Learn as you search (and browse) using generative AI | Hacker News Reader

Learn as you search (and browse) using generative AI

blog.google

162 pointsjonifico192 comments

Ironically, articles got pointlessly long because of Google. Everyone stretches their content because long form ranks better.

A few hours ago I wanted to know the difference between a typhoon and a hurricane. Here is the answer: they're the same thing, just in a different ocean. I clicked three or four articles and had to read for a few minutes to get to that answer.

It's even worse with videos. It takes 10 minutes to answer the simplest questions, because that's the ideal length for monetisation.

Google shifted ranking to length under the guise of quality to cover for the increased ad impressions they were getting for those junk rambling articles.

Google has been enshitifiying the web for a good while now, and I used to work there! I worked hard to make the core product more efficient but ultimately your internal goals just override the external ecosystem. And with so much power google was able to exert a shifting force on the web.

> A few hours ago I wanted to know the difference between a typhoon and a hurricane. Here is the answer: they're the same thing, just in a different ocean. I clicked three or four articles and had to read for a few minutes to get to that answer.

I just googled “what is the difference between typhoon and hurricane”

Top result is from the Red Cross, with this blurb conveniently extracted so I don’t even need to click into the article unless I’m curious more deeply:

If it's above the North Atlantic, central North Pacific or eastern North Pacific oceans (Florida, Caribbean Islands, Texas, Hawaii, etc.), we call it a hurricane. If it hovers over the Northwest Pacific Ocean (usually East Asia), we call it a typhoon.

With this Search Lab enabled Google just gives you the answer above all the results.

""" The only difference between a hurricane and a typhoon is the location where the storm occurs. If a storm is above the North Atlantic, central North Pacific, or eastern North Pacific oceans, it's called a hurricane. If it hovers over the Northwest Pacific Ocean (usually East Asia), it's called a typhoon. """

The search labs are not available in my country and I get the quick snippet response anyways

Yes, I can tell you the difference between a typhoon and hurricane. But first, let me tell you a story about how my grandmother once survived a typhoon ... etc. etc.

I'm just waiting for the shitty recipe sites to start hiding the recipes in the content...

* Rosemary - I remember the first time smelling fresh rosemary, it was the spring of 1989 and I was in my dear aunties kitchen...

* 1 boneless skinless chicken brest - Chicken breasts are a common ingredient in many dishes throughout the world. This hearty chunk of meat contains a high amount of protein and is quite versatile...

This is astonishing: https://archive.ph/fbLlZ

> "Let’s start with some research from Hook Agency, which claims the best content length for SEO in 2023 is between 1,760 and 2,400 words"

The source, however loathsome, is probably correct. The front pages of the internet have become saturated with garbage ultra-long-form articles... not even "articles," really, more like ramblings. And it's Google's fault.

We traded a sane web experience for a better search experience and now both are trash and filled with ads. The modern internet really sucks

That's actually not the main reason at all. It's so more ads can be inserted.

The classic example is mega-long intros for recipes. But if you look into it, the primary reason is to be able to include 10 ads on a page instead of 2.

For questions such as the difference between typhoon and hurricane, ChatGPT and Claude give accurate answers in just a couple of seconds. It’s such a different experience compared to the search engines.

Yeah, search engines give accurate answers for that question with no noticeable latency, providing a much better experience.

That was the difference you meant to highlight, right?

Worst part for flow is how fast chatgpt logins expire. Ironically probably some kind of anti-botting measure or is there a better explanation?

But how do you know if they’re accurate?

You don't. Just like before generative AI responses. If accuracy is required, you have to check the citations and make a judgement call. And no, people generally won't do that just like they didn't do that before.

Google's "featured snippets" make it too easy for content farmers to thrive. As long as someone has the exact search "What is ___?", "How do I do _____?", etc formatted as an h2 with a simple explanation underneath, they have a pretty good shot of getting a feature. There's really no quality control.

Although they do look similar, there are quite significant differences between a Hurricane and a Typhoon.

A Hurricane has a ~1000hp early Rolls Royce Merlin engine and a top speed of about 350mph; a Typhoon has a 2000hp+ Napier Sabre engine and a top speed of over 400mph.

The Typhoon was actually intended to be the replacement for the Hurricane, but challenges in the high altitude interceptor role led to the Typhoon taking on a more fighter-bomber role as the war went on.

A Tornado is of course even more powerful. It has a pair of Turbo-Union RB199 afterburning triple-spool turbofan engines with 9800 lbf thrust each (17300 lbf with afterburners on) and a top speed of Mach 2.2 with the wing swept back.

Edit: at full afterburner, it burns fuel at a heat output of 235 MW (expressed as horsepower: 315140)

The Eurofighter Typhoon houses two EJ200 engine with 60 kN (13,500 lbf) of dry thrust and >90 kN (20,230 lbf) with afterburners. I believe there are experimental versions of EJ200 with >100 kN thrust but since Typhoon is being retired across Europe and no oversea sales, we won't see Typhoon armed with them.

