With the unveiling of GPT-4, OpenAI has revolutionized the way that machines can understand and interact with human language, creating a new generation of AI-powered technologies that are more powerful and sophisticated than ever before. Without an understanding of how this technology works, and how it can change the world as we know it, most users don’t know how to feel about this information.
Everyone from Hank Green to the tech’s creators have expressed some trepidation around what’s coming next because of this tech, so how worried should we be? What exactly is this update, and how is it going to effect the world at large?
The easiest way to comprehend the power of GPT-4 is to compare it to its predecessor, the original base for ChatGPT: GPT-3.5. Here are a few of the major changes.
One of the biggest changes from GPT-3.5 to GPT-4 is the ability to process images. This kind of power can instantly produce image captions, descriptions, and more. In the future, we can expect to see AI-generated image descriptions used in a variety of applications such as accessibility aids, search engine results, and automated captioning.
For example, if a picture of the beach is fed into the model, the AI will generate a description such as:
“a sun-drenched beach with palm trees, white sand, and crystal-clear water”
In the launch video, OpenAI president Greg Brockman showcased the sheer power and creativity of the new Language Model. GPT-4 was able to handle complex prompts that its predecessor would have simply ignored. This will allow users to work with the AI for more complex and esoteric projects such as creative writing, lyric creation, and more.
These upgrades are great news for people who have been experimenting with ChatGPT, as we’ve already seen them attempt to write songs and even create sewing patterns and recipes. Artists will be far more successful working with this update to push boundaries and explore what might be possible in the future.
GPT-4 is able to understand the context of conversations and incorporate it into its responses. This enables it to generate more accurate and natural-sounding text, which makes the conversation easier to follow.
It can take information from upwards of 25000 words and synthesize it with ease. With this kind of context, the Language Processor can be used to generate more accurate summaries, and even have better success with natural language processing tasks like sentiment analysis.
Each iteration of GPT has included more data in its staring, allowing it to get a handle on more and more complex information and tasks. With this much sheer computation, GPT-4 was able to ace both the Uniform Bar Exam and the Biology Olympiad.
With a vast team of experts providing feedback throughout its training, GPT-4 is 40% less likely to “hallucinate” or produce false responses. These monitors were also able to keep the program from generating anything that has been disallowed. The new program is safer and far more accurate.
The speed with which this tech has made its way into applications is truly unprecedented. There are multiple companies making leaps and bounds to find out what GPT-4 is capable of.
Be My Eyes is a startup based out of Denmark that focuses on making the world at large more accessible for those with limited vision. They’ve recently launched a “Virtual Volunteer” feature that not only tells users what is in front of them but can also provide context and suggestions.
Where previous machine learning models would simply tell you there’s a piece of wood in front of you, GPT-4 is able to carry on a conversation and let you know if that wood is going to be a hazard. Other use cases that have already come to fruition include navigating public transit and processing what is important about different web pages. The more this program is used, the more accessible the world will become.
With its mission to create truly accessible education for everyone, Khan Academy is very aware of the different levels of help that students need. Since 2022 the company has been testing language modeling in a pilot program called Khanmigo. This virtual assistant would operate both as a virtual tutor and a teacher’s assistant.
GPT-4’s ability to carry a conversation and learn from its users allows Khanmigo to provide personalized help to each student in a class. When asked why it’s important to learn a certain concept, the language model can give specific answers based on what the individual student is interested in.
This program may even be able to help teachers understand what’s working and what isn’t in their lesson plans. It’s still in the early stages, but eventually, GPT-4 could change the face of education as we know it.
With the vast capabilities and speed of development, GPT-4 has generated quite a bit of buzz. This has launched the development team at OpenAI even more into the spotlight.
By getting to know a few of these developers, and hearing about their excitement and fears, we’re able to gauge the potential of GPT-4.
The GPT-4 team’s fearless, visionary leader is Jakub Pachoki. He’s been a part of the team since the very first iteration of the language learning program and says that not much has changed within the fundamental formula.
He’s always had a passion for pushing the limits of computers, something that has been clear since the very start of his journey with GPT. You’re likely to run into him at a coding or hacking competition, of which he’s won several.
In a recent tweet, Pachoki’s coworker Sam Altman praised Pachocki’s leadership, stating “we wouldn’t be here without it.”
As the CEO of OpenAI, Sam Altman has been on the razor’s edge of innovation for years. He’s excited about the creative possibilities that are available with advances in AI, but also worries about the possibilities for misinformation and job loss.
During an interview with CNBC, Altman said, “We can have a much higher quality of life, standard of living. People need time to update, to react, to get used to this technology.”
OpenAI’s chief technology officer, Mira Murati, helped develop DALL-E as well as ChatGPT. She’s passionate about how this technology can change education, helping fill the gaps in learning from one individual to the next.
With a focus on the collaborative nature of technological advances, Murati is excited to get more input from other developers, government regulators, and users. She’s ready for regulations to be put in place to prevent this kind of tech from being used in malicious ways.
With each wave of technological advancement, there’s a new boom of possibility. GPT-4 can do wonderful things like help the blind to see and change the way we learn and connect, but it also has the ability to spread misinformation and perpetuate the ideals of bad actors.
What happens next is up to us. With more time and collaboration between developers, users, and lawmakers, we can generate a future that’s worth chatting to AI about.