It's going to be real fun when more than half of "user" generated content on the net will have been generated by AI content farms: forum comments, code, videos, pictures, entire websites.
I honestly think this is how we reach this "dead internet" meme. We're probably less than a decade away from this.
EDIT: here's my bold prediction. In the next 10 years, a very successful content aggregator (imagine Reddit, Instagram or Tiktok), where content is exclusively and openly AI generated, will reach the Alexa top 100 ranking.
We will have HN users saying we're too old and we just don't get why it is more fun to mindlessly get your dopamine fix from scrolling and liking AI generated images.
Arguably, we're already there. Most of the Internet is content marketing, and that content isn't meaningfully different than AI output. That it's generated by protein bots instead of silicon ones is just implementation detail.
> Most of the Internet is content marketing, and that content isn't meaningfully different than AI output.
I argue that it is. Human content farms don't create long-form articles, artwork or deep fake videos by the truckload, in minutes, from just a little prompt. And humans are relatively expensive.
AI generated spam will be orders of magnitude cheaper, faster and thus incredibly more common.
If your site has a lot of fresh, unique content on a topic you appear to be an authority about related search terms and your search engine rank goes up. So pair a bunch of AI generated pages with ads or e-commerce and you’ve got a spam-driven revenue model and we have a worse internet full of regurgitated AI content.
I think their point is that if both real humans and spambots are using AI to generate their articles, then detecting which one is spam becomes that much more difficult.
This is where I see natural language AI causing a real kerfuffle. We're probably already reading bot spam articles without realizing.
The local newspaper where I lived until this summer used software called “Lede AI Sports Desk” to automatically generate high school sports articles with headlines like "High School A edges High School B in snug affair 54-52" and "School C drums School D in sound fashion 48-4".
This is clearly not actually using any process that could be called "AI" with a straight face since it's obviously just playing madlibs with different sets of vocabulary based on the scores and margins, but the results were invariably stupid and repetitive headlines and articles.
To be fair, the special case of the Turing test where participants only discuss recent sporting events seems substantially simpler than the general one. I'm not totally convinced you need GPT-3 level capabilities.
that is fair; my only real complaint is that if the software were just maybe 5% more clever, the headlines it generates probably wouldn't look as bad as they do.
Using facebook's latest https://galactica.org/ i can get wiki pages that are pretty informative summaries of topics that don't yet exist on the net (though not very good yet).
A lot of the blog posts i read here are manually generated blogspam. These timelines these will converge, so i don't see a huge difference .
For the last month or so, on a daily basis, I have been getting the same spam messages (subject line: "You've been chosen”) in my inbox, often classified as important. No amount of reporting them as spam seem to help.
However, what ended up in my spam folder are emails from Amazon informing me about winning $100 in their sweepstakes. (When I found them it was too late to claim the prize as confirmed by Amazon's rep.)
I get that spam is hard, but weighing an account owner's signals more than other signals seems like a good strategy. (And if it is not, please do correct me!)
I have been getting those emails too, and I noticed that they get automatically send to spam ~5 minutes after I receive them.
I seem to be getting a spam email of this kind almost every day for the last ~6 months. Maybe the spammer found a way to prevent fingerprinting by Google and each batch of emails only gets automatically marked as spam after so many people manually mark it?
It has been the same for me. A small but noteworthy fraction of legitimate emails go to Spam and illegitimate emails come into Inbox; in spite of marking "spam" or "not spam" multiple times.
I'm terrified at the thought of all this hollow, low-on-details content that's about to hit the internet. Interestingly, if this degrades Google's search result quality, it could be an opportunity for someone else to create a new search engine.
Honestly, no reason why we will have the same search experience as we do today in 5 years.
I expect search engines to be more like an AI assistant.
I don’t think Google can pull that off. Not because they don’t have the expertise for it (they do) but because their organisation is built around monetizing ads and an AI assistant might not work with it.
I'm thinking more like some of the demos I've seen of Adept AI, where you type in (or speak) a complex command and it does the job for you. Like telling the assistant to "find flights to New York between 15th and 18th December that are under $250".
