In our SPOT Spotlight Series writer, we discuss our members of our society by their workshop in science and Ai, write, and their sources of inspiration. Today, we are happy to share our conversations with them Maurua Di Pietro.
Mauro is a data scientist and a 10-contained Creator of experience in the bank industry throughout Europe and Asia. You studied a lot of money but taught about the program after graduation, her passion for writing tutorials violating the complex simple topics and involving explanations.
He wrote a impressive series of creating agents AI from the beginning using Python and Ollama. What motivates you to avoid tools such as Openai Apis or Paid Clouds?
I like to do my stuff, and I'm a great “Open-Source person.”
I appear in the first time of the machine study, where data scientists were used to train their models. I have a very nostalgic in those days, when “all you need” was not AwarenessBut a little dataset, Scikit-learnThe power is limited to Computing is not enough to make good separation. I remember partly a data testing, as I was very beautiful in planning. Today, we all use ChatgtAnd I've never trained the model for age … so I prefer to build from the beginning wherever I can.
Besides, I work in the bank and usually manage very sensitive data. Open design tools from scratch is a better choice, rather than leaning from the cloud services, where you want to invest in control and customize. You have full ownership on top of your infrastructure, avoid the seller lock, and maintain the privacy and security of data. And most importantly, it's free. So, as long as I can choose, I will always choose the “open / from scratch”.
About “from scratch”
I believe you're really learning when you try to do things yourself. Growth is rarely finding things for the first time.
In real use cases, it has never been running as planned, so one should know the gap between theory and practice. Satisfactory between the two, it is important to treat the idea as a variable than a solid framework. The vision provides models that work in good conditions, but the real world conditions come noisily, uncertainty, as well as issues (as a budget, time and behavior. Finally, it is in a grave in the midwor and make wise ideas produce a real value. Therefore, in order to deal with the hardships of the real world, you first need to be able to know the examples of education.
But not just AI: working in all … life is the process of trial and error. We repent of experiences: We try, we fail, to convert, and try again. That is a person's reading (and a machine).
Examining one agent, many agents, and chain-based of buildings. How does your agent's opinion give you up as you develop these kinds?
In the meantime, single agents is a way of travel and closest to the production. In particular, single agents are better than a variety of agents where the operational case domain is properly defined and can be treated successfully with one point of control. They are easy to design, test, and keep.
On the other hand, many agents Systems import additional complexities in the process of making decisions, which may be unnecessary or inconsistent.
Many agents that include the system, which is difficult to control, and the quality of the output is affected. Let us remember that any result from the machine's learning model you must always confirm.
Therefore, unless the work does not benefit from the spreading cliff, I recommend developing single agents first.
How do you live up to date and inspire when working with the common tools and ways in the border of AI research and development?
Oh, it's the hardest part, as I'm a very lazy person. Which calls me to stay up to date with the industry mixing curious, passion and form … I don't want to be left behind!
Like any other registrar, I learn a lot, especially to see the new trends up. In addition, I work together at all times and society to understand how some people come to the same problems. For example, many of my students touched me on LinkedIn begging for help in my topics. I always try to understand their charges for using, discussing the meeting can be the best way, and sometimes new ideas increase.
Establishment which often appears with disciplinary crooked disciplinary action by feedback from peers from peers and users. Therefore, I will say the best way to stay inspired to talk to people.
Then, if you have found that the inspiration flow like a gasoline, you really are “so far”, you need to digest with the handles (meaning, contributions, contributions to open projects, open prototypes).
Looking forward, what kind of problems or programs are most fun to build, or see some buildings, using AGents?
I see agents “Baby Ai”. With the NLP of modern NLP and computer view, it is very close to all the original AI ingredients.
When I was a child, at the 90s, the whole family found a computer at all household family members with it. Yes, I believe it will happen again. Soon, each family will have AI connected to all devices (calls, house, car …). Finally, robots will find on the Hardware side, and that AI helper will be a robot that always dreams.
According to me, I am very happy with AI to remove people from the daily activities. I can't wait to see my robots sending emails, the reservation appointment, and editing my day agenda, when I could cook me because the “originally” never died!).
To learn more about Mauro's work and stay up to date with her latest articles, follow him here on TDS and LinkedIn.
