Artificial intelligence models are causing “storms” recently. GPT-3, released in May 2020, is OpenAI’s new artificial intelligence (AI) development engine. Its updated version, GPT-4, released just a few days ago, is expected to have a lot more processing power than GPT-3. Let’s AnonyViet discover GPT-4 What’s the standout feature?

What is GPT-4?

GPT-4 is the latest version of the GPT series created by OpenAI. GPT-4 uses machine learning techniques to create human-like text and is considered one of the most modern language models available today.

Like its GPT-3 sibling, GPT-4 also possesses natural language processing (NLP) functions, the ability to perform many tasks such as creating a story, giving instructions in the form of sentences, etc. question – answer, writing, writing poetry, ….

Survey to assess the ability of GPT-4the CEO of Open AI said that it can score 1,410 on the SAT, scoring 4/5 or 5/5 on the advanced exam in the subjects of history, biology, art, chemistry and analytics, beat nearly 90% of the participants to pass the bar exam to become a lawyer, it also scored perfectly on at least 34 tests in various areas such as macroeconomics, math…

What new features does GPT-4 have?

Input information can be text, image, video, audio

GPT-4 is a comprehensive upgrade and is much more efficient than GPT-3. GPT-4 is a “multimodal” operating model, which means it is capable of handling many types of input information such as images, speech, text, and video. This is a much more optimal improvement than its younger brother, GPT3 – which can only handle text.

Example: You can upload a photo and ask GPT-4 “With the ingredients in the picture, what can I make?”

GPT-4 capable of identification and analysis based on the provided images. From there, give the answer “smarter” like a human.

For example with this picture, I asked it, “What if the rope breaks? GPT-4 replied, “The balls will fly away”. Pretty smart isn’t it!

Or a little quiz, you can ask GPT-4 about any meme picture

“Is this picture funny?”

– GPT-4: “This picture is funny because it is a combination of two unrelated things: Earth and fried chicken wings”

Multimodal intelligence is an essential aspect of AI technology, helping machines not only read (through text) but also understand (via video), see (through images). This helps GPT4 process information in a more complete, human-like way. From there, it is possible to give precise answers to highly complex questions.

Provide more accurate and in-depth information

OpenAI says it spent six months making GPT-4 more secure and accurate. According to the report, GPT-4 is 82% less likely to respond to content that OpenAI doesn’t allow and 60% less likely to fabricate content than its predecessor.

One of the ways to do this, is to enhance RLHF (Reinforcement Learning from Human Feedback) – that is the process of training the model by human feedback through the Like, Dislike icons that we see. when chatting with ChatGPT.

This helps information that GPT4 make fewer mistakes, make more accurate decisions, and in language conversations are “more human-like”

Handles documents up to 25,000 words

GPT-4 can analyze, process, generate text data volume of up to 25,000 words, 8 times more than GPT-3 with only 3,000 words.

Multi-language support

GPT-4 Flexible operation with many different languages. For example, you can ask it in Vietnamese and ask it to answer in French or any other language. Really a useful tool, especially for global business models, which work in many countries.

Application of GPT-4

Job GPT4 The birth will have important implications for the financial analysis, healthcare, marketing, etc., helping users find accurate and fast information. It can be information about products, services, education, health, etc.

For example, GPT4 can be used to diagnose health conditions through symptom images. or suggest marketing ideas based on customers’ browsing history, ..

Businesses can take advantage of turning GPT-4 into a “virtual assistant”, helping to process work quickly

GPT-4 is a tool available as an API, so programmers can use it to create useful applications, the most recent example being Duolingo or Poe.

Helping users communicate with each other easily, breaking down language and cultural barriers.

Support data analysis, report writing, thematic content creation

Epilogue

What do you think of these new features of GPT-4? Is it enough to “kick the rice bowl” of content creators or programmers? It is still too early to answer this question. Let’s wait for more detailed information of this GPT-4 version