Feeling inspired to write your first TDS post before the end of 2024? We\u2019re always open to contributions from new\u00a0authors<\/a>.<\/p><\/blockquote>\n
And just like that, 2024 is (almost) in the books. It was a year of exciting transitions\u200a\u2014\u200aboth for the TDS team<\/a> and, in many meaningful ways, for the data science, machine learning, and AI communities at large. We\u2019d like to thank all of you\u2014readers, authors, and followers\u2014for your support, and for keeping us busy and engaged with your excellent contributions and comments.<\/p>\n
Unlike in 2023, when a single event (ChatGPT\u2019s launch just weeks before the beginning of the year) stopped everyone in their tracks and shaped conversations for months on end, this year we experienced a more cumulative and fragmented sense of transformation. Practitioners across industry and academia experimented with new tools and worked hard to find innovative ways to benefit from the rapid rise of LLMs; at the same time, they also had to navigate a challenging job market and a world where AI\u2019s footprint inches ever closer to their own everyday workflows.<\/p>\n
Photo by Oskars Sylwan<\/a> on\u00a0Unsplash<\/a><\/figcaption><\/figure>\n To help you make sense of these developments, we published more than 3,500 articles this past year, including hundreds from first-time contributors. Our authors have an incredible knack for injecting their unique perspective into any topic they cover\u2014from big questions and timely topics to more focused technical challenges\u2014and we\u2019re proud of every post we published in\u00a02024.<\/p>\n
Within this massive creative output, some articles manage to resonate particularly well with our readers, and we\u2019re dedicating our final Variable edition to these: our most-read, -discussed, and -shared posts of the year. As you might expect, they cover a lot<\/em> of ground, so we\u2019ve decided to arrange them following the major themes we\u2019ve detected this year: learning and building from scratch, RAG and AI agents, career growth, and breakthroughs and innovation.<\/p>\n
We hope you enjoy exploring our 2024 highlights, and we wish you a relaxing end of the year\u200a\u2014\u200asee you in\u00a0January!<\/p>\n
Learning and Building from\u00a0Scratch<\/h4>\n
The most reliably popular type of TDS post is the one that teaches readers how to do or study something interesting and productive on their own, and with minimal prerequisites. This year is no exception\u2014our three most-read articles of 2024 fall under this category.<\/p>\n
\n
- \n5 AI Projects You Can Build This Weekend (with Python)<\/strong><\/a>
<\/strong>From beginner-friendly to advanced project ideas, Shaw Talebi<\/a> demonstrates that anyone can get hands-on with\u00a0AI.<\/li>\n- \nUnderstanding LLMs from Scratch Using Middle School Math<\/strong><\/a>
<\/strong>How do LLMs work? Rohit Patel<\/a> offered one of the most accessible and engaging explainers you\u2019ll ever find on this\u00a0topic.<\/li>\n- \nHow to Learn AI on Your Own (A Self-Study Guide)<\/strong><\/a>
<\/strong>For the self-starters out there, Thu Vu<\/a> put together a streamlined roadmap for studying the fundamental building blocks of\u00a0AI.<\/li>\n- \nThe Math Behind Neural Networks<\/strong><\/a>
<\/strong>To understand neural networks, \u201cthe backbone of modern AI,\u201d Cristian Leo<\/a> guides us deep into their underlying mathematical principles.<\/li>\n- \nText Embeddings: Comprehensive Guide<\/strong><\/a>
<\/strong>Embeddings make the magic of LLMs possible, and Mariya Mansurova<\/a>\u2019s thorough introduction makes it clear how and why they\u2019ve become so important.<\/li>\n- \nHow I Studied LLMs in Two Weeks: A Comprehensive Roadmap<\/strong><\/a>
<\/strong>Another excellent learning resource came to us from Hesam Sheikh<\/a>, who walked us through an intensive\u2014but accessible\u2014 curriculum to master the basics (and then some) of large language\u00a0models.<\/li>\n<\/ul>\nRAG and AI\u00a0Agents<\/h4>\n
Once the initial excitement surrounding LLMs settled (a bit), data and ML professionals realized that these powerful models aren\u2019t all that useful out of the box. Retrieval-augmented generation and agentic AI rose to prominence in the past year as the two leading approaches that bridge the gap between the models\u2019 potential and real-world value; they also ended up being our most covered technical topics in recent\u00a0months.<\/p>\n
\n
- \nIntro to LLM Agents with LangChain: When RAG Is Not Enough<\/strong><\/a>
<\/strong>Back in March\u2014and quite ahead of the curve\u2014Alex Honchar<\/a> published the definitive beginners\u2019 guide to working with\u00a0agents.<\/li>\n- \nUsing LangChain ReAct Agents for Answering Multi-Hop Questions in RAG Systems<\/strong><\/a>