A land route exists, so not technically overseas, this Typhoon is sometimes observed outside its signature range, eg over Yemen.

I had understood that the Hurricane was a stop gap interceptor/air superiority fighter until more Spitfires were produced. I'm quite surprised that the Typhoon wasn't seen as redundant, unless it was intended early in the design phase for bomber interceptor, but the need for that waned.

I got sucked into a clickbait recipe video for "2 ingredient cake". The video was just over 10 minutes long, which I think puts it in a different monetization category on youtube. I spent wayyy too much time out of my life trying to find out what 2 ingredients are in the cake.

Answer: apple sauce and gelatin

Next time you should copy the transcript into chatGPT and ask for a summary.

I assume this is all done server-side, with Google feeding your interests and investigations into your personal profile. I really wish this could be done on-device. I don't want Google slurping up even more data about web users.

I use FF mainly on desktop and mobile, and there's probably a great opportunity for Mozilla to build an offline, privacy-first summarization model.

Yes, what Mozilla needs to do remove more support from browser development and debugging, to fracture their efforts even further, to continue the line of successes that began with Persona and continued with Firefox OS.

I’m wildly speculating, but it seems that projects like llama.cpp are bringing SOTA models closer to the desktop. It’s only a matter of time before browsers embed a small LLM for various purposes, providing access to this local model via a JavaScript API that allows clients like Gmail to perform tasks on data locally. Apple would be strongly incentivized to do this, given their value proposition of user privacy.

I think the fact Google curiously ignored all the security problems raised by the WebGPU API suggests they are closer to trying to offload the GPU inference part of this to end users than people think.

Build as much of the model as you can in the cloud, run inference locally and push results back is probably the cost optimal way to run this stuff at scale.

Kagi has been doing this already for quite a while, well before google (see Kagi Universal Summarizer) with no info linked to your profile [0]

[0] https://help.kagi.com/kagi/getting-started/faqs.html#i-have-...

For the record I am a happy user, not affiliated.

> I assume this is all done server-side

I assume so too, and given that, it's incredibly frustrating (but not at all surprising) that they require users to use Chrome to be able to use the feature.

At Apple, the challenge our team faced with on-device models was managing their size and update-ability when every team has multiple models.

This doesn't change how or when Google sees the web pages a user visits. If they were using Chrome with all sync features turned on, their browsing history was already being sent up.

“Summarize this YouTube video” (given a link) is what I’d like to see and Google could easily do when there’s a transcript, but they’d probably rather make people watch the video.

I think Kagi might do this? Any other good solutions?

I used summarize.tech [0] once, and it worked fine, although it was quite repetitive iirc.

[0]: https://www.summarize.tech

Kagi has such a feature, its called Universal Summarizer [0]

[0] https://kagi.com/summarizer/index.html?url=https%3A%2F%2Fwww...

I've been using this one: https://glasp.co/youtube-summary

Exactly. It's already easy enough to scan an article in 15 seconds to get the overall gist. But scanning through a YouTube transcript, in the tiny box, without any punctuation, without paragraphs, without sections, takes forever.

ChatGPT plus with a YouTube plugin does that too.

Interesting. I see multiple plugins. Any there any you recommend?

I use VoxScript because the first plugin I tried didn’t work well (time-outs) and it was the second and last YouTube plugin I tried. It’s okay but sometimes I need to ask for a full transcript otherwise it will get only the first chunk of the subtitles.

I use the AAAsummarize.io chatgpt plugin to summarize youtube videos. Has worked really well for me.

so many chatgpt plugins start with an "A " in their name to "rank" first in the list of all plugins. throw back to the telephone directory days :-)

I typically run lengthy YouTube video transcripts through Anthropic Claude to get a summary. It's free and offers a 100k context window. After summarizing, you can engage in a conversation with Claude. You can ask specific questions about the covered topics or prompt it with queries such as "What were the standout points?". It's quite adept at extracting key highlights.

Looks like that is still invitation only.

Really? I was under the impression that it is available to everyone. Are you attempting to access the API? There might be a waitlist for that. But, it should be readily available for use here: https://www.anthropic.com/index/claude-2

"Unfortunately, Claude.ai is only available in the US and UK. We're working hard to expand to other regions soon."

This sounds like a fantastic feature. And honestly I think all of these negative comments are just the HN reactionary trend. Lots of people are going to use this tool and it's going to be great to bypass all those annoying sites that have a lot of crap on them. Being built into the browser makes this very useful.

Also, all this wailing and gnashing of teeth is really quite annoying. Look at you people embarrassing yourselves.

Privacy aside if it's done server side, I think this crosses another dangerous line.

With this, Google controls how content you read is summarized, they could introduce biases, intentionally or not, and I'm not sure we should trust such an actor for this.

Solution: dont use google.

