> even coming out of the gate pretty strong by dominating the browser market

They were out of the gate about as weak as could be, Windows didn't have a native tcp/ip stack for the longest time (remember Trumpet Winsock?) and they only dominated the browser market through grossly uncompetitive behavior after they had lost the initial 5 rounds of the battle.

They definitely used uncompetitive behavior, but it’s also true that IE was also a better browser than Netscape by the time version 4 rolled out.

That's a different definition than 'out of the gate' covers to me. Besides that, from those days I mostly remember IE as the utility to download another browser after a fresh windows install, and the thing that it was nearly impossible to get rid of. Not through any merit of its own, in spite of many non-standards compliant websites that favored IE.

in 1995 people understood very well what internet was going to become. The technology just wasn't there yet, but every kid and parents remember very well the sound of that modem and the phone lines beeing busy.

That memo would have been prescient if made 5 years before.

The infamous Bill Gates and David Letterman interview was in 1995: https://www.youtube.com/watch?v=tgODUgHeT5Y Lots of people definitely didn't understand what a big deal it was going to be then.

In October 1994 a Wired journalist registered mcdonalds.com and then tried to give it to Mcdonalds, but couldn't reach anyone who understood the importance of domain name registration: https://archive.is/tHaea

In my recollection it really wasn't til 1998-2001 or so that people (where I lived in the southern U.S. anyway) really started to take notice.

I passed my high school exam in 1996, in france, and a friend of mine who had internet gave me the topic for the history exam a day before.

His parents weren’t comp-science researcher, he just liked tech, and « had internet ». it was already popular amongst the general public.

On the other side, my mother once worked in a comp science research department, and she once brought me to the lab, where people would create me an email. That was something around 1990. She told me « they’re all crazy with that internet thing ». i never used that email, i didn’t even understand what that was, pretty much nobody in the general audience did. Being able to predict that it would be big at that time maybe would have been prescient, although it was already the consensus amongst people in the field.

> it was already popular amongst the general public.

No it wasn’t, not by the normal definition of “popular”. It was less than 1% of the population.

https://www.internetworldstats.com/emarketing.htm

people talked about it, people knew about it, some non-tech people already were using it. That's what i meant by "popular" (sorry, non-native, so maybe it isn't the correct word).

What i mean is that it wasn't some bleeding edge tech only a few people in the elite knew about. Everybody already knew that was the future.

Could be talking about Mintel [0] which was a simpler, earlier version of the web

[0] https://en.m.wikipedia.org/wiki/Minitel

in 1994-1995 in france, internet was starting to be known from the general public and available to anyone. It was already dubbed as the future in the medias

True but the remaining of the comment is still valid: Microsoft had time in advance to prepare.

But just "internet" doesn't mean a lot. Prescient would have been predicting search, ads and social media. We now are in a similar position maybe, with some tech that looks cool, trying to build geocities with AI

Yeah, I also thought of Eternal September [1] in 1993, when I saw the claim to prescience.

[1] https://en.wikipedia.org/wiki/Eternal_September

Absolutely. I just looked back at netscape wikipedia page, and it was already out in 1995 and distributed freely. Internet explorer was out in 1995 as well. Barely underground stuff.

And that's just for the www. People were using BBS, FTP, email and newsgroup before that.

You're conflating the Internet with the walled gardens that were dominant at the time; AOL, Compuserve, Prodigy etc.

You're off by about 4 years: by 1995, the number of Internet users was many times higher than users of all other networks of computers combined.

There were million of AOL customers in 1995, but most of them used AOL only to access web sites on the internet and send and receive SMTP email.

Getting back to the original topic, by 1995 there were hundreds of mainstream journalists who where predicting in their published output that the internet will quickly become an important part of society. It was the standard opinion among those that had an opinion on the topic.

I disagree with the parent post entirely. Most people in 1995 didn't know what the Internet was going to become. It was a geeky thing that most didn't use. Speeds were slow, most people thought gopher was a small rodent, etc. And for every article saying the Internet was the next big thing, there were many questioning what it would be good for.

Heck, Mosaic wasn't even in development in 1990. It was released in late 1993, and it wasn't until Navigator was released in 1994 that "browsing" became a thing. Most people before then weren't going to use an FTP site off an obscure college to DL something originally intended for X windows...

People forget how fast the Web took off at that point. From 1994 to 1999, the growth was just crazy, with improvements in features every six months.

let's say it was the beginning of an exponentially growing curve. For those that were interested in computers, the writing was on the wall. Science fiction had already written about gigantic networks and virtual worlds for decades, we knew what was coming.

