Foto van Mario Alberto Tamà
Mario Alberto Tamà

Unveiling AI with Luciano Rossi

Artificial Intelligence is one of the most hyped and fast-evolving technologies of the contemporary era. Today, with the help of our Team Leader and AI expert Luciano Rossi, we are going to better understand what AI is and how it can effectively help companies to better their processes.

Hello Luciano, thank you for being here with us today! 
I hope you will help us to shed some light on the AI topic: 

Hello Mario! 
Yes, I hope that presenting what we have done and what we are currently working on will help to clarify what AI is really about.

Wonderful! 
First of all, I would like to know what you do with AI right now and how AI would be impactful for the companies:  

As you said, there is a lot of hype about AI and a lot of expectations about what AI can deliver.  In reality, LLM (Large Language Models) can greatly help companies with some recurring processes, sales for instance. 
Just to give you an example of that, we played a bit with open AI in one of our most successful projects: Omniscius. The original project was to create an online platform to manage online and offline courses. The core of a course is the content you are going to teach, paired with exam questions. What we actually built and then presented here is a model that would generate or suggest questions for the exam and also for each question present a couple of possible answers. 

The best part is that this model is adjustable via prompt and you can easily define how many questions you want and then the level of difficulty for those questions and also how many right and wrong answers you would accept for each question. The general idea is to define a set of parameters for each model and then the model starts producing content based on the context that is provided. We developed this solution one year ago and now I’m sure that new models in open AI are smarter and can deliver more content with higher reliability.
 
We have to keep in mind that one of the drawbacks is that the models were and are prone to hallucinations. By hallucinations I do not mean a medical condition, what I mean is that the model tends to produce content that is basically nonsense. 

 Even the most advanced models are less prone to this behavior, but they are not immune. 

For instance, I asked chat GPT what MM Guide is and it thought it was a medical marijuana company. Probably somewhere in the data set of the model, MM Guide was related to this online platform that provides guidance on using medical marijuana. 

Bias are also issues that are intrinsic to the model. 
Since they are trained on a data set, they will respond to what they were trained on. 
If that data set goes in a certain direction, then the answers and the content produced will also tend to go in that same direction. 

what you are saying is that the data set is the most important part of an AI model? 
Absolutely. I would say that the success of a model is based on its dataset. It’s not just about the algorithm; there are several algorithms that we know are reliable and well-tested. But if you don’t have reliable, well-handled, and normalized data, these algorithms are useless.

Another limitation here is using free content from the internet to train a model. While it’s the widest asset of data in the world, it’s far from being the most precise. For instance, OpenAI used over one million hours of YouTube videos to train their models. To put this number in perspective, that is more than a century!

But you know, on YouTube, we can also find a lot of trash, to put it mildly, and that influences the overall process.

how can a company get a good data set?

Well, that’s an interesting question.

Nowadays, there are what we could call “public models,” trained on a more or less public dataset and provided as a service. This is how OpenAI works.

But of course, you can also fine-tune models using your own dataset to make them more accurate for your context.

Imagine a company that develops video games.

You want to focus on gaming context, and you already have a lot of information about how players play their games and what kind of games they want. This information is perfect to create a dataset for the models to produce valuable data. What you want to do next is use that dataset that you have created and then put it into a model. The model will use that dataset repeatedly, refining itself and becoming more precise in terms of your context.

Another example is an HR company with a huge database of CVs. What they would like to do is feed these CVs to the data models and then use the resulting dataset to ask questions to the model to evaluate candidates or classify them.

As I said, it all comes down to the quality of the dataset.

This is quite clear! 
Is there a proper way to ask these questions? 

As you might know, we use prompts to interact with these models. We strive to be as detailed as possible in the prompt description; this helps the model produce the best results.

You should always keep in mind that when you prompt something, you might be sending sensitive information because you have to provide context to the model.

These are concerns that might arise for companies when developing these types of solutions. On the other hand, there are already regulations and compliance measures in place.

For instance, there has been an AI act in the EU since April 2021 that is expanding day by day. One of the main topics of debate is the use of copyrighted data to train models.

What about AI agents?  
They also seem to be quite interesting elements of this new technology: 

Agents are tools that use AI to automate tasks for you. The best part is that they can communicate with each other to accomplish even more complex tasks.

We’ve been working with agents for quite some time now, and one of the best results we’ve had so far was pairing them with a crawler to gather information from the internet. The goal was simple: help companies identify potential leads.

We began by building a couple of agents that would connect to our website, search for information, and classify it. Once the information was gathered, it was presented in a more readable and relevant format. With these agents in place, we no longer need to manually search websites for company information or services offered. The agents automate this process, allowing us to scale up operations significantly. Instead of manually searching one website a day, we can analyze hundreds or even thousands of websites daily!

A direct use for MM Guide would be to search different websites to see if companies are looking for professionals or people in specific technologies that we can provide. Similarly, instead of searching for jobs, we could search for products or anything else.

For instance, we could compile a vast dataset of item prices and return not only the cheapest price but also the fastest delivery time or the cheapest delivery costs.

The possibilities are endless.

When should a company consider adopting AI? 

As mentioned earlier, there’s a lot of hype surrounding this technology, and many companies are eager to jump on the AI bandwagon merely to be associated with it. However, I believe that companies with specific needs that align with what these models excel at are ideal candidates. If a company generates content, classifies items, or summarizes content, then AI could be tremendously beneficial.

 

In such cases, it might be more efficient to develop a crawler and pair it with an agent. This setup allows the agent to leverage AI capabilities by interfacing with the crawler. Whether a company needs to classify potential leads, items, or suppliers, AI emerges as the most effective tool.

It’s worth noting that AI is not a one-size-fits-all solution. It’s crucial to assess each situation individually and determine where AI can truly add value.

What next?

The value AI brings to businesses is undeniable: from optimizing processes and enhancing content creation to automating tasks and improving decision-making. However, the key takeaway is the critical importance of high-quality, relevant datasets in developing effective AI models. Companies need to carefully consider their specific needs and the quality of their data when adopting AI solutions. With the right approach and tools, AI can be a game-changer, driving innovation and efficiency in ways previously unimaginable. As we continue to advance in this exciting field, staying informed and adaptable will be essential to fully leverage the benefits of AI.

If you feel that your company might benefit from AI, Luciano and the entire MM Guide team will be glad to share their insights and expertise with you, helping you to better understand the profound impact AI can have on the future of your company.

More to explore

From manual to automated: how software solutions are revolutionizing SMEs 

In the dynamic landscape of contemporary business, a significant transformation is underway within small and medium-sized enterprises (SMEs). This shift centers on the adoption of software automation, a catalyst for evolving traditional operational paradigms. This article explores the transition from manual processes to automated systems, highlighting how tailor-made software solutions are revolutionizing SMEs, enhancing efficiency, and driving innovation.

Privacy Preferences
When you visit our website, it may store information through your browser from specific services, usually in form of cookies. Here you can change your privacy preferences. Please note that blocking some types of cookies may impact your experience on our website and the services we offer.