how GPT-4 can be used to perform calculations using image data, detect elements within an image, summarise images with text, and even decode memes.

Expertise in Higher-Level Reasoning

GPT-4’s improved problem-solving precision stems from its expanded breadth of knowledge and skill set. To demonstrate this, OpenAI tested ChatGPT and GPT-4 on the same scenario-based question. In spite of the fact that they were both correct according to the facts, GPT-4 offered a more convincing defence of its answer and an improved method for solving the problem.

Larger Percentile Rank

After being put through their paces on a battery of benchmarks, GPT-4 passed a mock bar exam with a score in the top ten percentile, while GPT-3.5 performed in the bottom ten percentile.

More secure and congruent

When compared to GPT-3.5, OpenAI’s newly released GPT-4 is 82% less likely to respond to requests for restricted content and 40% more likely to generate factual responses, as determined by the company’s internal evaluations. For better behaviour, GPT-4 is built with more human input, from both ChatGPT users and domain experts.

Cautions and Risks

There are a number of known and possibly unknown limitations to GPT-4 that are being addressed concurrently. There are constraints such as social biases, scenario fabrication, and the provision of harmful advice. Like its predecessors, GPT-4 is not foolproof and can make logical mistakes. Furthermore, Open AI’s study notes that “GPT-4 generally lacks knowledge of events that have occurred after the vast majority of its data cut offs (September 2021).” It went on to say, “GPT-4 can also be confidently wrong in its predictions, not taking care to double-check work when it is likely to make a mistake.”

Facebook
Twitter
LinkedIn
Pinterest
Pocket
WhatsApp

Never miss any important news. Subscribe to our newsletter.

Leave a Reply

Your email address will not be published. Required fields are marked *