Tech

GPT-4: What The Most Advanced Language Model Can Do

GPT-4 is the latest AI model with advanced capabilities for tasks like information retrieval and its sources. Here's what you can accomplish with this AI tool.

GPT-4 is the latest AI model with advanced capabilities for tasks like information retrieval and its sources. Here’s what you can accomplish with this AI tool.

Four months after the release of the GPT-4 language model, its developer, OpenAI, announced that the system will become available for embedding in various applications. At first, selected service developers will be able to use the AI model, but in the future it will be opened to everyone via API (hardware-software interface). However, GPT-4 is already available to all regular users who can try out its capabilities. “RBC Trends” explains what are the advantages of the system and how to get access to it.

What is GPT-4

GPT-4 (Generative Pre-trained Transformer 4) is a large multimodal AI model that can process queries in the form of pictures and text, and then produce textual answers. OpenAI unveiled it in March 2023. GPT-4 performs at the “human level” in a variety of vocational and academic tests, the model’s developers claim. On average, it scores 88% or higher on these tests.

GPT-4 was trained using adversarial testing programs as well as ChatGPT, resulting in improved manageability. The model compared to its predecessor, GPT-3.5, is 82% less likely to respond to requests for forbidden content and 40% more likely to generate correct responses. “The difference becomes apparent when the complexity of the task reaches a sufficient threshold – GPT-4 is more robust, creative and able to handle much finer-grained instructions than GPT-3.5,” states OpenAI. In addition, in a query to the neural network, it is now possible to specify the style of the response, as well as assign a role to it.

However, the model still “hallucinates” and sometimes makes errors in judgment. In one example, the chatbot referred to Elvis Presley as “the son of an actor”.

GPT-4 is capable of summarizing text (Image credits: Openai.com)
GPT-4 is capable of summarizing text (Image credits: Openai.com)

Despite this, in May, Microsoft researchers released a paper titled “Sparks of general artificial intelligence: early experiments with GPT-4.” They stated that GPT-4 shows early signs of AGI (Artificial General Intelligence), meaning that its capabilities are at or above the human level.

GPT-4’s capabilities

Compared to its predecessors, GPT-4 has a number of advantages:

multimodality. The neural network accepts queries in the form of text, images, and videos, and is also able to work with documents that have text with photos, diagrams, or screenshots. At the same time, it has learned to solve more complex problems, including chemical or mathematical problems. GPT-4 gives all the answers in natural language, in the form of program code, formulas, and so on;

GPT-4 can analyze graphs (Image credits: datacamp.com)
GPT-4 can analyze graphs (Image credits: datacamp.com)

The AI recognizes images in the same way. AI recognizes images as accurately as textual information, recognizes their contents and sees details. For example, GPT-4 can be asked what is funny about a particular meme, and the model will explain the essence of the joke;

large memory capacity. Neural network remembers much more context (up to 25 thousand words). This allows you to conduct long dialogs with it, referring to the context;

the ability to play different roles. The author of a query can ask the neural network to act as a text editor or a mobile application developer. The role will determine the style of the GPT-4 response. It has also become easier to model queries to it. In addition, the neural network now knows different language dialects.

How to use GPT-4 in my country

On the ChatGPT website, you can access the latest version with GPT-4 only if you have a paid subscription. However, there are workarounds that allow you to try out the language model even for those who have never used ChatGPT.

Bing AI tool
Bing AI tool

“Microsoft’s smart search engine Bing

Microsoft’s Bing is a search engine with a built-in chatbot based on GPT-4, which has been adapted for searching the Internet. In it you can dialog, write code, generate pictures, and the chats themselves can be exported in various formats. In addition, when generating answers, the model provides links to sources, which is convenient for checking this information.

The search engine is available in most countries, but you will need a VPN to use it if that doesn’t work in your country. To enable Microsoft’s Bing chatbot, you need to:

  1. Install the latest version of Microsoft Edge browser;
  2. Enable VPN (Hola VPN, Browsec or other);
  3. In the Microsoft account settings, change the region (choose USA or another country);
  4. Log in to your Microsoft account on Bing.com or in the app and go to the “Chat” section; or just visit this URL: https://www.bing.com/new
  5. Select the mode of operation: creative, balanced, or accurate.