Regardless, if the future of search is less a list of pages and more a set of focused, customized results, I can't see how Google will work in that context. Google, so far, has shown zero ability to monetize anything that can't be papered over with ads.
> I'm terrified at the thought of all this hollow, low-on-details content that's about to hit the internet.
Honestly it feels like "hollow, low-on-details content" hit the internet at least 5 years ago. Developments in indexing and searching have incentivized writing of low-quality content that makes it to the top of results, and here we are
Already buulding one. Assuming content like this will be ad-supported (what is the other way to monetize it? nobody would pay to read it) Kagi is already successful at downranking sites with a lot of ads.
I mean, it works ok. I think the issue here is structured documents like that wikipedia article. If you used smmry on each section, leaving the headers and structure alone, you'd probably get a far more 'coherent' result, and it's not surprising that the tool is built for "single block" content.
It would probably not be very hard to build a mediawiki smmrly extension that does just that.
I noticed Grammarly just entered the AI summarization space for email. https://www.grammarly.com/recap. I haven't tested it yet, but saw the Gmail banner last week.
I found that in my current job I don't so much need a summarization service, but a translation service; what does the writer actually want me to know and to do? I got an email the other day with a lot of brain farts, grainy screenshots, and I had to read it five times just to boil it down to "change this tracked URL to that one".
I've seen the benefit of AI for completing sentences, or lines of code, and I can imagine the benefit to creative writing where things do not have to be factually correct, however writing in Notion is typically writing about something in the real world.
The problem with AI text generation is that it does not have that real world context. A great example is the first example on this page: writing a blog post announcing Notion's AI generation.
The AI text generator does not know the feature set, and therefore either the content is going to lack any detail about the product, or it's going to have incorrect detail, in both cases providing no value.
What are use-cases for this that people will actually benefit from? Creative writing - sure. Brainstorming? Maybe for very vague and basic topics that don't need any business context. But more than that?
I'll use github copilot as an example, it pulls context from your surrounding code files and code in the same file. Something like that gives it the context your thinking of and a similar technique may be able to solve that problem here.
I think there's a big difference between matching up variable names and the context of what you're working on in a file, and saying "brainstorm 5 marketing strategies for me" like in the examples here. With code, most of the context Copilot needs to do a good job is a) written down and b) parseable in very well defined formats. With business contexts most of the context won't be written down, or won't be in a structure that is useful, or won't be parseable.
Let's take the example of the marketing brainstorming. Let's say you're a super organised small business and you write down a report on every marketing experiment you try, and influencer marketing didn't work for the product, so you write a report on why TikTok creators are not a good direction for marketing.
To know not to suggest influencer marketing, Notion needs to: have all your documents, know that document is relevant, have the "idea" (let's brush past the fact this was essentially written on someone else's brainstorming list and learnt from), know that the idea is related to that other document(s), know that the conclusion was negative sentiment in the way that matters, and then know that because it's doing brainstorming, lists of ideas should probably be filtered by things that haven't worked in the past.
Maybe it's not the worst thing if it generates "influencer marketing", but if you ask it for 5 and it always gives you the 5 you've done before, then you start having to engineer your way around its stupidity – give me 10 ideas, or give me 5 ideas that don't include xyz. Is it providing value over, say, googling "small business marketing ideas" and reading an article full of ads titled "Ten things that will transform your small business marketing". That's low value content, but it seems most likely to me that this will produce largely similar results.
First off sorry totally missed you had replied and you took time to write a thoughtful response. I think your skepticism is completely warranted, but it sounds like a space where it’s worth a shot. If I’m wrong no biggy this is low risk, but if I’m not that’s a lot of upside.
I’ve seen copilot do some truly novel code synthesis, more than just matching up names. I’ve also seen GPT 3 generate compelling marketing copy and many other things so I’m admittedly biased in favor of potential solutions like this.