<\/strong>Showing us how agents and RAG can complement each other, Dr. Varshita Sher<\/a>\u2019s tutorial addresses the common need of answering complex queries on internal documents.<\/li>\n- \n17 (Advanced) RAG Techniques to Turn Your LLM App Prototype into a Production-Ready Solution<\/strong><\/a>
<\/strong>Building a rudimentary RAG pipeline is one thing; optimizing it so that it can actually work in a business context is another. Dominik Polzer<\/a> put together a comprehensive guide to the methods you can leverage to achieving that lofty\u00a0goal.<\/li>\n- \n12 RAG Pain Points and Proposed Solutions<\/strong><\/a>
<\/strong>On a similar troubleshooting beat, Wenqi Glantz<\/a> outlines a dozen streamlined approaches for tackling some of the most common challenges practitioners face when implementing RAG.<\/li>\n- \nChoosing Between LLM Agent Frameworks<\/strong><\/a>
<\/strong>It can be tough to make informed choices in an ecosystem where both major and emerging players release new tools every day. Aparna Dhinakaran<\/a> is here to help with sharp insights on the tradeoffs to keep in\u00a0mind.<\/li>\n<\/ul>\nCareer Growth<\/h4>\n
Data science and machine learning career paths continue to evolve, and the need to adapt to this changing terrain can generate nontrivial amounts of stress for many professionals, whether they\u2019re deep into their career or are just starting out. We love publishing personal reflections on this topic when they also offer readers pragmatic advice\u2014here are four that stood out to us (and to our readers).<\/p>\n
\n
- \nWhat 10 Years at Uber, Meta, and Startups Taught Me About Data Analytics<\/strong><\/a>
<\/strong>From the importance of storytelling and business acumen to the limitations of metrics, Torsten Walbaum<\/a> generously consolidated lessons based on a decade of work into actionable insights.<\/li>\n- \nWhat I Learned in my First 3 Months as a Freelance Data Scientist<\/strong><\/a>
<\/strong>Career switches are always tricky, and moving away from the structure of working at a company to the world of self-employment comes with its own set of challenges\u2014and, as CJ Sullivan<\/a> shows, with great opportunities for learning and\u00a0growth.<\/li>\n- \nHow I Became A Data Scientist\u200a\u2014\u200aNo CS Degree, No Bootcamp<\/strong><\/a>
<\/strong>For anyone just taking their first steps in the field, Egor Howell<\/a>\u2019s candid account of his path into data science is a must-read.<\/li>\n- \nI Spent $96k To Become a Data Scientist. Here Are 5 Crucial Lessons All Beginners Must Know<\/strong><\/a>
<\/strong>Offering a different perspective on entering the discipline, Khouloud El Alami<\/a> offers practical tips on managing your data science education so that you set yourself on the right\u00a0path.<\/li>\n<\/ul>\nBreakthroughs and Innovation<\/h4>\n
Staying up-to-date with cutting-edge research and new tools can feel overwhelming at times, which is why we have a particular soft spot for top-notch paper walkthroughs and primers on emerging libraries and models. Here are three such articles that particularly resonated with our audience.<\/p>\n
\n
- \nA New Coefficient of Correlation<\/strong><\/a>
<\/strong>\u201cWhat if you were told there exists a new way to measure the relationship between two variables just like correlation except possibly better\u201d? So starts Tim Sumner<\/a>\u2019s explainer on a groundbreaking 2020\u00a0paper.<\/li>\n- \nIntro to DSPy: Goodbye Prompting, Hello Programming!<\/strong><\/a>
<\/strong>In another exciting year for open-source tools, one of the standout new arrivals was DSPy, which aims to open up LLMs for programmers and make it easier to build modular AI solutions. Leonie Monigatti<\/a>\u2019s hands-on introduction is the perfect place to start exploring its possibilities.<\/li>\n- \nKolmogorov-Arnold Networks: The Latest Advance in Neural Networks, Simply Explained<\/strong><\/a>
<\/strong>KANs, \u201cpromising alternatives of Multi-Layer Perceptrons (MLPs),\u201d made a splashy entrance last spring; Theo Wolf<\/a> made their ramifications and potential benefits for ML practitioners evident with this accessible primer.<\/li>\n<\/ul>\nThank you for supporting the work of our authors in 2024! If writing for TDS is one of your goals for 2025, why not get started now? Don\u2019t hesitate to share your work with\u00a0us<\/a>.<\/p>\n
Until the next Variable, coming your way in the first week of\u00a0January,<\/p>\n
TDS Team<\/p>\n
<\/p>\n
\n2024 Highlights: The AI and Data Science Articles That Made a Splash<\/a> was originally published in Towards Data Science<\/a> on Medium, where people are continuing the conversation by highlighting and responding to this story.<\/p>\n<\/div>\n