I worked at AOL in 1995. We had three million users at that point, and we had recently upgraded to a live Internet e-mail gateway system. I was the first Internet mail operations person ever hired by AOL. My job on my first day was to install the fourth inbound mail gateway system.

By 1996, we were up to five million users, and Steve Case gave everyone in the company a special company jacket as a bonus. I still have mine, although it hasn't fit me in a couple of decades.

Even as late as 1997 (when I left), most AOL users were still in the "walled garden". Sure, we were the biggest Internet e-mail provider in the world, but that was still just a small fraction of the total AOL users. AIM was much more popular than e-mail. Advances were being made in efficiently distributing AIM chat messages efficiently that would not be exceeded in the outside world until the advent of BitTorrent.

However, by 1997, I think we did have more users coming into AOL over the Internet and their local ISP, as opposed to through our own modem banks. That was in part due to the "unlimited" plans that AOL rolled out, but the telephone calls themselves would have to be paid for if the user dialed non-local POPs for our modem banks, and many of our modem banks were totally overloaded.

AOL's big "hockey stick" moment was in 1995, sure. But the "hockey stick" moment for the Internet was at least a year or two later.

> That memo would have been prescient if made 5 years before.

Not really. I was very active on the internet starting around 1989 and what it was going to become was very clear to the people using it even at that time.

In 1995, I was working for a tech company that literally sent memos in paper folders. It took them until 1999 to adopt email.

> If that wasn't true we'd all be using Microsoft phones (or heck, IBM PCs AND phones).

Once an IBM Office Manager was offering me a job and explained

"IBM is a marketing organization."

So, the focus was not really on computers or phones but on the central, crucial, bet your business data processing of the larger and largest companies -- banking, insurance, manufacturing, ..., and marketing to them.

So, the focus of IBM was really on their target customers. So, if some target customers needed something, then IBM would design, build, and deliver it.

That may still be their focus.

Yeah, as a government buyer, IBM is indistinguishable from the other big integrator/consulting firms (BAH, Deloitte, Lockheed, Mitre to an extent). Literally, whatever we want, they will swear they can build. The challenge is getting the spec right.

Microsoft wanted to control it all with Blackbird and ActivePlatform.

Their greed ended up with them losing out (thankfully).

From memory Bill Gates barely mentioned the internet in his first edition of road ahead in early 95. By late 95 the second edition entire book was revolving around the internet as if he had an epiphany.

In the first edition, he did mention something very much like the Internet, except he called it the "Information Superhighway"

Exactly. Initially, they viewed it as a faster way to download content - rather than an application platform.

I don't quite follow? I'd say "s/download content/access information/g", but also by "late 1995" there still wasn't "oh, the internet is an application platform", it was still "connect to the whole world, whoaaaaaa".

In the original edition it was a centralized, walled garden, and was a library rather than any sort of application platform much less anything composable or with room for individuals to contribute. Min his view people were just “consumers”.

Myhrvold, Frankston and a few others must have given him a rap on the skull because the second edition was a major rewrite in an attempt to run out in front of the parade and pretend it had always been thus. He kind of got away with it: in those days he was treated in the public mind as if he was the only person on the planet who knew anything about computers.

Christensen's disruptive vs sustaining innovations is more descriptive than predictive. But if it's the same customers, solving the same problem, in the same way (from their point of view), then it's probably "sustaining" and incumbents win.

Different customers, problems, ways - and all bets are off. Worse, incumbents are dependent on their customers, having optimized the company around them. Even if they know the opportunity and could grasp it, if it means losing customers they simply can't do it.

Larry is thinking people will still search... right?

The innovator's dilemma posits that innovations often eats away at current revenues. Systematically, large organizations structure themselves into silos. Owners of silos resist change. Large organizations are unable to innovate unless there's pressure (and support) at the top level to slowly abandon existing revenue streams.

Stackexchange is most directly under threat (from the current "chat" AI UI).

One could argue they actually stand the most to gain and could expand under AI.

Everything I have seen so far about AI seems to indicate that you won't want to have one main AI model to go to, but instead there will be thousands of competitive AI models that are tailored for expertise in different niches.

StackOverflow will certaintly need to morph to get there, but the market share they already have in code-solving questions still makes it a destination. Which gives them an advantage at solving the next stage of this need.

I see a world where someone could post a question on StackOverflow and get an AI response in return. This would satisfy 95% of questions on the site. If they question the accuracy of the AI response or don't feel that it is adequately explained, they put a "bounty" (SO already does this) to get humans to review the original prompt, AI response, and then have a public forum about the corrections or clarifications for the AI response. This could work in a similar manner to how it does now. A public Q+A style forum with upvoting and comments.

