I’ve been using Google search for a long time, like the rest of world ( -china), and my experience is not as horrible as some people suggest, but it definitely feels like I have to battle SEO spammers and Google ADs, just to find what I’m actually looking for 
Based on the search experience in the GPT store, I’m not expecting an OpenAI search engine to be actually useful, but I am hoping for some competition in the search engine space.
LOL
You’re right though, and OpenAI directly competing against google’s core business would definitely make a statement
I’ve had such a hard time using Google that I basically use a combination of before:2023 and appending reddit to the end of the search.
As someone who made an automated blog posting Assistant in just one day it’s too easy to flood the search engine. Mine wasn’t even that good but it still… Worked. If I was a team with a couple weeks and experience with SEO I’m sure it could hit high ranks while providing no actual content.
Pages these days
Then, reading the articles of what Google is doing to (short-term) increase their ads profit by crippling the search is just gross and desperate. The decision-makers are simple fools that obsessed way too much in their measuring sticks.
I basically use ChatGPT w/ Browsing as a Google Search now. It does great. It seems like they’re caching pages in the back now and can return results crazy fast. I mean, it seems like OpenAI is building a RAG database that covers the entirety of the internet.
I don’t agree with what they’re doing. It will drastically change the economy of blogging and hosting websites, but shit, A LOT is drastically changing.
A HUGE benefit of using ChatGPT as a search is the instant continuation of conversation. StackOverflow is a prime example: You find a potential solution, but oh, it’s a year old, the library has changed. You leave a comment: “I have this issue but the new library API doesn’t match the soltuion?”. A response within a couple hours… maybe
It would, but a statement directed at who? (Probably investors)
I think it’s also worth reminding ourselves that Google’s core business isn’t the search engine, but the delivery of advertisment to your eyeballs 
I think the engine space needs a bit of “disruptive business practices” before things actually get better, but I honestly don’t care if it’s Microsoft or OpenAI doing it, and it’s unfortunate that it has to be a:
But the sad reality is, that actually publishing the SERP algorithm would just cause an immediate race to the bottom
That’s the thing!
Google already had all of this! They had potentially the most powerful RAG system. They could have used all the pages they had cached to create something truly incredible, but.
Meanwhile OpenAI and Microsoft are like: “Damn, we can cache the pages for instant LLM web-browsing responses”
They had the perfect recipe for it all, but they didn’t even bother. Google Home could have been huge, but as a endorsed installer it was pathetic how little resources they actually dedicated to it.
Now, they are nothing besides playing keep up. Destroying the actual organic side of their products in the capitalistic endeavor of pumping stocks and keeping investors happy.
From the article I linked:
These emails are a stark example of the monstrous growth-at-all-costs mindset that dominates the tech ecosystem, and if you take one thing away from this newsletter, I want it to be the name Prabhakar Raghavan, and an understanding that there are people responsible for the current state of technology.
These emails — which I encourage you to look up — tell a dramatic story about how Google’s finance and advertising teams, led by Raghavan with the blessing of CEO Sundar Pichai, actively worked to make Google worse to make the company more money. This is what I mean when I talk about the Rot Economy — the illogical, product-destroying mindset that turns the products you love into torturous, frustrating quasi-tools that require you to fight the company’s intentions to get the service you want.
The advertisements GREATLY DEPEND on search. The intermixing of Ads and Search is truly what ruined it. The ads team doing everything they can to push the other organic teams into ruining their experience in the name of $$$.
Without search we would lose finding these ad-sense ridden websites. Without search there wouldn’t be any ads to place in the results. YouTube can be the exception here but it’s also being crippled (happy premium user for years now though, I bet the ads team froths at the mouth because of youtube premium)
I GUARANTEE that there will be a documentary on Google and how badly they screwed up. On paper, they do great but looking around and listening: a lot of people are not using their products anymore and they are being over-shadowed by true, proper ambition. Coincidentally I bet the timeline start of this documentary will be late 2015.
