Google beat their original competitors — Altavista, Excite, Yahoo!, and so on — with an incremental improvement, not a new model. At the time (2000-2003), their search was better and faster, certainly, but not several orders of magnitude better.
In fact, Google was something of a reactionary model at the time, by rejecting the push for bloated, captive "web portals" and going back to the simpler, stripped-down user experience that Altavista had succeeded with originally.
Given how bad Google's results have become, I think you can certainly outdo it. Anecdotally, it's amazing how often I find myself reaching for ChapGPT these days to find basic facts that Google can't. I think Google is right to be afraid.
Pagerank (TM) was a Google innovation and the reason Google had much better results than the others and it took many years for the SEO spammers to catch up. That revolution, combined with the philosophy of the speedy, focused homepage and the elimination of irrelevant results by always using the ‘AND’ operator was why their popularity exploded.
At Voila, a subsidiary of France Telecom, a study around 2005 found that their search engine offered nearly the same answers as Google, while not in exactly the same order.
Despite this objective comparison, in users' view Google's results were much much better.
I remember a German CS professor telling me eyes in eyes with obviously a lot of schadenfreud "Google will kill you".
And I answered "No they will become as us, big, fat and ugly".
> At Voila, a subsidiary of France Telecom, a study around 2005 found that their search engine offered nearly the same answers as Google, while not in exactly the same order.
Order is a pretty big part of tbe value prop of a search engine, so ”nearly the same” results but “not in exactly the same order” is not as much similarity as you seem to be implying.
Come on, there's no absolute correctness for a result in being in Google's first versus second answer. The actual order depends a lot on the user's context.
Even Google doesn't change the order of answers based on the user's context. Otherwise it wouldn't say in Google Web Console "your site appears in nth position"
> Come on, there’s no absolute correctness for a result in being in Google’s first versus second answer.
“Absolute correctness” is an irrelevant standard. Order can affect satisfaction without "absolute correctness" being a thing that can even be assessed meaningfully.
> Even Google doesn’t change the order of answers based on the user’s context.
Yes, it does. [0]
> Otherwise it wouldn’t say in Google Web Console “your site appears in nth position”
I think that reflects the experience of a no-context user (e.g., not logged in or otherwise tied to a history to provide context.)
Jeff Dean even posted on G+ about how he essentially independently invented a variant of pagerank at DEC, but his supposed collaborator couldn't figure out how to open a tar archive, and that was the proximate cause of his leaving DEC.
Google is the initial assignee and a patent provides exclusionary rights to the inventor of a technology. Great if someone else thought of it earlier but no one published anything using it or thought to get a patent, because otherwise this very important patent would have been void and it wasn’t.
It wasn't just incremental. It was a massive improvement! I still remember how floored I was with the quality of results compared to AltaVista which was my main search engine at the time
Depending on the day of the week, I might be doing a standard enterprisey dev project or I might be doing a “DevOps” project automating some things around AWS (where I work in ProServe) using the AWS SDK.
Before ChatGPT, I would spend a lot of time searching on Google for the right API call and the response structure:
Problem: I need a simple Python script that list roles in an AWS account that has at least one of a list of command line specified policies.
Before ChatGPT: I would look up the api call to list the roles (the API service area of AWS is huge), then look up the API call to list the policies for the role and then look up the API call for the “paginator” that pages through all of the results since the API call only returns 50 results at a time.
After ChatGPT:
“Write a Python script that allows me to specify a list of AWS IAM roles that contain at least one of a given list of policies”.
I get “working” code and then I start polishing it.
“Use argparse with required parameters and use a paginator to iterate through the roles”
Again expected results.
“I need the list of ARNs printed out as a comma separated list”.
I copy the code into VSCode, test it, and I’m done.
Later on for another project, the customer prefers JavaScript. I copy the code into ChatGPT, and tell it to convert it into JavaScript. It works perfectly and is idiomatic JS along with proper async/await handling
Lol dude I just googled “script list IAM roles that contain policy” and this was the top result. Probably less typing than the usual silly authoritative tone people use when composing an “AI” prompt , like they’re Tony Stark or Captain Picard .
And trust me, once u google stack overflow enough pretty soon you won’t need to ask anyone how to output a csv
that’s true. But that code snippet didn’t meet all the requirements. I laid out. And it definitely was less typing using ChatGPT once you add all of the requirements I laid out.
And then there is the context switch. This is just one little script as part of a larger system where I’m designing an entire system. It’s much easier if I’m already writing out a long set of documentation, diagrams, slides, etc just to open ChatGPT and keep writing English.
Not to mention when you’re juggling a couple of projects and one is Node and one is Python.
And just in case you are trying to gatekeep, I’ve been programming for a few years - let’s just say my first hobbyist code was in 65C02 assembly language in 1986
I agree completely. Hilarious example. But, hell, if a product's pitch is "you'll feel like Tony Stark or Captain Picard" it will probably be successful.
The entire thesis is that ChatGPT can give better results than Google.
ChatGPT gave me a specific answer with the command line arguments that I specified and allowed me to specify a list of policies and output the result in the format I needed
Google gave me a code snippet that I had to modify for my use case.
In fact, Google was something of a reactionary model at the time, by rejecting the push for bloated, captive "web portals" and going back to the simpler, stripped-down user experience that Altavista had succeeded with originally.
Given how bad Google's results have become, I think you can certainly outdo it. Anecdotally, it's amazing how often I find myself reaching for ChapGPT these days to find basic facts that Google can't. I think Google is right to be afraid.