OpenAI, the San Francisco-based artificial intelligence company, on Tuesday, released its latest large language model (LLM), GPT-4, the successor to GPT-3, which was released in 2020 and is known for its ability to create human-like text, generate images, and computer code from almost any prompt.
OpenAI’s GPT-4 is a better language model than GPT-3, but not at all. According to OpenAI, GPT-4 has big improvements over its predecessor and can handle images as well as text, and —to be specific — it recently passed the US bar legal exam with results in the ninetieth percentile.
You could be wondering when we reached the 4th level of GPT or probably ask yourself what the initial iterations were and had. GPT-4 is a type of generative pre-trained transformer neural network that can perform various natural language processing tasks such as answering questions, summarizing text, and even generating lines of code.
ChatGPT is an AI chatbot application that uses GPT-3 and GPT-4’s language models that people can interact with. OpenAI describes GPT-4 as excelling at tasks that require advanced reasoning, complex instruction understanding, and more creativity.
Since 2018 when GPT-1 was first published, GPT-3 has made giant progress. The GPT-1 had only 117 million parameters. GPT-2 raised the bar to 1.2 billion parameters, and GPT-3 raised it even further to 175 billion parameters.
To define parameters, these are configuration variables that are internal to the model and whose value can be estimated from the given data. They are required by the model when making predictions, and their values define the skill of the model on your problem.
What’s new about GPT-4
GPT-4 can processes 8x the words of ChatGPT. The maximum number of tokens GPT-3.5-turbo can use in any given query is around 4,000, which translates into a little more than 3,000 words. GPT-4, by comparison, can process about 32,000 tokens, which, according to OpenAI, comes out at around 25,000 words —this allows the chatbot to provide a greater context in its responses, as well as handle larger text inputs. This allows it to summarize entire blog posts or even websites.
Users of GPT-4 have already reported several creative uses of the new model, such as coding entire video games in their browsers in just a few minutes and transforming a handwritten sketch of a website and turn it into a functioning website that successfully ran JavaScript and even generated additional relevant content to fill the site.
One of the biggest highlights is GPT-4 also has multimodal capabilities, that allow the chatbot AI to handle text, images, and eventually even video inputs, though only text and image inputs. Users can upload images and receive recommendations based on the image, for instance, a user uploaded a picture inside the fridge showing groceries and asked the AI to create a recipe based on that.
In a nutshell, GPT-4 performs better than its older siblings. It can answer maths questions better, is tricked into giving false answers less frequently, and can score fairly highly on standardized tests.
To use GPT-4, users need to subscribe to ChatGPT Plus, which costs $20 per month for premium access to the service. Reports indicate that, for starters, GPT-4 has a cap of 100 messages every four hours.
Alternatives are also growing
OpenAI’s GPT-4 is not the only LLM in the market even if it is the most popular AI chatbot right now. They are a few of the other chatbots that could fill in if you can’t get into ChatGPT. Google launched its own chatbot, Bard, in Feb. 2023, which uses a different language model called LaMDA, Microsoft’s AI-powered Bing, and Jasper, among others. Additionally, Huawei, Alibaba, and Baidu are also racing to roll out their own versions amid heated competition to dominate this burgeoning AI sector.
That said, OpenAI’s GPT-4 is poised to revolutionize natural language processing and artificial intelligence. We’ll watch the space to see how it fares.