Seems like that's the direction things are moving. Google's becoming heavily enshitified.

Becoming? I beg your pardon?

Their entire business was built on this. https://en.m.wikipedia.org/wiki/List_of_mergers_and_acquisit...

I know this is HN and all, but to wait all of these years of acquisitions, poaching, project terminations, Chrome shenanigans to destroy the internet, main product "SEO" degradation and literally every single sign of enshittification...Google has always been one of the prime examples of enshittification. That's why they are very successful at making money.

Of course I can, and do, avoid Google.

That's an individual solution. I'm interested in a world that's as good as possible not only for me, but for the other people too. Most people around still use Google. They might find this new feature useful and start using it, and possibly suffer this bias. Not only they may get manipulated individually, but it can have broader consequences, even to those avoiding the feature. If Facebook can influence which president get elected in a country, it will also affect people avoiding facebook (sometime for this very reason).

Avoiding stuff for oneself is a first step but it's hardly a solution, these big corporations have a non negligible effect on the world, still with you as a member.

We'll have to start to raise awareness around these things, in addition to all the subjects on which we already need awareness.

But same old same old I guess, Google already puts you in a bubble, just even more concerning I guess. It was manipulating us, now it's another, new, potential manipulation angle.

I share your frustration, but it's hard not to be pessimistic about this. Regular internet users will flock to these tools, just like they flock to social media, even though it's a clear detriment to our personal and collective well being.

If the Snowden revelations did anything is to show that people ultimately don't care about being exploited, as long as they can continue to enjoy technology. Trading convenience for privacy is not something most people would even consider.

Things will quickly spiral out of control once everyone has their personal AI assistant, that is controlled by a major corporation with government ties. It's the wet dream of any aspiring autocrat, and we're building that future.

Here we are, on a tech forum among people who work at these companies, and all we can do is type at the void. These companies continue to build products that on the surface seem enticing and useful, which is surely how they also pitch it internally so engineers that build it have a clear conscience, yet lead us to a further breakdown of society in ways we haven't imagined yet.

I'm not sure if all this is over the top, doom and gloom thinking on my part, but it feels frustrating to be part of this industry, seeing how it's tearing apart society, and having the concerns ignored and dismissed by both users and companies alike.

> but it feels frustrating to be part of this industry

There are places in this industry that are not so bad. Of course it has drawbacks and require effort, but you can get paid for doing things that align (as best as possible) with your values.

They already show heavily curated content, with a conscious and intentional bias into presenting, for example, groups in a light that doesn’t match reality.

You won’t find info about any merely “controversial” idea on Google, it’s all censored, it’s all disinformation already. Try even finding memes, you generally won’t find them. Try finding info that doesn’t paint some politicians in a good light, you won’t find it, even though it exists with the exact keywords you type in and the crawler returns a success for that page.

So what do you mean, while you already trust Google to give its bias to the current world?

What’s a good example that I can Google for but won’t find results (censored) or where only one side is presented?

I will Google it and report back…

Yeah. You gotta be searching for some real bonkers shit to have google not show it up.

The most completely batshit insane groups I’m willing to google anymore are flat earthers, young earth creationists, and Facebook mom groups, and google happily indexes and returns favourable results for all of these.

Even looking at off the deep end alt-right content like men’s rights, red pilled, trucker convoy, anti-masker, Tate, etc, google has zero issues return favourable results for all of them in a private window.

Really begs the question of just what level of illegal shit this dude is trying to get google to return.

Classic hacker discourse...

1. Bunch all people who disagree with you into one and claim they are some kind of group. Bonus for listing 12 or more "groups".

2. Raise suspicion about the user you're replying to. Maybe they belong to one of those people you just listed? Maybe they are a terrorist? Maybe they voted for the wrong guy? Should we call the police? Of course we should! He's even trying to make Google show "bonkers shit, bat shit and illegal shit".

No matter how hard you try to follow the line, one day you will find yourself in one of those groups who people like yourself try to bunch together, and be exposed to the same level of suspicion and hatred that you yourself are unloading in your comment.

People who make suspicious, dubious claims deserve to have their claims called out.

I am not going to give up on calling out this purposefully vague, bullshit commentary just because you dislike it. If you don’t like having your sides bullshit being called out, don’t post it.

It’s actually hilarious how eerily similar your group is.

>oh woe as us, we’re such big victims

Someone responds shutting it down

>why do you have to call out the posts. Can’t you just not respond

Every single time like clock work. It’s like you guys all belong to a forum of AI generated comments and responses and you just copy and paste them.

I don't think anybody is interested in giving you an example, because they expect that it will get them banned from HN, or at least have their comment flagged and removed by activists.

Google does censor content, and they are kind enough to inform the owner of a website by e-mail when they have censored the content from their search results, so it is a fact that is just undeniable.

What a nice escape hatch. I can't give you and example because I will be banned if I do, ergo there are cases that will get me banned. Not QED.

don't worry, it gets shutdown in 2 years

They will shove ads in it eventually and then it’ll become core to search.

