OpenAI working on web search product

“Google” is now a common verb meaning “look it up on the internet”

That’s a lot of inertia to fight against as a new search player.

I bet it’ll be an api that can power smaller scale, specific search products (and also bing)

Small correction: as recently as last March, the CEO & Cofounder of Neeva, which I had never heard of either until listening, was on the No Priors AI podcast promoting Neeva, which had launched less than 2 years prior. So not ancient history. The podcast is a fascinating way to learn that you should really never believe a founder who is giving off an aura of success and growth as their company stunned everyone by announcing their shutdown around 30 days later, which obviously indicates that the CEO had been having internal discussions about how they were probably dead in the water at the same time he went on a podcast acting very positively. Which, I would probably do too. It's arguably in the job description. No judgement. But it was interesting enough to really stick with me.

You don't disrupt a thing by doing exactly the same. So, I don't think it's likely that OpenAI is looking to build a website that gives you ten or so results and then you click one with maybe some ads on the side.

What they are more likely to do is to use the web for RAG type approaches and use natural language to help you find what you are looking for in a chat (or voice) dialog that maybe involves repeated and increasingly more fine grained results that match what you are looking for.

People do this manually with Google currently and it's time consuming. I remember doing this with Yahoo, Altavista, and all the other things that were around before Google came around to find academic papers for my research. Google made that a lot easier. But the next level of that might involve a lot less skimming through long lists of results, opening lots of websites, etc.

Let me add that more often than not, it's also absolutely useless when you DO know the exact thing you're looking for, shoveling literally any content at you at all cost.

> I kind of already use ChatGPT to surface things because Google's become absolutely useless if you don't know the exact thing you're already looking for.

I think Google no longer works even in searches you know the exact thing you're looking for. For example, searches for exact string marches no longer work at all.

https://support.google.com/websearch/thread/3544867/how-do-i...

While it's likely that this is the direction search is headed in, it's also sad because it's going to be yet another layer of abstraction over the world wide web. How long before the concept of websites itself is obsolete?

>give objectively better answers without all the weird SEO and advertising cruft.

i think you forgot a “yet” or a “for now”

> It's a wildly better experience searching for "traditional apple pie recipe" on ChatGPT than on Google due to the brutal amount of intrusive ads and over-optimization and the gap is just getting bigger.

So the differentiator is not that AI is inherently better but that the phase 1 of enshittification hasn't hit yet. There is no indication that AI solves enshittification, it's only a matter of time. In that case, why bet on it ? We already know the end result and it's shit.

Your comment seems to indicate that alternative search engines with different strategies might be better, because remaining in the minority naturally filters out the unneeded, and that's a more interesting strategy.

(it also implies that capitalism is a problem in itself and that is something I can only agree with)

> Google can't meaningfully improve their search without killing their profits and growth numbers.

This is a really good point and might mark a dead end for google. Most of their money is made when people click off of search and land on an advertiser's landing page. Almost everybody in our circle here on HN is unsatisfied with most of those pages they land on because those pages are designed to win clicks, not inform people.

If it turns out that people want answers more than they want search results then google is arguably stuck between a rock and a hard place.[1]

Content creators are in for a hard time too though. Already in Kagi I can ask a question and get a really good answer based on five or six sources without having to view any of those sources. If that's the future then those sources are going to be missing out on a ton of search traffic. Maybe it'll wipe SERPs off the face of the earth, but it doesn't bode well for search traffic for real content operations.

[1] Maybe we should prepare ourselves for "sponsored answers" -- LLM-generated word salad that still somehow tries to sell you a product.

MS is working hard to try to be the sole beneficiary of AI. They are using their power to favor their AI products above everyone else with their Copilot key, Copilot all over windows, and dark patterns with trying to force users onto Edge.

We need new anti trust laws that can be applied more freely just based on size of the company. And higher tax rates for very high profits. And aggressive breaking up of ones like MS. Until these things are done, their abuse will continue.

As moves by Netflix and most other streaming services indicate, ads and the surveillance-tech underpinning them are never going away.

You can pay and still be the product. Paying for ad-free Netflix doesn't exclude your account from telemetry collection or from stitching together other non-Netflix behavioural data that can be freely collected from data brokers.

I love perplexity, but I can see there days being more numbered.

I heard Sam talking in an interview where he mentioned that OpenAI would end up killing a lot of startups building on top of GPT, if they were just adding missing functionality and it's only the startups that are doing something much more unique that would make sense. I feel like this will apply to Perplexity at some point.

Sadly it's just the same with anything, much like a startup making an app that fills a need missing in an OS and then in the next OS update Microsoft/Apple add that functionality killing the startup/app overnight.

ChatGPT already has browsing functions, data cut off points are a big issue for some things, I love that I can talk to perplexity about things that happened that day and get the details, for example the mysterious gp2-chatbot that appeared, I asked perpexity about it the day that it happened and it gave me all the info in about 30 seconds that contained a bunch of different insights from several different sources that would have taken me much longer to find manually.

Would you mind sharing the user agent string/ips. Thank you.

You can't beat google.com with a paid product though. Even on a free tier, search.chatgpt.com is a harder sell to the layman, or anything with a "chatgpt" brand. Even if the results are better, it won't be easy to go against their moat.

Google also has a paid plan for Gemini Ultra. And there isn't anything particularly special about ChatGPT. Meta, Anthropic, and Google are rapidly catching up to GPT 4.