Good morning!
Today, OpenAI announced an update to their (pre-existing) search feature on ChatGPT. It's now rolling out for all users (paid and free users), but you need to have an account. You can search like you would on Google Search but then get a summary of the relevant sites, hopefully containing what you need.
It will also soon be available through the advanced voice mode feature. When talking to it, you can tell ChatGPT to search the web for up-to-date information, such as events around you.
That being said…
Plus (and pro) subscribers have had this feature for a while now, and frankly, it has yet to be ready to replace Google Search or even Perplexity.
The search function can be helpful for long queries when you don’t have the exact word but more of a general idea or up-to-date information. We usually use it to enforce ChatGPT to leverage existing content rather than its internal knowledge. On the other hand, Google Search works best with keywords.
TL;DR: If you know what you want and are searching for specific keywords, Google Search should be the first option. For long descriptions (some products you have seen a while ago but don’t remember the name), searching on ChatGPT is a good place to start.
OpenAI has published its Chrome extension (which has been live for a little while now), and it does the same thing as the new web search feature, but it is on the top search bar, replacing Google, and it’s not good. The application always routes you to the ChatGPT application. So, if you want to go on Netflix to watch your favourite shows, you type Netflix in the search usual search bar, and I open ChatGPT, run the search, and display the results. You then have to click on the URL and go to the website. It’s just not worth it and adds extra steps. They say in the demo that the links are now appearing faster and on top of the rest of the generation, which is a good idea, but you still have to pass through ChatGPT every time you search on Google. There are over 1M users for this extension, and the ratings are… not so excellent.
We tried some requests with the search and got not-so-good results. For example, we asked for a restaurant near the area with a score of 4.9 and higher. The results were that all restaurants were in the range of 4.5 and lower. The problem with big models is that they are still a bit dumb with numbers.
In the demo, they also asked what happened during the week of 23 December 2024, and the chat responded with some past events.
We kind of got the idea with the “web browser”, but it does not count in the bingo.
See you tomorrow for day 9!