Google has always introduced biases by virtue of ranking your results and deciding which content to surface.

This was always the case.

And this is worse.

It's hard to make a search engine. Any decision they make will create bias and shape the entire web.

It may be summarized to what they want, but it can be what I want too. All YouTube recommendations are already what I want to see. Videos from creators who share my world view, criticizing what I hate, and praising what I've already liked. If this Chrome's new feature is also personalized in this manner, I can read articles against my view and still get only what I want. My world view will become increasingly polarized, and it's already quite biased.

When I try to search up a topic on YouTube, the list of search results eventually STOP showing videos matching my criteria and instead show 10 videos in a row labeled "People also watched" that is unrelated to my search.

It's one thing to get recommendations in a place where recommendations are appropriate, and the dangers of the feedback loop you are talking about. It's another thing entirely to actively push it in other contexts.

"I know you are searching for programming topics, but might you be interested in this political outrage instead?"

> results eventually STOP showing videos matching my criteria

How long is that typical list? How often do you pick the things at the bottom of that list vs the results near the top?

I pick things from the bottom of the list a lot. I really hate when engineers try to guess what I want. I already know what I want. These “automatic” products are terrible.

What do you think is a good list length? One can assume search results are practically infinite, so you have to make a decision of where to stop. What would your heuristic / algorithm be for determining the termination point?

A list of every video where my terms show up somewhere in the title or description.

Do you want it sorted in some way? If so, why might you go to page 99999999999999 and pick the last item?

I’m not searching for some huge wildcard — I’m searching for some video content which must be generated by somebody and uploaded. There are not billions of results for “cool cover of a song I like” — there are hundreds of results at most. I want those hundreds of results.

Maybe you can just use a different product (search site) to get better results.

> One can assume search results are practically infinite

I don't think that can be assumed at all, actually. At least, it sure doesn't look that way when I search for most things, on YouTube or elsewhere. Most of them run out of relevant results pretty quickly and then start including obviously irrelevant results.

I did a test, searching for a pop song. I got 9 results for that pop song and then the "People also watched" section comes up. After 4 results with a +6 more button, it goes back to my pop song results.

What you're describing is a common UX dark pattern. The system must never show blank space in the UI for a search result, lest it drive them into the arms of a competitor.

We can thank Netflix for starting the trend of "we don't have this, but how about this other thing that we do have?"

I'm not sure that fully explains it here. I see more relevant search results below the "People also watched" section (just checked now). Further down, there is also "For you" and "Previously watched" which have videos of no discernible relevance to the query and these are also followed by again more relevant search results.

No, youtube will cut off your search results early to show you videos it think will engage you more regardless of how many search results there actually are.

Google does not give a fuck what you think you want, because they know better than you

The trick to get rid of that is to go to search filters and select "videos."

Well, I'm not quite at ease with trusting a third party to try to guess my (ever changing?) world view.

How do you know it works well? How do you know it misses nothing? How do you know it's not subtly biased (intentionally or not) in a way you would not notice, or you would think it's good enough?

> I can read articles against my view and still get only what I want. My world view will become increasingly polarized, and it's already quite biased.

Yes, of course we all do, but it's another thing to involve a third party in this process.

>All YouTube recommendations are already what I want to see

I wish I had your experience. My YouTube recommendations can definitely be said to be “my fault” in some sense of the word, but wherever the blame lies they’re still mostly terrible. YouTube will watch me skip a video for a whole week, but keep showing it to me “just in case.” I’d rather it were just a dumb search at this point.

With the new Shorts feature the YT has gotten so bad that it isn’t even funny anymore. I wonder if anyone there is using their product.

> All YouTube recommendations are already what I want to see.

I might feel more kindly toward YouTube recommendations if they did this for me.

I was ahead of my time 30 years ago when I achieved the exact same result by simply buying the right newspaper.

And if I wanted to find more in depth material to support my firmly held beliefs I just had to go to the right bookstore.

it's ok, they already read your email

I currently use the "summarize everything" extension for this. It lets you tweak your system prompt if you want to, but the defaults are pretty good. It's just another tool. Use it or don't, but it's saved me countless amounts of time digging through overly verbose prose looking for the interesting bits of information.

On the bright side, if it's fine serverside I wonder if it'll bypass paywalls?

There is no neutral source ofninformation. Legacy tv is often controlled by governments and corporations. Social media are also controlled, by corporations, and by some degree by governments. There is also bias, censorship and moderation.

People were Hunters before. Were searching for food. We do not need now to hunt. We had to hunt for information, we do not need now.

We rely more and more on services and corporations. We choose the jest one, which becomes monopoly after some time. After some longer period of time monopolystic corporation becomes monster.

Circle of life.

We may not want to use chatgpt or other bots, but eventually we will. Common people use the easiest route. They will decide, what is popular, and what will be used.

