What are the superior features of GPT-4 compared to ChatGPT?
(Baonghean.vn) - On its first day of release, GPT-4 surprised many users with its superior features compared to the previous version (ChatGPT), such as the ability to draft lawsuits, pass standardized tests, and build a working website from a hand-drawn sketch.
 |
Illustrative image. |
On March 14th, OpenAI announced the next generation of its artificial intelligence (AI) technology that underpins the ChatGPT chatbot. GPT-4 offers more powerful features than its predecessor (GPT 3.5 or ChatGPT), potentially changing how we use the internet for work, entertainment, and creativity. But it could also raise challenging questions about how AI tools might disrupt careers, allow students to cheat, and alter our relationship with technology.
What is GPT-4?
GPT-4 is the latest version of OpenAI's language modeling system. The previous version, GPT 3.5, supported the company's extremely popular ChatGPT chatbot when it launched in November 2022.
Because GPT-4 is a large multimodal model (emphasis on multimodality), it can accept both text and image input as well as human-like text output.
For example, with GPT-4, a user can upload a spreadsheet, and GPT-4 will be able to scan that spreadsheet and provide answers to the user's questions. GPT-4 can also read charts that the user uploads and perform calculations based on that data source.
Currently, the only way to access and use the GPT-4 version is to subscribe to OpenAI's premium paid service (ChatGPT Plus), priced at $20 per month. However, even with a paid ChatGPT Plus subscription, there are user limits, meaning users may not be able to access it whenever they want, a factor to consider before subscribing.
GPT-4 is an updated version of OpenAI's big language model, trained on large amounts of online data to generate complex responses to user requests. GPT-4 is now available through a waiting list and has been integrated into several third-party products, including a new version of Microsoft's Bing search engine. Some users with early access to the tool have shared their experiences and provided some compelling examples.
Here are five outstanding features of the GPT-4 version that previous versions like ChatGPT could not offer.
1. Analyze the entire image.
At its core, the biggest change in GPT-4 is its ability to work with user-uploaded images. One of the most impressive use cases to date is from an OpenAI video demonstration showing how a drawing can be transformed into a working webpage in minutes. The demonstrator uploaded the image to GPT-4, then pasted the resulting code into a preview showing what a working webpage might look like.
In its announcement, OpenAI also pointed out how GPT-4 was asked to interpret a joke from a series of images, including a smartphone plugged into the wrong charger, and describe why it was funny. While it sounds simple, analyzing a joke is more complex for AI tools because it requires the necessary context.
In another experiment, The New York Times showed GPT-4 a picture of the inside of a refrigerator and asked it to come up with a meal based on the ingredients in it.
The image uploading feature is not yet available on GPT-4 but is expected to be rolled out in the next few weeks.
2. Easier to write code
Some early users with little or no prior coding knowledge have used GPT-4 to recreate iconic games like Pong, Tetris, or Snake after following the tool's step-by-step instructions. Others have created their own games. According to OpenAI, the GPT-4 version can write code in all currently popular programming languages.
Arun Chandrasekaran, an analyst at global research and consulting firm Gartner Research, said: “GPT-4’s powerful language capabilities will be used for everything from storyboarding and character creation to game content. This could create more independent game providers in the future. GPT-4 and similar models could be used to create marketing content around previews, generate articles, and even moderate game discussion forums.”
Similar to the game, GPT-4 could change the way many people develop apps. One Twitter user claimed to have created a simple drawing app in minutes. While another claimed to have coded an app that suggests five new movies every day, along with trailers and details on where to watch them.
Lian Jye Su, an analyst at technology analytics firm ABI Research, said: “Writing code is like learning to drive; as long as beginners are guided, anyone can write code. AI can be a good teacher.”
3. Pass the tests with flying colors.
Although OpenAI claims that GPT-4 is inferior to humans in many real-world situations, it demonstrates “human-level performance” in various professional and academic tests.
OpenAI claims that the GPT-4 version recently passed the US bar exam with a score in the top 10% of test-takers, while the GPT-3.5 version scored below 10%.
According to OpenAI, the latest version also performs well in exams such as the standardized test administered by the US Bar Association Accreditation Council (LSAT), the widely used standardized aptitude test for college admissions in the US education system (SAT), the General Recognition Examination (GRE), and many Advanced Placement (AP) exams administered annually in May by the College Board (a non-profit organization specializing in education and standardized testing in secondary and higher education) in the US for students.
Last January, ChatGPT garnered attention for passing prestigious graduate-level exams, such as those from the Wharton School of Business at the University of Pennsylvania, although not with exceptionally high scores.
OpenAI says it spent months using lessons from its experimental program and ChatGPT to improve the system's accuracy and topic-following capabilities.
4. Provide more accurate feedback.
According to OpenAI, compared to its predecessor, GPT-4 can generate longer, more detailed, and more reliable written responses.
GPT-4 can now provide answers of up to 25,000 words, up from around 4,000 words previously, and can offer detailed instructions for even the most unique situations, from how to clean a piranha's fish tank to extract strawberry DNA. One early GPT-4 user said it provided detailed suggestions for flirtatious lines based on a question listed on a dating profile.
5. Streamlining work processes across different industries.
Joshua Browder, CEO of the legal services chatbot DoNotPay, said his company is exploring the use of GPT-4 to create “one-click lawsuits” to sue fraudsters. This is an early indication of the enormous potential of GPT-4 to change the way people work across industries.
"Imagine you receive a call, press a button, the call is transcribed into text, and a 1,000-word complaint is generated. GPT-3.5 isn't good enough, but GPT-4 handles this very well," Joshua Browder added.
Meanwhile, Jake Kozloski, CEO of the dating website Keeper, said his company is using the tool to better match users.
According to analyst Lian Jye Su, we may also see major advances in connected car dashboards, remote diagnostics in healthcare, and other AI applications that were previously impossible.
Limitations of GPT-4
Although OpenAI has made significant improvements to its AI model, GPT-4 still has similar limitations to its predecessors. According to OpenAI, the current GPT-4 is trained on a massive dataset dating back to 2021 and earlier, so it cannot provide the most up-to-date information, rendering some queries and searches useless.
OpenAI stated that GPT-4 can also fall prey to "simple reasoning errors or overconfidence in accepting clearly false claims from users" without verifying the results.
Gartner Research analyst Arun Chandrasekaran argues that this also reflects the reality of many AI models today. “Don’t forget that these AI models aren’t perfect. Sometimes they can generate inaccurate information,” added Arun Chandrasekaran.
Currently, according to OpenAI, GPT-4 users should exercise caution and use it with extreme care, especially "in high-risk environments".