I could see a use for having the AI describe the generalities (some of which could be provided in the prompt) and then have a human go in and fill in specifics or correct inaccuracies. Either way it amounts to a laborsaving tool, it's much easier to read than it is to write.
I see your point, although I'm not certain I agree. It's very easy to read what we hope to see, especially when we know the topic, rather than what the text actually says. This is why proof reading is so hard for the author to do. Copilot has a similar issue, pushing more effort from code-authoring to code-review, which I'd suggest humans are probably worse at.
> What are use-cases for this that people will actually benefit from? Creative writing - sure. Brainstorming? Maybe for very vague and basic topics that don't need any business context. But more than that?
Looking at my Linkedin feed: lots of lazy people excited about possibility of "creating content" like "100 blog posts on a given topic in a minute". So, in short: seo spam. Lots of seo spam.
Look into Jasper, the company was founded 18 months ago and just raised a 125M series A. The more impressive bit is they have 80k paying customers (their cheapest plan is $30 a month).
Who uses it? Mostly content marketers. GPT-3 is actually quite handy when writing because you can give it a rough outline of something and it can do a pretty good job making it sound good.
As for the lack of context, there's no fundamental reason you can't give these tools information to work off. In fact most of them ask you to.
I'm quite bullish on AI assisted writing - it feels like the internet in 94. For now no one understands it, but the fundamental value add is enormous.
AI is the new 3D printing. Cool novelty that will undoubtedly change the world somehow, but I hope we avoid letting it become this creation that is played out within 5years, overapplied, and shoved into any process whether it improves the product or not because its "the future". I'm about two AI support bots from snapping.
I know this wasn’t your point, but I’m a 3D printing enthusiast and I follow what’s going on in the industry.
I wouldn’t consider 3D printing played out, but I would say that it’s become more boring in the last few years. This is a good thing because it means that it’s being sustained by real work (e.g. medical, prototyping, micro-manufacturing) rather than hype.
I do agree that there was a hype phase, though, and the AI hype phase will probably be similar if not bigger.
Highly skeptical of AI applications like this, unless you have a personal active learning model that you can refine over time. Otherwise, it can only really answer broad canned questions like "10 ways async communication is more productive". As the other commenter pointed out, seems like it would be way more useful to use AI to summarize existing content and suggest links than to pretend like it can do the work of an employee with zero context.
I'm skeptical too, but at the very least this gets buzz with somewhat limited development and maintenance effort. Summarizing your content and building personal models is expensive, hard to debug and support, and it opens the possibility of attacks from competitors who "don't process your data" or "build models off customer data" (even if properly secured/isolated.)
I work for a company in the blogging space (not AI related), and we are seeing huge uptake on AI content generation. Something tipped in the last month or so.
For most bloggers, the hardest part of a creating a piece of content is starting it - staring at the blank screen. At best, they are using AI to jumpstart themselves, then they can flesh it out, rewrite, etc. Human nature being what it is though…
Overall, I think it is effective enough to be a complete disaster for the internet.
Definitely locks the user in, pushes engagement up, probably pushes the creation of new "blocks" up, etc. Defends against google docs doing it to them?
Classic issue with Universal Links, assuming you're on iOS. I implemented Universal Links for an app a few years ago. I think many developers think it's trivial to implement, follow the one-page guide to setting it up, and think the feature is complete. In reality it's much more complex to do for an app of any real size.
They probably need a wildcard in their routing to open a webview for stuff like this they can’t handle in the app. But yes universal links are tough to get right. Easy initial implementation and then surprise bugs forevermore :)
I am wondering the same thing. From their website: "Your data is encrypted at rest and in transit." (see: https://www.notion.so/help/security-and-privacy). I am not really sure what that means.
Decoding for all "we take your security seriously" pages on the web:
In transit means https (like your comment getting sent to this page).
At rest means the underlying store uses a key, like S3, or even HDD encryption. It's encrypted on write to the disk, and decrypted on read. This may be fantastic for when nation state black ops sneak into the data center and remove the server's drives for forensic analysis, useless for anything else. (To be fair, it's useful on a laptop. Not so useful in a data center.)
