The mosaic model can be a novel approach to creating AI.

   The mosaic model can be a novel approach to creating AI. 

The regular model for creating AI is the large language model, or LLM. That means that. Developers must have control over large data masses. And large code structures. All large code structures are complicated. Because they involve a lot of code. And there are always some kind of errors in the code. That means, when the number of code lines increases, there are more possibilities to make errors. Making the AI is like making a statue. When the system turns more sophisticated. And more advanced, the development process turns slower. When we want to make statues, the first steps are always fast.

 But when we are closing the goal, finishing details takes more and more time. The finishing process just before the statue or product is ready for delivery to customers. Takes most of the time in the development process. The same way. Finishing and detecting errors from the data structure takes most of the time in the programming process. The problem is this: the LLM requires many event handlers. And each event handler is an independent algorithm. That means LLM is a large number of algorithms that operate as a whole. 

The problem is that the LLM rises like dough. That means there are more and more algorithms that must operate as one system. And when the code mass. And the data mass that the system operates on is both large. There can be lots of errors. The AI is a tool that can detect errors in other programs. But the system is not very effective if it must search for errors in its own code. The outsider algorithm can make error detection. AI needs a model. That it can compare its code to. With the code model. Actually, the outsider AI makes the comparison process with two other language models. 

And the problem is: when a new AI model comes. There are no models that the developers can compare their code to. Without models, it is impossible to compare even billions of code lines, which can involve minimal errors. But when there are billions of code lines, and let’s say. There is an error in 1000 lines; searching those codes is a long-term process. That requires carefulness. 

The modular, mosaic model helps expand the AI. 

The mosaic structure or the morphing neural network. That can involve multiple independently operating small language models. It can make it possible to make a flexible and elastic system. Every brick in the mosaic structure is an independently operating server that runs an independent small language model, SLM. Each SLM is a module that involves a certain skill. If the system doesn’t need a module. It can put the server that runs the module into sleeping mode. That makes it more energy efficient.  When developers want to give a new skill to the AI. They can make a new SLM and then connect. That thing to the entirety. 

Researchers say that the future belongs to small language models, SLMs. SLMs are lighter; they involve less code, which makes the development process easier to handle. The SLMs can also operate as a whole. This means: each SLM can operate independently. But those things can also form the virtual entirety. That entirety is like a mosaic. Each SLM has certain skills. When the user sends signals to the system, the SLM knows. If it has that skill or capacity. If the SLM doesn’t have a match. That system forwards the message to another language model, which has the database. That involves data about the SLMs and their skills. And the system sends that information to SLM, which can make the job. 

The idea is that each bite of the mosaic can be written independently. Which means developers can make new modules for AI. And then collect them into  a mosaic. The mosaic-type developing means that. Programmers can control the code better. That also decreases electric use, because if those modules or parts of the mosaic are not required, the system can put its servers to sleep. A mosaic model or morphing neural network structure allows developers to make an elastic and flexible system.