Within the Writer Highlight sequence, TDS Editors chat with members of our neighborhood about their profession path in knowledge science and AI, their writing, and their sources of inspiration. At the moment, we’re thrilled to share our dialog with Maria Mouschoutzi.
Maria is a Knowledge Analyst and Mission Supervisor with a powerful background in Operations Analysis, Mechanical Engineering, and maritime provide chain optimization. She blends hands-on business expertise with research-driven analytics to develop decision-support instruments, streamline processes, and talk insights throughout technical and non-technical groups.
In “What ‘Thinking’ and ‘Reasoning’ Really Mean in AI and LLMs,” you handle the semantic hole between human and machine reasoning. How does understanding this distinction affect the best way you method mannequin growth and interpretation in your skilled work?
AI has generated enormous hype lately. Rapidly, many old-school ML-based merchandise are immediately rebranded as AI, and there appears to be a renewed demand for something that has AI slapped on it. Due to this, I consider that it’s now important for everybody to have a primary technical understanding of what AI is and the way it works, in order that they’re ready to judge what it may and can’t do for them.
The reality is that we supply plenty of baggage concerning the very nature of AI, originating in narratives from our sci-fi legacy. This baggage makes it straightforward to get carried away by all of AI’s thrilling and promising potential and neglect its precise present capabilities, in the end misjudging it as some form of magic resolution that’s going to alleviate all our issues. Non-technical enterprise customers are essentially the most liable to this overexcitement about AI, typically imagining it as a black-box superintelligence, capable of present right solutions and options to something.
For higher or for worse, this couldn’t be farther from the reality. LLMs — the principle scientific breakthrough all of the AI fuss is de facto about — are impressively good at sure issues (as an illustration, producing emails or summaries), however not so good at different issues (for instance, performing advanced calculations or analysing multilevel trigger and impact relationships).
Having a technical understanding of what AI is and the way it essentially works has immensely helped me in my skilled work. Primarily, it permits me to find legitimate AI use instances and to handle enterprise customers’ expectations of what can and can’t be achieved. On a extra technical degree, it permits me to differentiate the precise parts that must be utilized in particular contexts, in order that the delivered resolution has actual worth for the enterprise.
For instance, if a RAG software is required to go looking particular technical documentation and carry out calculations primarily based on data that’s present in that documentation, it signifies that a code terminal part must be included within the software to carry out the calculations (as an alternative of letting the mannequin straight reply).
The place do you draw the preliminary inspiration in your articles, particularly the extra philosophical ones just like the “Water Cooler Small Discuss” sequence?
The preliminary inspiration for my “Water Cooler Small Discuss” sequence got here from precise discussions I’ve skilled in an workplace, in addition to from mates’ tales. I believe that because of the tendency of individuals to keep away from pointless battle in company setups, typically some actually outrageous opinions may be expressed in informal discussions round a water cooler. And often, nobody calls out incorrect info simply to keep away from battle or problem their colleagues.
Regardless that such conversations are benevolent and well-intended — actually only a informal break from work — they generally result in the perpetuation of incorrect scientific info. Particularly for advanced and not-so-easy-to-intuitively-understand subjects like statistics and AI, we will simply oversimplify issues and perpetuate invalid opinions.
The very first opinion that pushed me to put in writing a whole piece about it was that ‘in the event you play sufficient rounds of roulette, you will ultimately win, as a result of the chances are about 50/50, and the outcomes are going to ultimately steadiness out.’ Now, in the event you’ve ever taken a statistics class, that this isn’t the way it works; however in the event you haven’t had that statistics class, and nobody calls this out, you might depart this dialogue with some unusual concepts about how playing works. So, my preliminary inspiration for that sequence was primarily misunderstood statistics subjects.
Nonetheless, the identical — if no more — misunderstandings apply these days to subjects associated to AI. The massive hype that AI has generated has resulted in folks imagining and spreading every kind of misinformation about how AI works and what it may do, and so they typically achieve this with unbelievable confidence. For this reason it’s so vital to coach ourselves on the basics, irrespective of whether it is statistics, AI, or every other subject.
Are you able to stroll us by your typical writing course of for an in depth technical article, from preliminary analysis to closing draft? How do you steadiness deep technical accuracy with accessibility for a basic viewers?
Each technical publish begins with a technical idea that I wish to write about — as an illustration, demonstrating easy methods to use a particular library or easy methods to construction a sure downside in Python. For instance, in my Pokémon post, the purpose was to clarify easy methods to construction an operations analysis downside in Python. After figuring out this core technical idea that I wish to deal with, my subsequent step is often to seek for an acceptable dataset that can be utilized to display it.
