Does it still make sense to use Google’s search engine? The answer is becoming increasingly complicated, especially after OpenAI’s announcement that ChatGPT Search is now available for free to all users.
Additionally, there’s a Google Chrome extension that allows you to set it as your default search engine—which is exactly what I did to see if it could replace Google or other alternatives. The truth? It performs well, sometimes even surprisingly well, but in other cases, not so much.
For the tests, I installed the Google Chrome extension and typed a search term in the address bar. As soon as I did, the browser displayed a message saying it had detected the search engine change and asked if I wanted to undo it. I clicked Keep it to confirm ChatGPT Search as the default search engine in Chrome.
From there, I entered various search terms, just as I would on Google. I began by typing “What’s ChatGPT Search” to see the results.
As shown in the accompanying screenshot, the response was accurate. It was also very current, referencing the launch of ChatGPT Search for both free and paid users. The sources cited included The Verge and the New York Post.
One significant advantage of this alternative to Google is that ChatGPT Search is a “conversational search engine.” When I asked about browser extensions, it quickly recommended four options.
The fourth option is the official OpenAI one. As shown in the image, the results aren’t just text predictions, as can happen when using ChatGPT normally. Instead, it summarizes information ChatGPT Search extracts from web pages, then synthesizes and displays it as a search result.
Still, the search engine is based on the GPT-4 model, which comes with some limitations. The engine acknowledges the risk of “potential inaccuracies.”
The results and texts displayed by ChatGPT Search come from various media outlets and websites, meaning any “inaccuracies” are often due to errors found on the websites accessed by the OpenAI search engine.
So, if the information isn’t reliable, the answer won’t be either. It’s important to be cautious about the sources ChatGPT Search uses to determine how much we can trust the provided answers. But be aware, the search engine can also misinterpret or incorrectly synthesize information, even if the source is reliable.
I tested this by creating a table of soccer world championships and runners-up. While Wikipedia provided the correct data, ChatGPT Search failed to list France as the runner-up in 2022.
Here’s another interesting fact. When I pointed out the error, ChatGPT Search corrected both the data and the table. I tried to mislead it by claiming Spain won the 1930 World Cup. However, despite apologizing for any confusion, it explained that Uruguay defeated Argentina 4-2 in the final of that championship.
The versatility of ChatGPT Search is also evident when solving coding problems. I asked it to write a Python program to count links on a random website, and the answer was straightforward. More importantly, it included the sources it relied on to provide the solution.
Was the program correct? It didn’t work on the first try. The script asked for the website where I wanted to count the links. I tried xataka.com, but the message read: “Invalid URL ‘xataka.com’: No schema provided. Maybe you meant https://www.xataka.com?”
What did I do? Of course, I asked ChatGPT Search. After I explained the error, it provided revised code, explaining the problem (I hadn’t included “http://” or “https://” when typing the URL).
The second program it gave me worked fine and counted the links—500 links to the Xataka homepage, to be exact.
ChatGPT Search can be overly informative for some questions. For example, when I asked for Tom Cruise’s age, in addition to correctly stating his date of birth and age (from Britannica), it also added some extra trivia. While this information may be interesting, it may be unnecessary for a straightforward search.
Interestingly, ChatGPT Search isn’t particularly visual and tends to focus on textual information. However, if you ask for images of Tom Cruise, for example, it will display them. You can click on any image to view it in its original size, with the image source at the bottom.
ChatGPT Search’s behavior here is commendable, though it’s surprising that it doesn’t display visual results more frequently. It prefers to stick to your request, unlike Google, which tries to guess your intent and shows links, images, and videos, when applicable.
ChatGPT Search also surprised me with a final test. Most chatbots and large language models on the market struggle with a simple question: “How many R’s are in the word strawberry?” These models often misinterpret the question due to how they break it down into tokens. When I asked ChatGPT Search, it gave the correct answer but based on a reliable source that addressed the issue accurately.
I wanted to test it further, so I asked how many R’s were in the phrase, “Round and round the rugged rocks the ragged rascal ran.” And here, ChatGPT Search showed the limitations of relying on ChatGPT: It counted 18 letters incorrectly when the correct number was seven.
Check, ChatGPT Search. Your move.
Images | Solen Feyissa (Unsplash) | Xataka
Related | Secret Santa 2024: How to Use ChatGPT, Gemini, and Copilot as an AI Gift Exchange Generator
View 0 comments