San Francisco-based AI research company OpenAI announced on March 14, Tuesday, the launch of GPT-4, the successor of GPT-3.5 that powered ChatGPT.
According to OpenAI, GPT-4 (Generative Pre-trained Transformer 4) is a “multimodal” model, meaning it can generate content from image and text prompts.
GPT-3.5 vs GPT-4
“In a casual conversation, the distinction between GPT-3.5 and GPT-4 can be subtle. The difference comes out when the complexity of the task reaches a sufficient threshold,” says the company.
While GPT-3.5 is limited to about 3,000-word responses, GPT-4 can handle over 25,000 words of text, allowing for use cases like long-form content creation, extended conversations, and document search and analysis.
OpenAI says it spent around six months making GPT-4 safer and more aligned.
The fourth-generation GPT is 82 per cent less likely to respond to requests for disallowed content and 40 per cent more likely to produce factual responses than GPT-3.5.
Capabilities & limitations
Throwing light on the capability, OpenAI says ,”GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5.”
OpenAI evaluated GPT-4 through a variety of exams designed for humans.
During these evaluations, it performed well and often outscored the vast majority of human test takers.
For example, on a simulated bar exam, GPT-4 achieved a score that falls in the top 10 per cent of test takers. This contrasts with GPT-3.5, which scored in the bottom 10 per cent.
On March 14, Greg Brockman, President of OpenAI, shared a “developer demo livestream” of the latest edition of their AI chatbot named GPT-4 on YouTube.
During the demo, Brockman requested the bot to conduct tax calculations, which he referred to as “how to work with the system to accomplish a task that none of us like to do but we all have to.”
Check out the video below:
Despite its advanced capabilities, GPT-4 also suffers from similar limitations that plagued earlier GPT models, says OpenAI.
Similar to its predecessor, GPT-4 also lacks knowledge of events that have occurred after the vast majority of its data cuts off in September 2021.
“It is not fully reliable (e.g. can suffer from “hallucinations”), has a limited context window, and does not learn from experience. Care should be taken when using the outputs of GPT-4, particularly in contexts where reliability is important,” says the company.
However, the company revealed that GPT-4 scored 40 per cent higher than GPT-3.5 on internal adversarial factuality evaluations.
Currently, GPT-4 is available on ChatGPT Plus and as an API for developers to build applications and services.
01
Job board for modern workforce: How Remote Talent helps jobseekers find truly remote, distributed work