The latest release of OpenAI, GPT-4, is the model for artificial intelligence most powerful and impressive yet from the company behind ChatGPT and DALL-E.
Now available for some users ChatGPTGPT-4 has been trained on a massive cloud supercomputing network linking thousands of GPUs, custom designed and built in conjunction with Microsoft Azure.
The company has unveiled the powers of the language model on its blog saying that it is more creative and collaborative than ever. While ChatGPT with technology GPT-3.5 only accepted text input, GPT-4 can also use images to generate captions and analysis. But that’s just the tip of the iceberg.
That is why in this report you will learn what is GPT and of course what is GPT-4, what are the improvements with its previous version and how to access it.
What is GPT-4? Is it the same as ChatGPT?
To properly explain what GPT-4 is, It is necessary first of all to explain what GPT is. Generative Pre-trained Transformers (GPT) or translated as Generative Pre-trained Transformers are a type of deep learning model that is used to generate human-like text. Common uses include:
- Answer questions
- make summaries
- Translate text to other languages
- Generate code
- Create blog posts, scripts, stories, conversations, and other types of content
There are endless applications for GPT models, and you can even fit them on specific data to create even better results and that’s where GPT-4 comes in.
This is the fourth version of these models developed by OpenAI, the most powerful artificial intelligence company today. This is a great multimodal model and it was announced on March 14, 2023.
Multimodal models can encompass more than just text: GPT-4 also accepts images as input. Meanwhile, GPT-3 and GPT-3.5 only operated in one mode, text, which meant that users could only ask questions by typing them.
In addition to the new ability to process images, OpenAI says that GPT-4 will also “exhibits human-level performance in various academic and professional benchmarks”.
Of course, if you wonder if it’s the same as the famous ChatGPT, not exactly. If ChatGPT is the car, then GPT-4 is the engine: a powerful overall technology that can be adapted to many different uses.
What makes this model better than the previous version?
Now works with images too
The most notable change to GPT-4 is that it is multimodal, allowing it to understand more than one modality of information. ChatGPT’s GPT-3 and GPT-3.5 were limited to text input and output, meaning they could only read and write. However, GPT-4 may receive images and ask you to understand this information.
For example, in the OpenAI report showing the tool, GPT-4 was shown an image with ingredients and asked what recipes could be made with them. At first he responded with the elements that appear in the image, that is, he recognized them and then he began to give a multitude of recipes.
Parameters: more is not synonymous with better
Regarding the differences with the previous models, one that must be emphasized from the beginning is based on the concept of “More power on a smaller scale”. OpenAI, as usual, is very cautious when offering all the information and parameters used to train GPT-4 in this case.
At the moment it is known that GPT-3 has 175,000 million parameters and it is estimated that GPT-4 will exceed it but not with a huge difference as has circulated on social networks.
Regarding the assumptions that many media make about the parameters with which GPT-4 is trained, comment that OpenAI has already confirmed that this information is being saved in order to protect itself from competition. In other words, it prefers not to say it to prevent other companies like Google, which also develop large language models, from having a reference to beat.
With tools like ChatGPT It has already been shown that the number of parameters is not everythingbut architecture, and quality also play an important role in training.
Smaller language models such as Gopher (280 billion parameters) and Chinchilla (70 billion parameters) have already shown that they can keep up with larger natural language models such as Megatron-Turing (which holds the title of the largest neural network, with 530,000 million parameters).
It’s harder to fool
One of the biggest drawbacks of generative models like ChatGPT and Bing is their propensity to get off track from time to time, generating toxic information that alarms people. They can also confuse the facts and create misinformation.
OpenAI says it spent 6 months training GPT-4 using lessons from his “adversarial testing program” as well as ChatGPT, which resulted in the “better results of the company in terms of factuality, management capacity and refusal to leave security barriers”.
Can process much more information at once
Large Language Models (LLMs) may have been trained on billions of parameters, which means countless amounts of data, but there are limits to the amount of information they can process in one conversation.
ChatGPT’s GPT-3.5 model could handle 4,096 tokens or around 8,000 words, but GPT-4 beats those numbers up to 32,768 tokens or around 64,000 words.
This increase means that while ChatGPT could process 8,000 words at a time before it started to lose track of things, GPT-4 can maintain its integrity during much longer conversations. It can also process long documents and generate large format content, something that was much more limited in GPT-3.5.
OpenAI admits that GPT-4 has similar limitations to previous versionsstating that it is still not completely reliable and makes reasoning errors.
However, GPT-4 significantly reduces hallucinations (inventions) relative to previous models and scores 40% higher than GPT-3.5 on factuality assessments. It will be much harder to trick GPT-4 into producing unwanted results, such as hate speech and misinformation.
While English is still its first language, GPT-4 takes another big step forward with its multilingual capabilities. It is almost as accurate in Mandarin, Japanese, African, Indonesian, Russian, and other languages as it is in its native language. In fact, it is more accurate in Punjabi, Thai, Arabic, Welsh and Urdu than the English version 3.5.
Therefore, it is truly international and your apparent understanding of concepts, combined with outstanding communication skills, could make you a truly next-level translation tool.
It is already being used in conventional products
As part of the GPT-4 announcement, OpenAI shared several organizations that are currently already using the model.
These include an AI tutor feature developed by Kahn Academy that is meant to help students with assignments and provide teachers with lesson ideas, and an integration with Duolingo that promises a similar interactive learning experience.
Duolingo’s offering is called Duolingo Max, and it adds two new features. One will give you a “simple explanation” as to why your answer to an exercise was correct or incorrect, and will allow you to ask for other examples or clarification.
The other is a “role-play” mode that lets you practice using a language in different scenarios, like ordering coffee in French or making plans to go hiking in Spanish (currently, those are the only two languages available for the game). function).
Intercom, on the other hand, recently announced that it is updating its customer support bot using the model, promising that the system will connect to a company’s support documents to answer questions, while payment processor Stripe is using the system. to answer questions from employees.
Finally, currently GPT-4’s imaging capabilities can only be accessed through a single app: Be My Eyes Virtual Volunteer. This app for blind and partially sighted people allows them to take photos of the world around them and request useful information from GPT.
How to get access to GPT-4
Open AI has yet to make GPT-4’s image input capabilities available via any platform because the research firm is collaborating with a single partner to get started. However, there are ways to access the text input capabilities of GPT-4.
The best way to access is with a subscription to ChatGPT Plus, which grants subscribers access to the model at the price of $20 per month. However, even through this subscription, there will be a user limit, which means that you may not be able to access it whenever you want, something to consider before making the investment.
Of course, there is a free way to access the text capacity of GPT-4 and that is by using Bing Chat. On the day that OpenAI introduced GPT-4, Microsoft shared that its own chatbot had been running on a slightly older version. Light of GPT-4 since its launch five weeks ago. It is free to use, but requires registration through a waiting list.