Neither of these means it's encrypted between reading disk and sending over HTTPS.
Neither one means employees, servers, or clients, can't see your data.
Thanks for the clear and detailed response! It's what I feared it meant. It's a pity they need to use such deceptive language, I find.
I really liked notion as a product. This is kind of a wakeup call for me that using the cloud pretty much means giving up control - and not just theoretically.
Looking at their privacy policy page [0], they mention that customer data is used for "Developing new products and services and improving the Services". So by using the service you give them rights to fine tune on your data.
Who's bearish on AI used like this? Seems like everyone is packaging the cutting edge work/research and putting into this same "bucket".
Will it go the way of 3D tvs? Interesting for 2 years then gone? We'll see.
Not to mention these companies are 100% training their "bot" with your data. Notion will learn from user's notes. Replit will learn from their new "bounty marketplace" they launched. Github from the repositories. Etc.
Sad to know that instead of a proper API or offline support the notion team has decided to spent their time on AI.
Notion turned 8 this year and it still doesn't have the aforementioned offline support, no repeating dates and events and no plugin support.
Why companies prioritise a product nobody asked for instead of features which a large majority of users are vocal in support of is beyond me. I can only hope a competitor will force them to shift their focus.
Less snidely, it's probably a lot more interesting to poke at AI integration than building another REST API. I bet there are more programmers putting their hand up to "explore that area" at the stand ups.
Yeah we've lived in a good job market. Can't assume it'll last forever but giving programmers some autonomy is likely the best way to move a software company forward so I wouldn't be surprised if that continues
I doubt they walk into work every morning and pick from a big board on the wall, but if you don’t have some influence over the kind of projects you’re working on, you should get a new job.
> Less snidely, it's probably a lot more interesting to poke at AI integration than building another REST API
People love to work for developer focused companies with a lot of freedom, but in all honesty most paying users do not care in the slightest what you are interested to put your hand up for.
Because AI generates attention and therefore new users/subscribers. Stuff existing subscribers want is a lower priority. Also, the developers probably think AI is cool to work on and offline support isn't.
Agreed, their search (literal string matching only) was outdated long before they were ever a company, but they do this instead of sticking one mid to senior level DS on that?
This is also almost certain to lead to more documents, not better ones, making the abysmal search experience worse.
Unfortunately every product focused AI/ML/DS team out there is currently churning out these thin wrappers over GPT3, I know because my PM can't stop pushing this nonsense.
The issue is that because a lot of non-technical people haven't played with GPT3 too much, there is huge initial engagement with these "features" so the stats around usage look great, even though anyone with a few brain cells can identify that the root cause of this is the novelty effect.
PMs also get the mistaken belief that, because writing is hard for most people, it must be a big blocker for these marketers. But for a marketer writing copy is like writing code for a seasoned dev. Similarly copilot is fun to play with, but I don't know any developers out there who seriously use it (save your replies I know some enthusiasts must exist). Additionally, unlike copilot, these GPT3 copy outputs always only look decent at first glance, but contains enough oddities that they are rarely if ever useful in their displayed form.
Expect floundering "AI" teams to churn this for out en mass at any remotely content related SaaS company for a while.
Honestly even the examples on the promotional page are quite bad.
I’m a big believer in AI as an accelerant for working, especially after playing with Stable Diffusion. But these examples are like almost wasting my time not making me work faster.
Based on the two examples I looked at on the posted page, one was a brainstorm ideas list for how to promote Notion AI which was fairly obvious. It did not include anything novel, just literally spit out what a quick Google search for “ideas to promote a new product” would have. I suppose you could say that it saves you the search but if you aren’t someone who understands promotion inherently then you’d need more info in the first place and if you are then this list is unnecessary as you’ll already have a promotion strategy in mind.