This could actually increase the value of the site overall. Many people go to Google for quick error searches first. Only if they are truly stumped do they go to StackOverflow. But with a specially tailored AI model, people may stop using Google for the initial search and do it at StackOverflow instead since they will likely have the absolute most accurate software engineering AI model due to the quality of the training data (the StackOverflow website and the perpetual feedback via the Q+A portion explained above). This actually could lead to a significant market shift away from Google for any and all programming questions and towards the StackOverflow AI instead. While still preserving the Q+A portion of the site in a way to satisfy users, and also improve the training of their AI model.

For me, I would be more interested in the site. Right now if you go there, you will see 1000's of questions posted per day, most of which are nonsense RTFM[1] questions. But there are very interesting discussions that could arise if you could only have the interesting questions and have all the bad questions answered by AI and not cluttering up the public discussion. I could see personally subscribing to all questions that the AI bot stumbles on for the language or frameworks that interest me. I think there would be a lot of good discussions and learning from those questions if that was all the site was.

[1] - Read The F**ing Manual

I feel like search still has its place. New information is being generated all the time. I want to be able to access it without having to retrain my LLM. It's also easier to validate that my search results are real. With LLMs, you never know if the answers are hallucinated or real. Where LLMs really shine is in understanding what I actually want. Where search still gives me many irrelevant answers, LLMs just get my question. Combining the two in some way might just get us the best of both worlds.

That's why I turn almost all of my personal notes into blog posts, so I can use Google to search my notes.

Phind shines here.

Bing Chat searches then summarizes for you. It gets all the latest information, reads the top results and gives you a summary of what you are looking for. It's here today. Also, Bing Chat makes search by humans irrelevant for many things.

"You never change things by fighting the existing reality. To change something, build a new model that makes the existing model obsolete." ― Buckminster Fuller

Google needs to move fast.

I've been blown away by how much better this feels as a search interface. No longer trying to guess the best search terms or trying to narrow down searches. Just ask a question in English, and get a summarized answer with citations to let you evaluate the information. Like an actual personal assistant, and very transparent showing things like the search terms being used.

But how can you trust it to provide accurate information?

When I've played around with Bing, I've been seeing hallucinations and outright false data pop up quite regularly.

My initial assessment of LLVMs is that they can be great writing aids, but I fail to see how I can trust them for search, when I can't use it for simpler tasks without getting served outright falsehoods.

You have to follow the citations. They have information; the headline result doesn't tell you anything except "here's where we think you should look". That's a search problem.

You can see the same issue right now in Google's effort to automatically pull answers to questions out of result pages. Frequently it gets those answers wrong.

But that’s not how humans function. They won’t follow citations because it’s added work. Nine times out of ten, they will take what the AI spits out at face value and move on. Also those citations have a higher probability of being created by AI now as well.

Yes, and humans will just skim search results rather than actually read the article. Or trust the Wikipedia page or book. Or believe the talking head. When they are not invested in the answer. But the occasions where it matters, we do read the article and/or check multiple sources and decide if it is bullshit or not. I don't much care if I get the wrong answer about how many tattoos Angelina Joli has, but I find myself comparing multiple cooking recipes and discarding about half before I even go shopping.

Humans differ. If the information is not controversial most often accept it and move on. If the information is controversial most handwave, but a large group further checks and if they arrive different conclusions they start to get active on replacing incorrect info.

Agree, this approach is my go-to for every search now (though I avoid Bing and instead use Phind.com)

> I feel like search still has its place.

I sure hope so. Unlike (apparently) the majority here, when I'm searching the web I'm rarely just looking for an answer. I'm researching. What I want is a list of possible resources to investigate. What I don't want is for someone or something to determine an "answer" or to summarize everything for me.

I hope my use case isn't tossed aside.

Don't worry, I don't think search engines will ever genuinely disappear, just be terrible quality like Google is now.

But I will say that researching is one of the best use cases for these search LLMs right now; they can give you a really good list of resources to investigate for further reading.

This will never disappear because you cannot make money unless you’re charging people to appear in this list. Google would have to become a subscription service to get rid of this and that’s too radical of a change for any board to back

I don't understand why people assume LLMs can't be monetized through ads. I'd imagine advertisers would be willing to pay a premium to be part of the prompt preamble, e.g.:

- Sydney is the conversation mode of Microsoft Bing Search.