I consider that that is essentially the most difficult and time-consuming half — discovering an excellent, open-source dataset that may be freely used in your evaluation. Whereas there are many datasets on the market, it isn’t so trivial to seek out one that’s freely out there, with full knowledge, and fascinating sufficient to inform an excellent story.
In my opinion, the flavour of the dataset you will use can have a huge impact on the recognition of your publish. Structuring an operations analysis downside utilizing Pokémon sounds rather more enjoyable than utilizing worker shifts (eww!). Total, the dataset ought to thematically match the technical subject I’ve chosen and make for a considerably coherent story.
Having recognized the technical subject of the publish and the dataset I’m going to make use of, I then write the precise code. It is a slightly easy step: write the code utilizing the dataset and get it to run and produce right outcomes.
After I’ve completed the code and I’ve made certain it runs correctly, I begin to draft the precise publish. I often begin my posts with a quick intro on what initially sparked my curiosity on this particular subject (for instance, I wished to make a complex visualization for my PhD, and the searoute Python library made my life simpler), and the way this subject may be helpful to the reader (studying my tutorial explaining API calls to the Pokémon knowledge API may help you perceive easy methods to write calls to any API).
I additionally add some temporary basic explanations, wherever acceptable, of the underlying theoretical premise of the use case I’m demonstrating, in addition to a brief introduction to the code libraries that I will likely be utilizing.
In the principle a part of the technical publish, I usually present easy methods to construction the code with Python snippets, and current step-by-step explanations of how every little thing is enjoying out and the outcomes you might be anticipated to get if every little thing runs accurately.
I additionally like so as to add GIF screenshots demonstrating any interactive diagrams which are integrated within the code — I consider they make the posts much more fascinating, straightforward to know, and visually interesting to the reader.
And there you’ve gotten it! A technical tutorial!
What initially motivated you to start out sharing your information and insights with the broader knowledge science neighborhood, and what does the method of writing give again to your skilled apply?
Again in 2017, whereas writing my diploma thesis, I stumbled upon Medium and the In the direction of Knowledge Science publication for the very first time. After studying a few posts, I bear in mind being fully mesmerized by the abundance of technical materials, the number of subjects, and the creativity of the posts. It felt like an information science neighborhood, with writers of various backgrounds and at totally different technical ranges — there have been articles for each degree and for varied domains.
However other than appreciating the technicality of the tutorials that allowed me to be taught and perceive extra about knowledge science, I additionally preferred the creativity and storytelling of the posts. Not like a GitHub web page or a Stack Overflow reply, there was a sure creativity and artistry in a lot of the posts. I actually loved studying such posts — they helped me be taught a whole lot of stuff about knowledge science and machine studying, and over time, I silently developed the need to additionally write such posts myself.
After eager about it for some time, I reluctantly drafted and submitted my very first publish, and that is how I revealed with TDS for the primary time in early 2023. Since then, I’ve written a number of extra posts for TDS, having fun with every one as a lot as that first publish.
One factor I actually take pleasure in about writing technical items for TDS is sharing issues that I personally discovered difficult to know or particularly fascinating. Typically advanced subjects like operations analysis, possibilities, or AI can really feel scary and intimidating, discouraging folks from even beginning to learn and be taught extra about them — I personally am responsible of this.
By making a simplified, easy, even seemingly enjoyable model of a posh subject, I really feel like I allow folks to start out studying and studying extra about it with a mild, not-so-formal begin and see for themselves that it isn’t so scary in spite of everything.
On the flip aspect, writing has vastly helped me on a private {and professional} degree. My written communication has vastly improved. Over time, it has develop into simpler for me to current advanced, technical subjects in a approach that enterprise non-technical audiences can grasp. Finally, placing your self ready to clarify a subject to another person in easy phrases forces you to fully perceive it and keep away from leaving ambiguous spots.
Trying again at your profession development, what’s a non-technical ability you would like you had centered on earlier?
In an information profession, crucial non-technical ability is communication.
Whereas communication is efficacious in any discipline, it’s particularly crucial in knowledge roles. It’s basically what bridges the hole between advanced technical work and sensible enterprise understanding, and helps make you a well-rounded knowledge skilled.
It’s because, irrespective of how sturdy your technical abilities are, in the event you can not talk the worth of your deliverables to enterprise customers and administration, they received’t take you very far.
It is very important have the ability to clarify the worth of your work to non-technical audiences, communicate their language, perceive what issues to them, and talk your findings in a approach that reveals how your work advantages them.
Knowledge and math, as invaluable as they’re, can typically really feel intimidating or incomprehensible to enterprise customers. With the ability to translate knowledge into significant enterprise insights after which talk these insights successfully is in the end what permits your knowledge evaluation tasks to have an actual affect on an organization.
To be taught extra about Maria’s work and keep up-to-date together with her newest articles, you may comply with her on TDS or LinkedIn.
