July 24, 2025

Samaira Mody 

7th Grade

New York 

In recent times, AI has been used more commonly by more and more people. One popular AI that is a part of the OpenAI company is called ChatGPT-4o. It’s an AI chatbot that uses OpenAI’s GPT (Generative Pre-trained Transformer) model. This model was released on May 13, 2024. OpenAI was founded in December 2015 and is headquartered in San Francisco. ChatGPT-4o is the flagship model of OpenAI LLM. LLM means Large Language Model, and being a Large Language Model means that ChatGPT-4o can adapt to speaking in a way that humans can (in other words, it can generate a human way of speaking). It can also generate datasets of text, sentence structures, and an understanding of the real world. The “o” in 4o means “omni,” which is “a reference to the model's multiple modalities for text, vision and audio” (www.techtarget.com). ChatGPT-4o can be useful to help find answers to everyday problems, but in a way that humans can understand.

ChatGPT-4o is built on a technology called Transformer Architecture. Transformer Architecture can be simplified into five steps, which are tokenization, embedding, positional encoding, transformer block, and softmax. Tokenization is a dataset of  “all the words, punctuation signs, etc.” (Medium.com). Once you type in your question or input, ChatGPT-4o takes each individual word and all the prefixes and suffixes, as well as the punctuation, and matches it to a token in the AI’s library. The next step in Transformer Architecture is embedding. Embedding is when ChatGPT-4o takes the tokenized input and turns those words into numbers. Next is positional encoding, which is the step where the AI puts all the numbers in order. The next step is the transformer block. The whole purpose of each transformer block is to predict the next word in a sentence. 

The transformer block is made up of two parts: the feed-forward and attention components. The feed-forward component helps the AI to find patterns between the tokens that add to the deeper meaning of the input. The attention component of the transformer block is used to see which part of the input is most important while processing the information. All of this, therefore, helps the AI to respond to the input. The final step is the softmax, which helps the AI to create an output. As an example, when asking ChatGPT-4o to write a story, the AI will turn the words into probabilities, where the highest probability will go first, and so on. So when the AI is writing a response, “once” would have the highest probability, “upon” would have the second highest probability, etc. This is a process that just helps ChatGPT-4o to organize its response back to us.

By now, it is clear that ChatGPT-4o has a lot to offer, so let's dig a little deeper into what it can do. According to godaddy.com, ChatGPT-4o has the “ability to process audio, visuals, and text simultaneously.”. ChatGPT-4o can also help businesses by serving as an “assistant,” whether by analyzing data, making images, or just giving helpful information. ChatGPT-4o has had a major upgrade in terms of how it generates images. It can now create the same character in many different styles (such as making the same AI-generated character sit, stand, eat, etc.), add stylish text to posters, and even do spectacular image editing. ChatGPT-4o is also accessible for small businesses because of its affordable price range. 

Another perk of having ChatGPT-4o is that it has more than 50 languages to choose from, which makes it easier to communicate with others. “You can now take a picture of a menu in a different language and talk to GPT‑4o to translate it, learn about the food's history and significance, and get recommendations” (openai.com). ChatGPT-4o can also make products that will help you make money, such as custom photos, t-shirt designs, fonts, etc. ChatGPT-4o is also faster, more efficient, and more accurate than ChatGPT-4, and its development is a sign of how quickly our world is moving toward becoming fully automated.