The second example was the AI writing a blog post introducing Notion AI. This is what I really referred to as a noise machine. The reason you need a blog post like that in 2022 is not for people to read but for Google’s indexer to read so that it gives your page and site all the points and you get better SEO. In other words the whole concept of this has gone off the rails where we have computers writing articles for other computers to read, which will ultimately be fed to another “AI” to learn from to write better articles for computers to read. Why are we perpetuating this circle jerk? Wouldn’t it be better to fix some core issues?
Lastly, I think the tech that allows you to turn a sentence or two into a coherent article is quite exciting. But I think this is somewhat the wrong application for it.
Writing is one field that I think won't be able to achieve the 'human' touch anytime soon. Art? Sure, as we have already seen with DALL-E and the like. Music? Doubtful (look up Kandrake-Z for an example), but not sure if it will expand beyond plunderphonics and noise. But writing? Meh.
I don't know if anyone even uses content-generating AIs like this for writing - but I'd be glad to change my mind by seeing some hard numbers. I mean, except for incomprehensible content-generated articles that pop up on searches only because of PageRank's flaws.
I'm very wary of all these companies building their offerings on GPT-3. It's dependent on OpenAI until it's cheap enough to host your own instance, and then they wouldn't have a proper service to offer anyway.
Also, GPT-3 completion is pretty hacky, and anyone who needs it on an information organization pipeline is not taking it seriously anyway.
I will join others to conclude that, being VC backed it what makes you do things like this and is a signal that company will work on more shallow features and focus on growth for growth sake.
Notion is solid app and idea and it would be a shame to became just one more thing that is just noise.
But it understands your work habits! Also, it uses machine learning to understand your work habits! This makes you more productive by understanding your work habits.
Yeah I thought at first the video was skipping or rewinding or something when I tried to play-pause-play my way though it as it kept repeating essentially the same thing over and over.
Honestly the web link to iOS app connectivity across the board seems completely broken these days. I always just end up in the home screen of the app in deep linking to.
332 comments
[ 3.2 ms ] story [ 296 ms ] threadThis one should be promising!
I honestly think this is how we reach this "dead internet" meme. We're probably less than a decade away from this.
EDIT: here's my bold prediction. In the next 10 years, a very successful content aggregator (imagine Reddit, Instagram or Tiktok), where content is exclusively and openly AI generated, will reach the Alexa top 100 ranking.
We will have HN users saying we're too old and we just don't get why it is more fun to mindlessly get your dopamine fix from scrolling and liking AI generated images.
I argue that it is. Human content farms don't create long-form articles, artwork or deep fake videos by the truckload, in minutes, from just a little prompt. And humans are relatively expensive.
AI generated spam will be orders of magnitude cheaper, faster and thus incredibly more common.
This is where I see natural language AI causing a real kerfuffle. We're probably already reading bot spam articles without realizing.
This is clearly not actually using any process that could be called "AI" with a straight face since it's obviously just playing madlibs with different sets of vocabulary based on the scores and margins, but the results were invariably stupid and repetitive headlines and articles.
Are you suggesting that humans do not create spam too?
But humans create spam too!
A lot of the blog posts i read here are manually generated blogspam. These timelines these will converge, so i don't see a huge difference .
For the last month or so, on a daily basis, I have been getting the same spam messages (subject line: "You've been chosen”) in my inbox, often classified as important. No amount of reporting them as spam seem to help.
However, what ended up in my spam folder are emails from Amazon informing me about winning $100 in their sweepstakes. (When I found them it was too late to claim the prize as confirmed by Amazon's rep.)
I get that spam is hard, but weighing an account owner's signals more than other signals seems like a good strategy. (And if it is not, please do correct me!)
Is this for real?
Are you joking? Surely this is not something that exists.
I seem to be getting a spam email of this kind almost every day for the last ~6 months. Maybe the spammer found a way to prevent fingerprinting by Google and each batch of emails only gets automatically marked as spam after so many people manually mark it?
I expect search engines to be more like an AI assistant.
I don’t think Google can pull that off. Not because they don’t have the expertise for it (they do) but because their organisation is built around monetizing ads and an AI assistant might not work with it.