Great, another excuse for people not to read the article.

We've got enough problems with people reading a headline and running with it. I hate to think of google summarizing some scientific study - particularly some "pop" scientific study - and have everyone confidently reciting it as though it's been conclusively proven without making any effort to look at the source.

It also introduces a second order problem when the summarization algorithm is updated/changed. Anyone reading an article before the update can get one summary while everyone reading after the update can potentially get a totally different summary.

It's like when people talk about Google search results without realizing their search/browsing history has put them in a particular bubble. There's no guarantee any two people's result ranking will be the same which can affect their understanding or perspective on a topic.

Shouldn't they worry about Search being incredibly low quality nowadays first?

They're literally featuring a wrong ChatGPT answer as an answer to this question: https://www.google.com/search?q=country+in+africa+that+start...

The second item is "Better understand coding information in AI overviews". Is there really such a big demand for this from the average google user, or is it just something that's easy to do with these language models?

Articles written by an AI and then summarized by another program. Nothing can go wrong here.

Then they tell us our Java app consumes too much CO2.

Why would an AI tell you that your java app uses too much CO2? What am I missing?

Google searched has encouraged long pages for SEO. Now Google sells us the idea that Google can help you reduce the useless words on the internet.

Maybe de-index receipes talking about how this soup was my grandma’s preferred soup. It’s not so hard, you still have the full power on search engines, Google! For at least 2 more full months! Do it!

You are the problem, Google.

They are too busy turning their browser and search experience into crap. For instance what's with search results in gmail and google drive? I literaly search for a file by name, doesn't find it, I search by extension ... it finds it? Then there's chrome - right click on selected text to search on google, instead of opening at tab it opens a crammed side bar? Has google gone insane?

I've noticed gmail search has gotten significantly worse. Maybe my inbox corpus is becoming too large or google may be dedicating less server resources to search, but the frequency of failed inbox searches has significantly increased over the past year. (Failed as in I try to search for an email I know exists, but can't find it)

Looks like my inbox is at 250,000 emails. Wonder if search degrades with inbox size. (Same with drive?)

From my experience with corporates this is poor product management. I think these are not bugs but rather poor design choices made by people that have no clue other than how to tick boxes of "inovation" delivered. Whether features make sense or not the product manager's KPIs will look good.

Articles are getting too long? Since when has this been a problem? I think it's the amount of crap ads injected while I'm trying to read the article that is the main problem.

selling the solution to a problem they created

SEO rears it's ugly head again - yes, articles have gotten much longer, why? Because the Google bots think longer content = more authoritative content.

It's not actually a direct correlation, but enough people take that as gospel that they will pad out an article to make it super long just for the SEO value.

Similar to recipe websites giving their entire life story before a recipe.

Have you tried to find a recipe online? Every one comes with a life story because of SEO.

Ironically Google is the main cause of articles getting really long and full of fluff.

They already have your browsing data, folks. This is just caused by the LLM push at Google. They got a top-down mandate to sprinkle LLMs on everything and this is the result.

I'll use AI to puff out articles with nonsense so they get higher ranks, and then Google will use AI to shorten them again.

The Dead Internet theory will soon be indisputable.

Anyone who has spent time with AI chatbots knows how often they are just completely wrong, often in surprising and confusing ways. It's hard enough to parse articles for accuracy and coherency without first passing them through as random error-adding machine.

Humans are worse. Obviously if you don't know how to discern garbage data from legit data, you're going to have a bad time, but these emerging tools are strictly better than anything that has existed before.

The range of errors we expect from humans is more narrow, more relatable and we have more experience parsing them. Adapting to parse AI-written material will require developing new skills, and I'd argue those skills will be more difficult to develop because the errors/inconsistencies generated by AI are not as relatable to humans as errors introduced by humans.

So if you opt in then your website won't get clicks -> no ad revenue for you since google will "steal" the clicks via their summarization.

And if you opt out google will eventually just push you down.

This has already been the case for a while with "Zero-click searches". It's when you look up a trivial piece of info that can be said in one line. Google will show that sentence on the search results. You don't have to click through to the full page.

The best part is that product wasn't even that accurate, and has no qualms about showing you straight up bullshit or a lie stolen out of context from whoever did the actual research.

Google is ahead of the curve in how ready they are for LLMs!

I wonder how much they're projecting to spend on this? Or maybe they're prepared to eat a loss on this just to stay relevant in the face of Bing/Kagi/OpenAI?

Results from these models are orders of magnitude more expensive than traditional search. Maybe they have some benefits of scale where they can cache and serve up the same summaries to many people?

I use the "summarize everything" extension that uses my OpenAI key, so it costs me a fraction of a penny any time I summarize something. Subsidizing something like this at scale would be pretty dumb.

I understand the worries about privacy and monopoly power.

But this doesn't change that this is a useful feature.

