Comment by hooande | Hacker News Reader

Comment by hooande | Hacker News Reader

I do not seeing ChatGPT as a replacement for Google Search. Here are my last seven searches (in reverse order):

1. looking up cost of a dental procedure 2. checking the stock price of tesla 3. job postings at Flipper Devices 4. how to push to a remote git repo 5. looking for a machine learning paper by title 6. current information about the fighting in Bakhmut, Ukraine 7. the translation of the russian phrase "bozhe moi"

At most two of my last seven could be answered by ChatGPT. You can do this with your own search history for comparison [0]

The reality is that the majority of my queries are for real time information of some kind. ChatGPT just cannot do this. I've seen no papers or news about an LLM that can update in real time, and you'd think this would be OpenAI's #1 priority. It's not going to be an easy problem to solve without writing a lot of code that isn't related to the core technology of LLM summarization.

The ability of ChatGPT to condense information and inspire creative thought has value. I use it every day. But the market of people who regularly need summaries of concepts or creative inspiration is very small. Especially relative to the multi billion dollar market that Google already has an iron grip on.

Google has not shipped a product based on their LLMs (PaLM, etc) simply because it would not move their bottom line by a significant amount. The majority of queries, especially for non-hn users, are for weather, sports scores, stock prices, movie showtimes and reviews, etc. Most people do not search for code snippets. And if I'm using google for > 50% of my queries, I'm going to keep using google as the default.

Obviously OpenAI can invest some of their billions into connecting to APIs for various information. But it will take them a long time to catch up to what Google has already had for decades. And there is no guarantee of success. I see ChatGPT as ultimately being a compliment to google, not a replacement. Personally I'll continue to use Google as I have been, with ChatGPT as an additional resource.

[0] = you can go to https://myactivity.google.com, click "Filter by date and product" at the top and limit it to just "Search"

Looking at my own search results, the thing that jumps out to me the most is that my search queries are almost entirely topic based and chatGPT assumes I want to talk to it or give it instructions, which aren't really the same thing.

For example my searches are all shorthand: "terraria switch", "spouse", "vampire's kiss", "a way out", "broomball", "wes anderson".

Sure I could have asked ChatGPT "is terraria available for the nintendo switch?", "what is the definition of the word 'spouse'?", "what is the movie 'vampire's kiss' about?", "when was the game 'a way out' released?", "what is broomball, how does it work, and where is it played?", "what movies has wes anderson made?" -- but that's a whole lot more typing and google had the answers I needed directly in my face (or one click away in the case of broomball) without me even asking the actual question. I'm frequently amazed at Google's ability to know what I need without me telling it much (I don't even bother to type anything correctly anymore since I know it can fix even incomprehensible spelling mistakes).

I, on the other hand, switched to ChatGPT on 80% of my Google queries. I still use Google for index searches (ie what's that website's domain?) but not for knowledge searches, which is what I do most of the time when "searching". Google cannot maneuver their legacy business fast enough to compete, nor safe enough not to risk cannibalizing their own revenue stream. And Google search is not just one product anyway, but a portmanteau of shopping assistant, live tickers, calculator, condensed Wikipedia, etc. OpenAI, or the next LLM business, really has a shot as a knowledge query disruptor - and then some. It's the most powerful software sidekick I've ever had, and the idea of an AI sidekick market is a card that trumps that of any index search on steroids.

When you go to the library, sometimes you know the book/author, and that's both the librarian and googles domain. But most of the time you want the knowledge, which libraries are terrible at helping you out. Now we got the sage. Once the sage gives not only the wisdom but the way to the wisdom (citations, links, recommendations) then it's whole new ballgame.

Live data from the web, index searches etc. are all add-ons in this new ballgame. "Plugins", like what chatGPT does with tabular data, will be added with time, making the search business completely obsolete.

Incorporating the tech into how current searches work for the best of both worlds is obviously the future.

After having used ChatGPT, it has become clear that the workflow for discovering complex information in Google is usually just complete garbage in comparison. It's basically a complete dice roll to land on an actual good link these days, and more importantly, Google searches have no context, which is ChatGPT's killer features.

Ex: I asked ChatGPT how ECDSA works, and it will give you a small quip about how it works overall, but then you can ask it go deeper, provide specific examples, ask it what provides the trap door function, how multiply is defined in such a system, etc. You can massage the answer output to give you what you want instead of a vomit of extraneous noise that links to internet forums/blogposts/papers will be able to provide.

Google can't possibly offer the same experience as it is right now.

> I asked ChatGPT how ECDSA works

I’d search for ECDSA, find and read a textbook, or even primary sources. I’d read other peoples code, but I prefer to write my own, or at least re-write it. I’d look for answers on Stack Exchange, but I’d rarely ask my own. I prefer documentation over tribal knowledge. I never fully trust what other people say, because (at best) much can be lost in communication.

Some people prefer to ask others for help, and those people will probably like using AI. Some people prefer to try and figure out stuff for themselves and they are not going to want to rely on AI any more than they’d want to rely on a colleague for answers.

And in the future, managers are going to be looking at the people relying on AI and wondering if they can just cut out the middleman.

This seems to ignore how fast ChatGPT popped up as a refinement of drastically worse precursor code. We're talking "give it a year and it'll leave current ChatGPT in the dust" rather than the previously held notion of "give it a decade and it might be a few percent better".

If you don't see the tech underlying ChatGPT as a replacement for google search, ask yourself why, because the only thing currently holding ChatGPT back as the only search engine you need, is information integrity: is what it's claiming actually true. Give it a year, OpenAI isn't sitting back, pretty sure the next publication will -yet again- blow us all away with how much of an improvement they effected.

well put. and you want context, source, and different perspectives and forms of media for queries like "current fighting in bakhmut" or "cost of dental procedure" - the first answer is likely to be unsatisfying in some way. If it is, you're immediate wondering what the source is, and looking to add context by seeking other source. Not something that's going to be (as) easy with chatGPT. Google actually leads you to the source. That's one of the cool things about it. ChatGPT doesn't do that and seemingly never will.

Surely someone is exploring language models that can be constantly updated in real time?

OpenAI is a non profit funded by people like Sam Altman, I’m not sure they don’t just have billions to invest.

On the other hand, I just used ChatGPT today not to search for how to write 5 relatively simple Python scripts. But go ask it to write them for me and then I copied a sixth one in that I wrote and it gave me a summary of the purpose of them good enough to use for documentation.

http://you.com/chat uses GPT-3 and search, it could answer some of those questions.

Reply on news.ycombinator.com