This is just my opinion, but I feel like the hype around GenAI is making us forget about traditional machine learning.
These days, it seems like developers want to use LLMs for almost everything. Classification, predictions, anomaly detection, data analysis… somehow, the solution always ends up being an LLM.
But what happened to clustering, regression, decision trees, random forests, and all the other ML algorithms?
Personally, I think we've become so fascinated by what LLMs can do that we're sometimes overlooking simpler, more efficient approaches. If I want to segment customers, why not start with clustering? If I want to predict sales or prices, why not explore regression? If I want to detect fraudulent transactions, why not consider traditional classification models combined with rule-based systems?
Of course, LLMs have their place, and I'm not arguing against them. They're incredibly powerful. But I don't think every problem needs generative AI.
What concerns me is that we're slowly shifting from asking "What's the best approach to solving this problem?" to asking "How can I solve this using an LLM?"
Maybe I'm wrong, but I feel like we're paying so much attention to what's new that we're starting to overlook what already works.
Good engineering isn't about using the trendiest technology. It's about understanding the problem and choosing the right tool for the job.
Sometimes that tool is an LLM. Sometimes it's a traditional ML algorithm. And sometimes, it's just a few well-written rules.