But beyond the necessary regulations, it is important to understand that artificial intelligence (AI) is here to stay and that its growth will be exponential.
In our opinion, organizations should adopt AI, but not with the idea of saving costs, but to enhance the capabilities that people already have. And for this it is necessary to train people to learn to use these tools, that will in the future be a commodity, as Excel is today, for example.
Currently, the key is that the teams working inside the companies know how to make the most of these solutions, this will ultimately allow companies to become more competitive.
In other words, organizations must understand that, thanks toGAI tools, their collaborators will be able to do much more. So now is the time to investigate how these solutions can help them add value and enhance the capabilities of the different teams within their company, whether they are marketing specialists or programmers, to name just a few.
The ChatGPT-4 model has numerous aspects that are still being worked on and that bring substantial changes. In this regard, we can highlight five advances:
The expansion of the context window, which will go from the current 4 thousand tokens (about 3 thousand words) of GPT 3, to the 32 thousand tokens of GPT 4. You will be able to have about 25 thousand words, 50 document pages, present simultaneously when you answer something. This will generate an important leap when a lot of contextual information is required to solve a task, such as the comparison of several academic documents or the analysis of large programming projects.
Multimodality: soon ChatGPT-4 will have the ability to process images, which will somehow “give eyes to the interaction”, which until now was only by prompt. This new ability of GPT-4 allows the artificial intelligence to reason about what it sees in images and documents, for example, explaining a diagram or analyzing the result of a certain action on an image.
Intuitive interaction through graphical interfaces: this capability is now being automated and will make it easier to connect with other tools (such as those in Office). This will mark a major shift in how artificial intelligence and traditional software interact and coexist.
Extending GPT’s capabilities with external tools or other AI models: it will be possible to install plugins to solve tasks that GPT does poorly, such as mathematical operations. Microsoft has extended this idea by creating a plugin “marketplace” within GPT chat, allowing the model to connect with external applications, for example, managing the purchase of a flight or booking a restaurant reservation. A recent study entitled “HuggingGPT” explores the idea of further expanding its potential by using GPT as a coordinating brain to access and use different models available in Hugging Face. With access to all these capabilities GPT-4 will be much more powerful than today.
Metacognition: the model will be able to establish a loop in which it analyzes its goal, plans the steps to follow, executes them and, in an iterative way, corrects errors, analyzes its outputs, generates tests, and plans the next steps until completing a goal set via prompt. An example of this approach is the experimental work called “AutoGPT”.
At Baufest we train company personnel and help organizations to understand how they can take advantage of these tools and extract the maximum potential for their business. We can also partner to design their own GAIstrategy.