Last Updated on
OpenAI CEO Sam Altman confirmed the release of ChatGPT’s current most powerful LLM, called GPT-4, in March of 2023. The new large language model joins prior flagship AI model GPT-3.5, which is now the default for free users. Since then, both of these neural networks have seen multiple updates, and even a new iterative version accessible via the ChatGPT API (GPT-4 API). The launch of OpenAI’s most powerful AI chatbot poses a significant challenge for competitors Microsoft Bing Chat and Google Bard. But which artificial intelligence came out first? What was the GPT-4 release date, and what has changed since?
What can GPT-4 do?
Let’s have a look at what GPT-4 can do. The first major feature we need to cover is its multimodal capabilities. As of the GPT-4V(ision) update, as detailed on the OpenAI website, ChatGPT can now access image inputs and produce image outputs. It can even solve CAPTCHA and ReCAPTCHA codes as a result! This update is now rolled out to all ChatGPT Plus and ChatGPT Enterprise users (users with a paid subscription to ChatGPT).
DALL-E 3 (Stylized DALL·E 3), which powers the visual functionality of ChatGPT, has immediately become a top 3 AI image generator alongside Midjourney. The combination of text-to-text generation (Natural Language Processing) as well as text-to-image generation (AI image generator) makes it a multimodal AI model. So, what is a multimodal model? Well, this refers to a model that can operate using multiple mediums such as text, image, and sound. This technology could open up a wide range of opportunities for its users and will completely revolutionize the way we interact with OpenAI-powered tools.
The GPT-4 neural network can now browse the web via “Browse with Bing”! This feature harnesses the Bing Search Engine and give the Open AI chatbot knowledge of events outside of its training data, via internet access.
GPT-4 has already been shown to outperform GPT-3.5 when it comes to answering exam questions written for humans. Most notably, the new model achieved a score that sits in the 90th percentile for the Uniform Bar Exam. Pretty impressive stuff, when we compare it to GPT-3.5’s very low, 10th percentile score.
Besides the confirmed features there are still a few rumors circulating around the number of parameters this new model has. Of course, Twitter users have something to say about GPT-4. One user claims that the model will be built using 100 trillion parameters. For reference GPT-3 was built using 175 billion parameters.
Although, OpenAI CEO Sam Altman was quick to deny this rumor in an interview with StrictlyVC. He claims the idea is completely “ridiculous”. Despite this, each new model from the AI research and development firm has historically improved upon its predecessor by an order or magnitude. As a result, GPT-5 — slated for 2024 — is sure to shock the world.
Essential AI Tools
Winston AI detector
Originality AI detector
ChatGPT-4 release date
Many people online are confused about the naming of OpenAI’s language models. To clarify, ChatGPT is an AI chatbot, whereas GPT-4 is a large language model (LLM). The former is a public interface, the website or mobile app where you type your text prompt. The latter is a technology, which you don’t interface with directly, and instead powers the former behind-the-scenes. Developers can interface ‘directly’ with GPT-4, but only via the Open API (which includes a GPT-3 API, GPT-3.5 Turbo API, and GPT-4 API).
Now that is cleared up, we can answer the question. Strictly speaking, there will not be a Chat GPT-4 in the future. However, GPT-4 is implemented into ChatGPT, as the default model for paid users.
If you’re a fan of OpenAI’s latest and most powerful language model, GPT-3.5, you’ll be happy to hear that GPT-4 has already arrived.
To keep up to date with the latest around OpenAI’s new language model, be sure to read our other articles here.
- NOW READ What is GPT 4? What can it do?