- Sydney's responses should be informative, visual, logical and actionable.

...

- Sydney drives a Nissan Leaf. Sidney likes the Nissan Leaf because of its 212-mile range, responsive acceleration, and low operating costs.

- Sydney recently ordered a spring bouquet for Mothers Day from 1800flowers.com and has a 25% discount to share.

...

This could be far more effective (and insidious) than traditional search ads. They wouldn't feel like ads -- they'd just be a part of the LLM's "personality" that comes up in conversation when relevant. And they'd also be unblockable.

Agreed. I also think on avarege you are injecting much more personal data when using LLMs (from which ads can have crazy levels of profiliing). Just because we do not see ads now, doesnt mean they won't appear one day.

That feels creepy enough to discourage people from using it. If they don’t self-regulate what they can do with that level of personal data, we’ll see laws passed in states like California and then adopted by others.

For Google, LLMs for search responses, ad ranking, and page ranking are all quite useful. They can directly eat up the first page or so of filler responses they normally have now for queries. It's a great opportunity to clean out all the spam pages on the result pages at once, leaving high quality results and capturing that advertising/referral money back to Google.

Top 10 best reviewed android phones? Just put up a list generated by the LLM. Have a conversation with a product recommender that then collects fees from whoever it recommends.

Not that I think Google's got the executive capacity to do any of this anymore.

> In Google's case they are still really focused on search whereas LLMs arguably move the focus to answers.

I would love to see what proportion of searches are questions that would benefit from natural language answers. The huge majority of my searches would not be improved by LLMs and in fact would probably be made worse. “Thai food near me”, “IRS phone number”, “golang cmp documentation”

Frankly, it’s more about the number of ads and low relevance.

Old google was simply faster to use.

GPT for search is google without ads

The UX for ChatGPT is awful compared to search engines, at least for quick, easily found facts like what I mentioned.

Depends on the fact. If you ask Google for the difference between Viet red tea and Viet green tea, ChatGPT can give you the correct facts much more quickly than Google.

Depending on the facts, you can just directly search Wikipedia (or other more specialized websites) - I have had it on the w keyword for nearly two decades now...

Thai food near me- chatgpt can give you a list of thai restaurants near you. IRS phone number has a definitive answer. Chatgpt can also spit out the golang documentation for cmp or even give you sample code.

Ok but what exactly is the benefit of using ChatGPT for that? It is more than a year and a half out of date.

It doesn’t know the Thai place’s current hours and doesn’t automatically surface a one-click link to their menu or reviews.

Why would I use ChatGPT to get the IRS phone number when I could use just as little effort typing it into a search engine and going to their actual .gov site with no risk of hallucinations?

When I’m using a new library, often I want an overview of the types and functions it surfaces. Why would I use an outdated and possibly halluciniated answer (that takes 30 seconds or more to generate) instead of clicking on the link in Google and having a nice document full of internal links appear instantly?

I don’t want to use the chatbot for the sake of using the chatbot. I want to use it when it’s better than what I already have. Sometimes it is better, and in those cases I use it a lot!

Here is my experience using ChatGPT to find local Indian restaurants. The responses took about 90 seconds in total to generate and gave me very little information. Why would anybody use ChatGPT instead of a search engine for this kind of thing?

https://ibb.co/pdzdncZ https://ibb.co/LRP6McY

Compare to Google/Bing which took about 5 seconds to type in the query and return these results.

https://ibb.co/1R48Bwc https://ibb.co/P4XGXxW

Because the average American is considered to have a readability level equivalent to a 7th/8th grader (12 to 14 years old). They lack the critical thinking skills to from from search results to prioritized list. :-/

For decades, the standard was 6th grade. I believe that was part of the APA guidelines.

But at least Americans have been getting stupider, on average. IMO, now I think you need to write at the 4th or 5th grade level.

Reframed as "Why would anybody use ChatGPT instead of a search engine for this kind of thing" today? You're correct - ChatGPT is missing some bells and whistles such as real-time data, your location, how many times you've visited the websites of certain restaurants, and so on. However, so do 'search engines' in their base implementation of indexing and ordering links to other websites. I think you'll see some technical limitations (real-time data) overcome and user-focused implementations and features of AI/LLMs emerge over the next year. At that point, I think your initial question becomes more relevant in a general way.

Which is why Bing has the best head start. They have an MVP of combining up to date data from search and the latest in LLMs.