PS: That being said, I have started developing a distaste for google search. I have been served so much SPAM lately, that I am now starting to associate that search box with low-quality content.

This is where you start to sympathize with the news people in Canada who want google to pay for links. If google just summarizes at scale, interdicts the traffic and revenue, where's the incentive to create content?

This is a failure mode for the information economy.

If your only incentive for creating content is harvesting attention and eyeballs, then the internet is better off without your content.

With that way of thinking, you exclude almost every content, ever.

Most people do not produce content to be left unread.

I just like the fruits of my labour to be attributed to me, not to a large corporation.

I'm not sure that is the case: TV and radio use that model and are/were valuable.

I do agree with you.

Google's behavior in regard to informations reminds me a lot of the "torrent leeches" of the past: siphon everything you can, give nothing back, ..., profit. But at a huge scale with an impact on society far more than "a few more bytes on a personal hard drive".

Good feature, I might actually start using chrome again if recipes are summarized how I expect them to. Generally speaking, there's already a problem with people only reading the titles and basing their opinion of off that. Hopefully this will encourage people to read more, since it will be significantly shorter than the actual article. This won't affect the people who care about the topic.

I'm not really worried about bias, since from my experience, summarization software will inherit the bias of the creator. Sure, it's behind generative AI, which can introduce more bias but it's not like the original article is hidden. No one is forced to use it.

LLM's already summarize articles for you. This approach to me combines the downsides of individual articles and the downsides of LLMs. I guess Google likes it since you presumably might be finding these articles through Google search.

Reading is not a binary process and you can scale your engagement with any written work. If you use a variety of techniques, there is probably no need for a machine summary.

For example, start by skimming and scanning an article and you may be able to pull out all the salient points and satisfy your reading goal or choose to go deeper with your analysis depending on time and interest and what you are trying to accomplish:

https://advice.writing.utoronto.ca/researching/skim-and-scan...

Skimming isn't really possible for people that rely on screen readers. I think effective summarization would be useful for them. Whether this is an effective implementation, however, remains to be seen.

It takes much more time to even skim an article than to run it through a summerizer. Not to mention that it’s generally hard to skim a YouTube video.

I do believe contextual link previews help with overall browsing UX. I like how Wikipedia has a summary when you hover over a link. Where this summary comes from does matter.

For those who don't like when Google summarizes content for you but like the idea of getting a link preview with a summary provided by the website owner via meta tags before navigating to a third-party website - I've built a product called Linkz.ai [0] - that allows website owners to install link preview popups with 1 line of code.

[0] https://linkz.ai

At the risk of sounding like an old man — heck, I guess I am — is this innovation ethical for humanity's future? A one-click "tl;dr" button for the web?

There are studies saying overall human attention span is shrinking. Anecdotally, I know people in their 20s who use TikTok daily, and they're no longer able to watch a movie without losing focus.

Squeezing all aspects of our digital life, including learning new skills or reading the news, into a bite-sized "summary," is not going to lead to a better-informed and more productive society.

> Anecdotally, I know people in their 20s who use TikTok daily, and they're no longer able to watch a movie without losing focus.

I'm almost 40 and haven't been able to do this since the smartphone arrived; it's not Tiktok, it's just that the phone is right there, and movies are slow-paced (arty thrillers) or monotonous (lengthy CGI fight scenes).

If we're going off anecdotes, we've also seen an explosion in popularity of the 2-3 hour podcast and season long Netflix binges.

Doesn't this generation have an absurdly long attention span compared to the generation of the 30 minute TV show with three 2-3 minute commercial breaks?

There's things that are different though. People listening podcasts are sometimes multitasking so that's divided focus right there. In terms of modern TV and movies, shots are actually shorter these days. To its worst extent you can end up with these sorts of scenes:

https://www.youtube.com/watch?v=by4UZ-79MK4

This sort of thing is why Canadian media wanted a cut of big tech profits. Imagine how harmful this is to websites that need to grow audiences, particularly news websites

Most news websites have a ton of duplicated content. That info overload is dying to be minimized.

And news orgs are going to keep coming under assault cause they perform a similar function to search engine robots by surfacing new info. As soon as some one copies that info, the value of the info is 0.

There is already data saying most content created is never consumed by anyone just cause of how much copying and duplication happens.

What is a good example of a Canadian news organization website that is not overwhelmingly populated by headlines and stories that are:

- clickbait

- hyperbolic

- gossip

- propaganda

- unoriginal

- unimportant

Great question. Going to assume you mean English language and share my own findings:

- CBC: Propagandic, unoriginal, unimportant

- Global News: Clickbait, gossip, unoriginal, unimportant

- Toronto Star: Hyperbolic, propagandic

- The Globe and Mail: Unoriginal, unimportant

- The National Post / Postmedia network: Hyperbolic, propagandic

- CTV News: Clickbait, gossip, unoriginal, unimportant

Anything I missed?

I have found the best strategy is to take the sum of multiple sources to get a slightly less awful whole.

