Hacker News Reader

Comment by tptacek | Hacker News Reader

So, Google already does this, right? And most of the feedback I've seen is: "how the hell do we turn this off".

OpenAI isn't going to beat Google at the search game any time soon, and yeah Google's AI results have mixed popularity now. Doesn't seem like the best use of OpenAI's focus to me.

Microsoft does it. In fact they came up with Bing integration before GPT-4 release, although UI is a bit different.

Kinda hilarious that nobody remembers...

Bing has been doing this for a while, using GPT-4. I don't see how OpenAI can substantially improve on that experience.

Have you used perplexity? It is very useful, I rarely google stuff any more.

I tried using it, results were often irrelevant to my queries or outdated. Maybe I was using it wrong, but I never got a useful search result out of it.

It would be cool if it could collect and aggregate information.

"What is the mean and standard deviation of the AQI along the current fastest driving route from Palo Alto to Lassen National Park, averaged over the driving time"

"What is the easternmost supermarket before Yosemite that is at least 2000sqft in size"

etc

I think Open Ai can do this better than google for the simple reason that nobody pays for google search, ads do. And with ads the incentive make the search bad. Google has become an advertising monster. If only google could get people to pay for proper search but that is not very likely and problematic since it would cannibalize their ads based shitty search business.

Open AI or other AI companies could capitalize on that because they already have hooked their users up with their LLMs, search could be another feature and has the potential to grow from there.

The problem with the Google one is accuracy. It told people to eat rocks. OpenAI just needs to beat that.

Obvious errors make for good comedy and create obvious moments for reflection on what is actually being done. Subtle errors lead to fractured realities.

Google makes shittons of the subtle variety now as well.

It appears from the demo example, OpenAI is not winning in the accuracy department either: https://x.com/kifleswing/status/1816542216678179083

> In ChatGPT's recent search engine announcement, they ask for "music festivals in Boone North Carolina in august"

> There are five results in the example image in the ChatGPT blog post :

> 1: Festival in Boone ... that ends July 27 ... ChatGPT's dates are when the box office is closed [X] 2: A festival in Swannanoa, two hours away from Boone, closer to Asheville [X] 3. Free Friday night summer concerts at a community center (not a festival but close enough) [O] 4. The website to a local venue [X] 5. A festival that takes place in June, although ChatGPT's summary notes this. [Shrug]

Not saying the SearchGPT will or will not be accurate, but this demo was certainly made in After Effects. Who knows where the copy came from.

Even though placeholder copy is common in promos, a search engine demo is the one case where you need completely accurate copy.

They are both training on the same garbage. OpenAI's accuracy isn't any better.

Reply on news.ycombinator.com