Bill Gates had a dinner recently with the team behind OpenAI before the release of GTP4. He said the last time he had a demo like that was when he saw the first graphical interface at Xerox Parc. That demo led him to pivot Microsoft to produce Microsoft Office (see the full interview here). Today he sees the same promise with AI. That is very high praise from a man that has been at the forefront of some of the most important moments in technology.
In this two-part post, I will show you some mindblowing AI use cases that companies have come up with for GPT4. In Part 2, I will share some publically traded companies that are set to benefit from this revolution.
Before that, I have three developments that we haven’t seen that I am excited about:
Apple - Has been deploying neural engines on its devices for years. iPhones have onboard neural-specific chips (M series) that can do very advanced processing without the need to go to the cloud. They have tonnes of data and a user who is always overwhelmed and could do with help. I am sure Apple will produce something very interesting.
Email & Messaging - There have been a raft of AI email writing plugins produced, with mixed resutls. I would love to see a Large Language Model (LLM) models put to use on my emails by Google. They would have deep context, our voice and also some technical understanding. This can really help drive efficiency into daily work. If done right, this will be another “mind-blown” moment in AI.
Company-specific LLMs - models trained on individual (private) company data that is a hybrid public/private LLM. This would mean that a company’s knowledge/processes are at your fingertips teased out by promotes and not in some database / doc in someone's folder that nobody ever uses.
Onto the demos!
Adobe Firefly
Adobe has been at the forefront of using AI in its software suite since 2019 when it launched Sensei. Leveraging machine learning, Adobe was able to achieve some “wow” moments with video and still images. Today they released Adobe Firefly; watch the video below for some truly remarkable feats that will cut hours from the creative process and open up Adobe to a whole new generation of creators.
Text 2 Music
Lots of AI-powered music websites are being launched; trained on massive libraries of music, they can take text prompts and create some astonishing music. Purists say it lacks soul, but honestly, I think it’s better than most things I hear on charts these days because I am old and grumpy.
ChatGPT Plugins
ChatGPT became a platform overnight by allowing plugins from data-rich websites. You can easily imagine a feature that will be available very soon where you can search flights, book a restaurant, find out the best food to eat, book a cab and create a full itinerary with just a few chat prompts. This really reminds me of WeChat in China, which has become an operating system upon which other “mini programs” run. Similarly, OpenAI wants companies to connect their data and allow users to stay in the chat and connect to companies and perform actions without leaving the chat.
Canva AI
Canva didn’t want to be left out of the action. They also announced a raft of AI tools that do everything from text to image to writing a sentence and creating an entire presentation from scratch! Very impressive what Melanie and her team have managed to pull together in such a short time.
Github CopilotX
I vaguely remember being very young when programmers would need to write low-level code to get a computer to do anything, even the most basic tasks. Over time we went from generalised software to super specialised software. But all these programs are always guessing the user’s use case and building for that. Imagine a future where a prompt is basically a human language-powered computer programme. That can store, manipulate and display data in any form you want. CopilotX is a big leap in this direction, giving existing programmers superpowers and new programmers a turbo-charged engine to create.
Generative Roblox
Games are going to be a huge area where “text to X” will produce amazing results. Robolox’s 200mln users are already creating complex maps and environments for others to play in. However, the work can be tedious and require some technical knowledge. Now, with just text, you can create objects, maps and entire levels. Text-to-game is going to democratise the creation of games.
Nvidia
It is often said that in a new industry, buy the guys that make the picks and shovels. Nvidia is that company. Their AI chips have a very strong lead amongst competitors, but more interestingly, their software platform that provides the communication layer, called Cuda, has developed specific programs and processes that help large language models be trained and deployed. This makes it even harder for companies to move away from the Nvidia ecosystem even if they wanted to.
Runway Gen2
Runway is one of the most advanced text-to-video companies in the market. Their latest version, which allows users to type in prompts and generate video, is astounding. This will dramatically cut the production cost of content which is at all-time highs right now with all the OTT majors fighting over subscribers’ attention.
DoctorGPT
Not that it is recommended but, GPT4 also saved a dog’s life (link)