OK Google, now summarize YouTube videos so I don't have to watch the entire videos or the ads.

Luckily there are AI tools out there do this.

Here is one...

https://www.summarize.tech/

Interestingly, OSX has had a Summary Service for as long as I can remember: Simply select text > App (Menu bar) > Services > Summarize[1]. I still use it from time to time, and it's pretty good, and completely offline!

[1] It's off by default these days, you'll have to turn on the Summary service under service settings.

The example really doesn't sell it to me. The search "what is the most common element on the periodic table" currently returns a big display with "Hydrogen" as the answer, followed by a brief explanation.

That's all I need for such a query. In comparison any text written by a "generative AI" is mostly noise.

Seems popular enough that all I am getting out of it is "Can't generate key points right now".

i use https://chrome.google.com/webstore/detail/cbgecfllfhmmnknmam... for summaries with GPT4 (API Key)

There are other Chrome extensions that will do stuff like this. I'd never choose to have Google do it for me if I could have a third-party extension that I trust do it instead (or even one I don't trust 100%, but which offers some level of anonymization).

But yeah, keep using Chrome because it's allegedly "faster" or so you don't have to test the page you are developing on 3 different engines.

Welp... I wondered when Google would start encouraging me to put my work profile into Firefox, I guess today is the day.

As if people needed even more encouragement to be too lazy to read? Awful feature, this is bad for society.

Hard pass. I don't want my browser doing anything except displaying webpages.

Net effect: fewer clicks to source websites, fewer clicks on ads on said websites.

Good. The adtech industry needs to die in a fire.

Google Reader will aggregate articles for you.

And of course there is no way for a publisher to opt-out of this without removing them from the search engine entirely.

Just another trick to spy on your internet activity.

No thank you.

No thanks.

Hard pass. I don't want Google, or even a trustworthy neutral party, to decide my view of the world. (Compromised as it may be already.)

We're all unique and unpredictable. In a given article, a random sentence or point considered throwaway by most people and algorithms could turn out to be meaningful to me.

I've been trained not to even trust the basic facts they pull of out content like movie showtimes or whatever. Once you see it wrong a few times you realize it's folly not to keep the responsibility of finding information yourself.

Google is like a trustworthy neutral party except for the trustworthy and neutral bit

for non political issues it should fine, eg to describe an algorithm or other non-controversial tech issues for example.

As long as they dont own a product in that field, yeah.

And algorithms and some tech things are probably the only things left that aren't political issues.

Please note if you're reading this, that this comment was written in 2023, before mathematics and physics became highly contended political issues.

It’s summarizing. That is by its very nature a reduction of an article to certain bias elements. It’s not rocket science or some evil magic. Jeeze you’d think HackerNews never used anything that wasn’t a home brew open source libertarian approved product if you believed the comments here.

Not to mention the clear preference here for comments that summarize an article to avoid actually reading it. Or poison the well sufficiently for a person to say “oh well I already know what it says”.

> some evil magic

It can be. Summarizing process can be tweaked very slightly to give more space to certain POVs, make them sound more reliable than others via word choice etc. The subtlety can make it undetectable, yet it can have strong effects when applied on mass scale.

I don't know how this will look like, but it is a very powerful technology to sway the public opinion one way or another.

Can you give some examples and point out the subtleties?

You can already try this with GPT, e.g. "Can you summarize XYZ while giving more attention to its negative impacts on the freedom of individual", contrast with "... to its benefits on the collective wellbeing".

Both will be factually correct, but they will each manipulate the opinion of the non-expert reader by giving more attention to specific facts.

GPT seems to be very strong in being able to present information with some specific accent, e.g. there are many examples of "write this in the style of [person, literary style, political slant]".

"It’s not rocket science or some evil magic"

It sure is.

If you had told people in 2019 that a computer program could summarize large topics and write cogent output you would have been placed in the asylum.

Sorry, but that's incorrect.

Leaving aside the mind-blowingly large amount of research on summarization that existed in 2019, I'll say this: by then the task was so we'll understood that there were multiple tldr bots on Reddit.

There's been an "AutoSummarise" tool in Microsoft Word since 2003 [1]

Whats different these days is people might actually trust the tool enough to use it :)

[1] https://www.officetooltips.com/word_2003/tips/getting_to_the...

In 2013 it was already done by kids: https://finance.yahoo.com/news/yahoo-acquires-summly-app-150...

Actually, we’re not unique and often predictable. There usually is someone who says “I’ll never use the product because…” And it’s often a top comment. e.g. Smartphones with GPS, Amazon Alexa,…

To me it’s simply a noisy discussion with little value. Don’t use it but let’s not waste time debating it. No one’s mind is going to change.

I’d love nothing more than for an algorithm to filter out the noise and extract any information and facts. Learning whose opinion I trust on specific topics would also be useful.