Since the bot works more like a search assistant, to get a more detailed answer, you need to ask it to clarify information. There is also a “Create” tab in the chatbot, where there are templates for business letters, blog ideas, etc., but it is only available in preview mode for now.

Chat-bot in Telegram

You can also try out GPT-4 capabilities in ChatGPT-bots in Telegram. The advantage of this method is that it does not require registration or additional settings, and the bots themselves work for free.

Examples of using GPT-4

The RBC Trends editorial team tried out the GPT-4 capabilities, and here’s what we got.

Describe a brief history of discoveries in physics in the last ten years.

— Here is a brief history of discoveries in physics over the last ten years:

In 2015, gravitational waves – fluctuations in space-time predicted by Albert Einstein in 1916 – were first discovered. They occur when massive objects such as black holes or neutron stars collide. This discovery opened up a new way to “hear” the cosmos and study gravity and the origin of the elements.

In 2012, the Higgs boson, an elementary particle that explains why other particles have mass, was discovered. Its existence was predicted in 1964 as part of the Standard Model of particle physics, but finding it took decades and required the use of the Large Hadron Collider, the world’s most powerful particle gas pedal.

In 2016, quantum superconductivity at room temperature – a phenomenon in which a material has no resistance to electric current – was created. Previously, this phenomenon had only been observed at very low temperatures close to absolute zero. Quantum superconductivity could have revolutionary implications for energy, transportation and computing.

In 2018, the first photo of a black hole – an object with such strong gravity that not even light can escape from it – was taken. It required combining data from eight radio telescopes around the world to create a virtual telescope the size of Earth. The photo showed the black hole’s shadow against the radiation it bends.

In 2020, a new state of matter was discovered: the topological insulator. This is a material that conducts electricity only through its surface, not its volume. Such materials may have unique properties such as the quantum Hall effect or the quantum spin effect. They may also be useful for creating quantum computers and other devices.

Generate a top of the ten most popular songs in the world based on search data from the last five years.

— Here are the top ten most popular songs in the world based on search data from the last five years:

  • Shape of You – Ed Sheeran;
  • Despacito – Luis Fonsi and Daddy Yankee feat Justin Bieber;
  • Blinding Lights – The Weeknd;
  • Dance Monkey by Tones and I;
  • Old Town Road – Lil Nas X feat Billy Ray Cyrus;
  • Bad Guy – Billie Eilish;
  • Rockstar – Post Malone feat 21 Savage;
  • Someone You Loved – Lewis Capaldi;
  • Senorita – Shawn Mendes feat Camila Cabello;
  • Sunflower – Post Malone feat Swae Lee.

What tools should a digital artist in 2023 possess?

— A digital artist in 2023 should be proficient in different tools that help them create quality and original illustrations for different purposes and platforms. Here are some of them:

Adobe Photoshop: it is one of the most popular and powerful programs for editing and creating bitmap graphics. It offers a wide range of tools, filters, brushes and effects that allow the artist to realize any idea. Photoshop also supports working with layers, masks, vectors and 3D objects. However, Photoshop can be difficult for beginners and requires high system requirements.

Vectornator: it is a free app for creating vector graphics on iPad, iPhone and Mac. It has an intuitive interface that adapts to different devices and supports Apple Pencil. Vectornator allows the artist to create logos, icons, illustrations, layouts and other vector images using different tools such as pencil, brush, shapes, text, etc. Vectornator also supports importing and exporting files in different formats including SVG, PDF, PNG and AI.

Autodesk Sketchbook: is a professional drawing and sketching app for a variety of devices, including Windows, Mac, Android, and iOS. It has a simple and easy-to-use interface that does not distract from the creative process. Autodesk Sketchbook gives the artist access to over 190 brushes that mimic different drawing materials and techniques. The application also has tools for working with perspective, symmetry, gradients, etc. Autodesk Sketchbook supports working with layers, masks, blending modes, and other raster graphics features.

Related Articles