Old habits are hard to change sometimes. Time and again, regardless of what the task or query for information might be, my go-to source of information is Google. Whether on my phone or computer, my muscle memory instinctively opens a browser window or clicks into the search widget and starts typing. My “google foo” has been honed and developed over more 32 years (yes, before Google existed) starting with Gopher on a DEC 5000. Regardless of the search engine – Altavista, Yahoo, Ask Jeeves, DuckDuckGo, etc. – I’ve learned which kind of results to favor and how to quickly refine the search further. Because I have a pretty decent typing speed, it just always feels more efficient than trying something new.

Maybe you’ve heard of that “something new”, a little tool called ChatGPT and related variants??

In the last week, however, I’ve been reminded in a couple of ways that my old habits need some tweaking. When my instinct is to start typing into a search bar, I’m trying to do a quick analysis if traditional search is more efficient for the task at hand than asking a Large Language Model (LLM).

• Hours of a local business? Google.
• Definition of a Postgres or database term? Google (generally)
• Finding alternative approaches to writing a query? ChatGPT/Copilot
• Converting a series of values or text into another form? ChatGPT/Copilot
• Asking for an analysis of the fantasy baseball team I help manage? ChatGPT/Copilot
• Cooking dinner, hands messy, and wondering what certain additives are in some of the food ingredients? Copilot voice conversation… which was way more natural and realistic than I expected it to feel.

At the same time, I had a friend message me this week to complain about some step-by-step instructions he had received from ChatGPT regarding the administration of some networking equipment. It was quickly evident that multiple steps contradicted each other and caused more confusion than help. Hallucinations are, and will continue to be, a problem that LLMs suffer from for the foreseeable future.

There is no magic bullet. Both traditional search and LLMs still require the end user to have enough knowledge or experience to judge a good answer from a “bad” one. The more experience we have in a certain area, the better able we are to refine our searches and interactions to attain a better, more helpful answer. But the last week has reminded me that my first, instinctual approach might not always be the best one. Although LLMs have their own problems too, I have found that the ability to converse in more long-form conversation often draws out questions and details that I might not have gotten using traditional search.

There is a space for both technologies, and personally, I think there will be for a long time to come. But I’m glad that I keep being pushed to explore my options and learn which technology is best for the job at hand.

Now, about those fantasy baseball trade waivers…