GPT-4: The Next Frontier of Artificial Intelligence

Artificial intelligence (AI) is one of the most rapidly evolving area of technology today. AI has the potential to transform various aspects of our lives, from entertainment to education to business. Development of AI has come a long way, and one of the most significant advancements in this field is in the development of Natural Language Processing (NLP) models. These models have revolutionized the way machines understand and process human language, and they have many applications across various industries. Understanding and generating rich natural language is not easy since we, humans, tend to be complex, ambiguous and talk in context use references.

GPT-4

One of the most popular NLP models is the Generative Pre-trained Transformer (GPT) series. GPT models are deep neural networks that use unsupervised learning to train on vast amounts of text data. These models can generate human-like responses, answer questions, and even write essays, among other things. The most recent model in the GPT series is the GPT-4, which was released this week. It is more powerful and capable than its predecessors, with more advanced features and capabilities.

GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, shows human-level performance on various professional and academic benchmarks.

GPT-4 has improved its ability to manage dependencies, coherence, and commonsense reasoning. Long-term dependencies refer to how words or phrases in a text relate to each other across long distances. Long-range coherence refers to how sentences or paragraphs in a text form a logical and consistent whole. Commonsense reasoning refers to how a text reflects general knowledge or common sense about the world. GPT-4 can better capture these aspects of natural language than GPT-3.5, because it has more memory capacity, a better attention mechanism, and a deeper understanding of semantics.

Some of features of GPT-4 include:

Enhanced Language and Code Generation GPT-4 generates more coherent, human-like language. GPT-4 has already shown that code generation has been also improved.
Better Understanding of Context GPT-4 is expected to have a better understanding of context, making it easier for it to find the intended meaning of a sentence and supply more correct responses.
Multilingual Support GPT-4 will support multiple languages, making it easier for it to communicate with people who speak different languages.
Multimodality: support for text and image input; at current moment OpenAI only has text-version of GPT-4 open to public use.
Longer Context Relevance. has context relevance for longer texts (even 50 pages). There are 8k (about 10-13 pages of text) and 32k (about 50 pages of text) versions available using OpenAI API (when API becomes available).

Image understanding (multimodality) is a very interesting aspect. GPT-4 not only brings in understanding what is in the picture, but also context and what it means. Just like in the picture below, where GPT-4 can explain and joke from a meme picture. Picture taken from OpenAI GPT-4 article.

Some Use Cases of Conversational AI with GPT-4

Customer Service: used to improve customer service by supplying instant responses to customers’ queries. It can also be used to generate automated emails and chatbots – with some care. It is good to remember that GPT-4 can also make up its answers (OpenAI calls this “hallucinates facts”).
Speech interface. Combining speech to text and text to speech capabilities to GPT-4 can elevate that to next experience level.
Language Translation: With its multilingual support, GPT-4 can be used to improve language translation services. It can accurately translate text from one language to another, making it easier for people who speak different languages to communicate. This is getting closer and closer to Star Trek Universal Translator.
Content Creation: GPT-4 can be used to create content such as news articles, blog posts, and even novels. It can generate original and coherent content, saving time and resources for content creators. Overall Generative AI is used to create already images, videos, 3D objects, Skyboxes and much more.
Image understanding. When this feature comes out, GPT-4 can understand the image with context. This is different from image recognition, which can only pick what is in the image. GPT-4 can understand the big picture. Combined with code generation people have created web sites from a sketch with this. Microsoft Power Apps had the same feature, but it was a specific use case while for GPT-4 this is but one use case. Image understanding will be very usable in so many scenarios that this will require a separate article later. On the bad side, Captcha just became meaningless (unfortunately) – it is time to invent better solutions to replace it.
Summarization: since GPT-4 understands longer context relevancies, it can create better summaries.
Code generation: There has been already examples of GPT-4 generating code for games like Tetris , Pong , Snake and game of life. And even combining couple of games together. You can create a website html by describing it. Another example that was shown that GPT-4 was used to create a Python code that generates 3D object based on text prompt; Python was required step since GPT-4 doesn’t have capability to generate .obj files directly. I wonder what code generation will achieve when CODEX in Azure OpenAI gets updated with these capabilities.
Content Idea generation: GPT-4 can be your personal sparring partner for ideas. You can use the chat to innovate and ideate with the AI and come up with new ideas for content. And of course, ask AI to generate great titles for the content.
Teacher and learning: I have already seen great examples how GPT-4 has been used as a teacher. Embedding AI of this caliber can transform a lot of training experiences. Imagine a virtual teacher that adapts to your skill level and learning style and is relentless.
Personal assistant. With proper integrations to business systems, natural language understanding GPT-4 supplies, takes us even closer to have our personal assistant. I am already using GPT-4 as sparring partner for ideas and helping with innovation.

Bing Chat and GPT-4

Bing Chat was revealed to be one of the first platforms that is using GPT-4. With GPT-4, Bing Chat can supply more correct and relevant responses to users’ queries. It was hinted when Bing Chat was launched, but since there was no official confirmation, it was not sure what model was used. During first week or so Bing Chat was really great (I wrote a post about it How Chat AI is Going to Revolutionize Our Work and Search ), until it was toned down. Now it is regaining capabilities one by one, with more safeguards in place. If you want to experience GPT-4 – just get started with Bing Chat!

Conclusion

GPT-4 is an exciting development in the world of AI. Its advanced features and capabilities are opening doors to improve communication and interaction possibilities. Its potential uses are vast, and we can expect it to be used in various industries soon. When AI is getting better and better, it is also more and more important to think about Responsible AI and what it means. How we use these tools? What is the effect? These tools can boost our efficiency big time, but as with all power tools: safeguards are necessary.

If you want to try out GPT-4 already today, you can use Bing Chat or get ChatGPT+ subscription.

The Future of Work with AI – Microsoft Event 16^th of March 2023

It is time to reinvent productivity again. Microsoft has a public event later today – at 4pm CET / 8 AM PT.

I can imagine, combined from rumors that run wild , that this means bringing AI to Microsoft apps. Word would be one good example of this. I do suspect that GPT-4 will be included on whatever is going to be announced. I will come back to that on a later blog post.

Join the event via LinkedIn!