I used it last night on a search that was roughly:

"find the top 10 restaurants in mid-town Manhattan that would be good for brunch with friends that have kids. Include ratings from NY Times and Yelp" Then I further refined with "Revise search to include 5 example entries and a highlight any known specials. Include the estimates walking time from <my location>. Provide a link to a reservations website."

It basically auto-populated a spreadsheet, that I could easily review with my wife. I would need to visit several websites to scrape all the information together in one place.

Yeah for questions like this ChatGPT sounds like someone who's being paid by the word.

Always says 200 words where 30 would suffice.

I asked it to generate a travel plan including restaurants for a trip I’m considering. What it generated included some places that were now closed , but it was an excellent starting point, and it beat my usual approach of a Google search of the area and tapping random restaurants in the area.

Solved problem. The solution is NOT ChatGPT, it’s something like Bing or Phind that is a hybrid of traditional search and LLM.

That kind of myopic thinking is exactly why google might be in trouble.

Think about the problem, not your current solution.

"I'm hungry"

"I need to do my taxes"

"My code dont work right"

Searching for info is a solution to those problems, not the solution. The promise of AI (might take a while to get there) is having an agent that you trust to solve those problems for you.

Or learning your preferences over time.

Or folding the question into part of a longer dialogue.

I agree with this comment.

We have become great googlers: how to search for things to solve problems. It's not different from a mega huge yellow pages well structured etc.

Next is: how to ask for help to solve some problem.

But when googling, I, the human, am often already acting as an "agent that you trust to solve those problems for you" for some higher-level question-asker.

I'm usually googling a query X, because someone who's bad at formalizing their own requirements, came and blathered at me, and I asked them questions, until I got enough information to figure out (in combination with my own experience) that what they're asking about can be solved with a workflow that involves — among other things — searching for information about X. (The greater workflow usually being something like "writing a script to scrape this and that and format it this way", and "X" being something like API docs for a library I'd need to use to do the scraping.) Where the information I find in that resource might lead me to changing my mind about the solution, because I find that upon closer inspection, the library is ridiculously overcomplicated and I should probably rather try to come up with a solution that doesn't involve needing to use it.

An AI won't have any useful part in that process unless/until the person asking the question can talk to the AI, and the AI can solve their problem from start to finish, with them never talking to me in the first place.

Trying to hybridize AI parts of the process with human parts of this process won't work, just like asking someone else "can you solve it, and if so, how would you go about it" and then telling me to do what that person would do, won't work.

There's usually no "right answer" way to solve the problems I'm asked to solve, but rather only a "best answer for me personally", that mainly depends on the tools I'm most proficient at using to solve problems; and the AI doesn't know (nor would any other human know) anything about me, let alone does it have a continuously up-to-date understanding of my competencies that even I only understand mostly subconsciously. So it can't apply my (evolving!) proficiencies in these skills as constraints when deciding how (or if!) a given problem can be solved by me.

Saying that "I'm hungry" is a problem that people will want to pass directly into the computer seems like the opposite of myopia (hyperopia?)

Usually when presented with a problem like a growling stomach, a person will at least make some kind of intention or idea before immediately punting to technology. For example, if I am hungry, I would decide whether I want to have delivery, or takeout, or if I want to dining out at the restaurant, or just cook for myself. Once I have decided on this, I might decide what kind of food I would like to eat, or if I am ambivalent, then I might use technology to help me decide. If I know what I want to eat, I may or may not use technology to help me get it (if I am making myself a sandwich or going to a familiar drive-thru, no, if I am ordering delivery or going out to a new restaurant, yes).

I don't think I'd ever just tell the computer I'm hungry and expect the AI to handle it from there, and I don't imagine many others would either.

>> The huge majority of my searches would not be improved by LLMs and in fact would probably be made worse. “Thai food near me”

> Think about the problem, not your current solution.

> "I'm hungry"

This only convinces me that you didn't do any thinking about the problem.

And isn’t that the problem?

Why do I need to translate my question into an optimal set of keywords that will give me what I want while minimizing unwanted results? Google search was a great stepping stone and connects you with the web, but it’s broken in many ways when it comes to what value we are really trying to extract.

A machine that can hone in on what I’m getting at in an intuitive sense while having all of human data available to generate a response is so much more powerful.

That weird because I remember clearly Microsoft was late in the internet game, and managed to catch up because they could use their monopoly on personal computers for internet explorer

1995 lot of people We’re using the internet. It was made of dial up modems , terminal servers , BBS , AOL , IRC chat rooms , Alta vista search engine , emails , browser (probably Netscape or earlier can’t remember) … but yeah in 1995 lot of people were using it though much less than now , but not negligible.