Of course, others will tell me why this doesn’t work for them and they want the added noise. Seriously, the entire point is that some people will want the product while others don’t.

I’m not trying to change your mind, and I’m not trying to stop anyone else from the discussion. It’s simply not for me, in general.

I find it quite helpful to understand why people don't use a product or are unhappy with it. If there's none of that, all I'm left with is reading marketing claims.

Good, then you should set your filter appropriately.

Depending on the effectiveness of the product, you may trust it for certain articles but not others. Certain sources but not others.

If you don't think you'll change anyone's mind, and aren't trying to, then why are you posting?

In all fairness, plenty of people post just to express their opinion or share their knowledge without any intention of changing anyone's mind about anything.

Person 1: “An electric vehicle will never work for me because…”

Person 2: “I want an electric vehicle because…”

Is that a better example to convey my point?

That's the sort of exchange I want to see. It's useful and educational to hear what works and doesn't work for people, and why. It's not noise at all to me, it's teaching me about the upsides and downsides people have found about a thing in their real lives.

I agree entirely.

I have no interest in AI summarizing anything at all for me. Not because of AI, but because I have no interest in anyone or anything summarizing for me. Too much is lost.

> a random sentence or point considered throwaway by most people and algorithms could turn out to be meaningful to me

This is what I crave for in good articles and books. There's some big insight just thrown in in the middle of the sentence because that pattern is already obvious to the author. It's also what AI-generated content is currently lacking because it sticks to the task at hand (but I believe it will change).

Also, your critique applies just as much to human-generated summaries.

My issue with it is that I don't want my browser to send home what I read and view although I guess that ship has sailed some time ago.

I feel like this is knee-jerk over-generalization.

Having Google summarize a too-long recipe (a problem they created, by the way) into just the simple ingredients and instructions is NOT "deciding your view of the world".

But it won't taste the same without knowing the authors childhood trauma!

Maybe Google decides certain ingredients aren't fit for public consumption and changes the recipe in their "summary." Or maybe the FDA tells Google it has to do this.

On a mass scale, maybe 0.2% of people would double check the source page. The rest would be unknowingly influenced to eat "healthier" by Google.

Yes, this seems unlikely. But so has much in the last 5 years.

Why are we deliberately charging head-first into social man-in-the-middle attacks? We already don't trust each other enough. LLMs lie, and lie often. Why should we trust them for anything?

Would you be OK asking a LLM what food is safe for an infant, or a pet dog? Without checking the source?

I think you’re saying that your concerns are moot because nobody would use LLMs the way you fear people will use LLMs?

Ok, but people won't use it just for recipes. If it's successful and common place google will monetize it.

The beginning could just be "see this related product or service." Which won't mess with the content. Perhaps that's fine.

I'd argue though that any summarizing tool inherently has a bias. It must choose to ignore certain details and make decisions about what to highlight.

As we understand LLM's more and the stuff that summarizes folks controlling them will be able to make those decisions and the money and power behind that will absolutely abuse it.

That's not even considering the effects this would have on journalism and writing in general if most of it gets summarized.

Too tinfoil for me. Bicycles are bad because someone will try to use one like a helicopter.

And humans already summarize and (mis)characterize. I just don’t see blaming a tool that is used by people as they see fit.

I do if that tool is going to be more centralized then what it replaces.

Seems like a pretty facile argument. Bicycles and helicopters don't need to be trained on ridiculously largest datasets on ridiculously large super computers.

Pretty sure the long recipes are due to others not being able to copy them if you have long essay describing it or something?

The essay is more protectible by copyright than the recipe itself, but including it doesn’t protect the recipe itself.

I actually think the issue is that the long essay provides more space to put advertising alongside or interrupting the flow.

I'm not sure why this particular idiosyncratic thing attracts so much mystery, but the answer is that there are long essays about the recipes because people like them.

I assumed the long essays were for SEO (inflates the amount of cooking-related keywords). Which would make it kind of ironic that another Google product is now stripping the SEO content back out.

Most articles we read (even on HN) have very little useful information compared to lines of sentences, maybe this can act as a summary and we can decide if we want to read the whole article.

It's just important to decide what you use it on.

I find these tools useful for articles these days because everything is clickbait and full of bloat and nonsense. Even if it's just to do a first past to figure out if I want to read the full article.

But I do understand your point, I feel this is very important ' a random sentence or point considered throwaway by most people and algorithms could turn out to be meaningful to me.', i've had this issue with book summary services, i've read a book and took completely different meanings or found insight in certain paragraphs that were completely glossed over in book summarys that just changed it into generic sounding nonsense. We need to connect the dots ourselfs and to do that, I believe is not to have someone elses summary of the situation.

However, for everything else a tool like this is useful, there is too much noise in the world, we need tools to filter it and help us understand what is relevant and what is not.

I'd stay away from anything Google does. They have proved to be sectarians in politics and machiavellian in business and privacy

Reply on news.ycombinator.com