Duarte O.Carmo, machine learning consultant based in Copenhagen, shares his honest take on AI tools, local models, and the reality behind the hype. Here's what you'll learn from this episode:
1/ 🎯 Don't throw LLMs at every problem. Duarte explains why binary classification problems don't need ChatGPT when logistic regression works better. Learn his framework for choosing the right tool - whether it's traditional ML or LLMs - based on what you're actually trying to solve.
2/ 🔧 Real AI coding tools comparison. From Neo Vim to Cursor, local models to Claude 3.5 Sonnet - get the unfiltered truth about what actually works. Duarte runs models on his MacBook Pro M3 and shares when local beats cloud, plus why agent mode isn't always better.
3/ 📚 Documentation still beats AI for learning. Despite trying MCP servers and LLMs.txt files, Duarte found nothing replaces 35 minutes reading good docs. Discover why libraries like Polars trip up LLMs and how to learn new tech effectively in the AI age.
4/ 🧠 The mentality shift in AI-assisted development. It's not about writing code anymore - it's about solving problems one level up. Learn how to focus on architectural decisions and business logic while letting AI handle implementation details.
5/ ⚖️ When AI hype meets production reality. Duarte shares real consulting experiences where companies want to use LLMs for everything. Get practical advice on evaluating AI solutions, building proper datasets, and why looking at your data still matters most.
Connect with Duarte O.Carmo:
Connect with me:
Check out our awesome sponsor, zerobang, your data and AI partner: www.zerobang.dev
