• By quade
• 13 March 2024

How Big Models Are Being Trained to Handle Smaller Programming Jobs

With the advancement of digital programming through AI, there is also a shift in the task requirements. It isn’t enough to handle the big tasks now, businesses now want an AI that can fully automate the process by handling smaller tasks. If AI is to be the future of the workforce, that means fully automating the process.

For that reason, Microsoft has begun developing a tool to handle smaller tasks in your business. This new AI model has been called Task Matrix.AI and is touted as being able to complete many smaller jobs, similar to ChatGPT4. With that in mind, there are hopes that Microsoft can revolutionize the workforce with this new technology and let the AI act similarly to a project manager.

How Does TaskMatrix.AI Work?

Since your business will need AI to perform multiple smaller jobs, the foundation model must be more specialized. It will not be compatible with other standard AI models. Instead of simply modifying and using the programs of other AI models, TaskMatrix.AI is set up with different Application Programming Interfaces or APIs. Together, these different APIs allow the software to access different pools of knowledge and communicate with each other.

That is why the model was programmed with four key components, a conversational foundation model, the API platform, an API selector, and an action executor. Each component serves an important role in making the model functional.

The conversational foundation model helps determine the user input across different mediums such as video, images, and text. This will generate code that the API can act on where the API platform comes in as it contains all the API documentation and its vast repository. With so much data available, it will be essential to search and pull out the important information instantly. That is where the API selector comes in and chooses the API  that fits the task requirement. Here the action executor will turn that code into action and allow the model to accomplish its tasks.

TaskMatrix.AI also uses ChatGPT as its foundation model, supplying it with information to better understand user instructions. They hope that TaskMatrix.AI can include automation, robotics, scrolling researching the internet, and many more. To accomplish this, the model must be programmed differently to serve its unique role. But simply performing tasks isn’t enough for an AI model, the team behind TaskMatrix.AI also believes that the AI has the potential to continue learning and further improve its capabilities.

An AI Assistant?

As AI advances, the development has raised the potential of AI tools functioning as an assistant. This is based on how the AI responds as developers hope to train it to the point where it can give interpretable responses. When it reaches that point, it can provide you feedback based on your work, allowing the AI to assist throughout the working process.

There have already been some promising advances as some teams have shown that TaskMatrix.AI has been used to make PowerPoint slides and process images almost independently, freeing you to work on other things.

The only times you will need to interact with TaskMatrix.AI during that process is by typing the instructions for complex tasks, placing descriptions for what you want to be done, and asking them to make edits if necessary. Hopefully, you will not have to do it often as TaskMatrix.AI can understand human language and intention quite well through text interactions.

TaskMatrix.AI Test Results

This knowledge was seen in a recent test where the model was asked to create a pink flower on a green background. The researchers added a single instruction to extend it to “2048 × 4096”. The model immediately understood the request to change colors and created a colorful flower against green leaves, effortlessly swapping out the APIs.

Another major test the TaskMatrix.AI was given was to create a PowerPoint presentation. The task was to create a set of slides with each one introducing a different tech company.

This can allow it to perform tasks like inserting texts, resizing images, and adjusting slides. Its training in ChatGPT has allowed the model to make small adjustments like that as shown when it managed to inset and resize five company logos after obtaining them online.

While these preliminary tests are promising, the AI team expects more challenges. A model as powerful as this requires plenty of training and maintenance to keep the high-quality AI functioning. More importantly, it requires a large API platform that can handle both the task requirements and any security and customization needs.

At geniusOS, we are eagerly anticipating the results of these tests to see whether or not this AI can live up to standards. If future tests prove this good, we believe this AI can make a huge difference in how we work and allow us to automate more of the programming process. If you have any questions about our working process, you can reach out and contact us here.