Fair point, and I agree 100%
I have already sold off all my Google stonks, and I recommend others do the same. I did so exactly because I don’t agree with the “growth at all cost” mentality. In my opinion it’s not the right approach to take, when you’ve already reached a state of a defacto monopoly on internet search.
Maybe, but I think there’s a chance that Google will come out victorious, now that they’re getting some actual competition, it might force them to actually be serious about providing a better service to users 
From a developer standpoint, it should also be faster and easier for Google to reverse a regression than for anyone else to build a competing product from scratch.
So what’s the conclusion about gpt2?
Possibilities:
v2 assistants 15%
gpt4.5+ architecture 15%
prank 10%
gpt4 incremental 15%
gpt-2 architecture 5%
other 40% (some new RAG / search approach)
Other and v2 assistants might be the same thing. If they say it’s just a simple test version of v2 assistants, then it goes to that. But if they don’t, and call it their search version than it’s other, even though under the hood it shares a lot of similar architecture.
I vote other. Maybe more “reasoning” and “memory” on the backend via RAG or other methods?
And yeah, OAI’s naming conventions are kinda weird haha…
One of the things we’re trying to define is what does “search” mean.
Does that mean LLM + curated Knowledge bases (RAG)
Or LLM + Bing?
Or is OpenAI creating some kind of new search engine for the backend?
My experience when LLM does searching has been very underwhelming. I’d rather just do it myself. There are serious problems I’ve found with this:
If by search they mean LLM + very powerful and compelling curated knowledge bases though, that sounds very exciting. Some incredible things to be done there.
I can’t remember what it was exactly at the moment, but I recall reading something recently about new stuff OpenAI wants to do on the back-end - ie reasoning of some sort and memory.
I think the search offering they have is separate from GPT2-chatbot? Maybe they’re related, though?
Is it Amazon that’s backing the other big AI search engine play?
I think for the reasoning they either have more layers, a different set-up, or maybe run through a first model to “plan/reason” then a second model to give the output. We can do that now with two-steps, but it would be good and bad for them to take control and do that stuff on the back-end.
I’m betting it’s this. They are storing all the information onto Bing and then using it as an ultra powerful RAG database.
That made sense, but what do you mean by storing information onto Bing?
My thought (hope) is that Bing isn’t involved except perhaps as a tool for staff to find knowledge bases well before hand to curate. And they might also be using Google 
Bing is a human search engine, I think, and not one optimized for LLM retrieval.
Though that might be a cool product. Search engine api optimized for LLMs.
Amazing technological breakthrough in machine learning algorithms propel AI cognition by magnitudes. We merely demonstrate it using GPT-2 from 2019. Wait until you see these revolutionary advances on GPT-5!
/jk
More likely just getting community eval on gpt-4-2024-04-09, optimistically with some attention layers turned back on and some auto-corpus-to-context RAG. Until that free GPT-4 was discovered by Reddit.
I wonder if GPT2 refers to a GPT with a two-shot on the back end?
Enough hype, I look like an idiot telling my colleagues that something big is coming any day now (for the past 5 months 
)
I’ve noticed that web browsing sometimes returns instant results without even (seemingly) querying, loading, parsing, and returning the page.
So I’m thinking that Microsoft (Bing) is caching pages (which Google has stopped doing), and with OpenAI have created a badass ultra RAG database.
Yeah, they both have a TON of data… Data is the new gold…
The lack of fanboyism in this forum is quite disappointing. 
im-a-good-gpt2-chatbot and im-also-a-good-gpt2-chatbot is up on lmsys as per Sama’s tweet. You need to use battle arena and click new round. Usually get it on my second or third try.
I’ve been comparing it against gpt4 and it is absolutely an upgrade. How the upgrade is done, whether it’s MoE with assistants v2, more self reflection, or something else I can’t say. But the results are usually superior to gpt4.