> I don’t think Google can pull that off.
Does Google Assistant not count? Searching for things with the assistant is certainly one of its main features.
Regardless, if the future of search is less a list of pages and more a set of focused, customized results, I can't see how Google will work in that context. Google, so far, has shown zero ability to monetize anything that can't be papered over with ads.
All the search results are from experts, and have been proofread and stuff. None are ad-supported, and very few are paid-placement.
Honestly it feels like "hollow, low-on-details content" hit the internet at least 5 years ago. Developments in indexing and searching have incentivized writing of low-quality content that makes it to the top of results, and here we are
I’ve been toying with the idea of making a browser extension that summarizes pages into ~5 bullets, <20 words each. And maybe a picture.
Someone please make this.
Too lazy to read walls of text. Too lazy to build the extension myself.
I imagine you just need to wrap that in a browser plug-in. Or maybe already exists.
[1] http://autotldr.io/
[2] https://smmry.com/
Try for example https://en.wikipedia.org/wiki/Functional_programming/ as an input to smmry - you don't get the idea
It would probably not be very hard to build a mediawiki smmrly extension that does just that.
https://tldrthis.com/
https://www.summari.com/products/chrome
Etc etc
[looks like whatever you're writing has already been written in this workspace]
Of course no one will lay them off for that.
The problem with AI text generation is that it does not have that real world context. A great example is the first example on this page: writing a blog post announcing Notion's AI generation.
The AI text generator does not know the feature set, and therefore either the content is going to lack any detail about the product, or it's going to have incorrect detail, in both cases providing no value.
What are use-cases for this that people will actually benefit from? Creative writing - sure. Brainstorming? Maybe for very vague and basic topics that don't need any business context. But more than that?
Let's take the example of the marketing brainstorming. Let's say you're a super organised small business and you write down a report on every marketing experiment you try, and influencer marketing didn't work for the product, so you write a report on why TikTok creators are not a good direction for marketing.
To know not to suggest influencer marketing, Notion needs to: have all your documents, know that document is relevant, have the "idea" (let's brush past the fact this was essentially written on someone else's brainstorming list and learnt from), know that the idea is related to that other document(s), know that the conclusion was negative sentiment in the way that matters, and then know that because it's doing brainstorming, lists of ideas should probably be filtered by things that haven't worked in the past.
Maybe it's not the worst thing if it generates "influencer marketing", but if you ask it for 5 and it always gives you the 5 you've done before, then you start having to engineer your way around its stupidity – give me 10 ideas, or give me 5 ideas that don't include xyz. Is it providing value over, say, googling "small business marketing ideas" and reading an article full of ads titled "Ten things that will transform your small business marketing". That's low value content, but it seems most likely to me that this will produce largely similar results.
I’ve seen copilot do some truly novel code synthesis, more than just matching up names. I’ve also seen GPT 3 generate compelling marketing copy and many other things so I’m admittedly biased in favor of potential solutions like this.
An h1 (or h2 I suppose) on the site is "Let Notion AI handle the first draft."
That makes sense! That's a lot easier than starting with a blank page.
Even if the AI is writing nonsense, you still benefit by virtue of Cunningham's Law.
I see your point, although I'm not certain I agree. It's very easy to read what we hope to see, especially when we know the topic, rather than what the text actually says. This is why proof reading is so hard for the author to do. Copilot has a similar issue, pushing more effort from code-authoring to code-review, which I'd suggest humans are probably worse at.
Looking at my Linkedin feed: lots of lazy people excited about possibility of "creating content" like "100 blog posts on a given topic in a minute". So, in short: seo spam. Lots of seo spam.
Who uses it? Mostly content marketers. GPT-3 is actually quite handy when writing because you can give it a rough outline of something and it can do a pretty good job making it sound good.
As for the lack of context, there's no fundamental reason you can't give these tools information to work off. In fact most of them ask you to.
I'm quite bullish on AI assisted writing - it feels like the internet in 94. For now no one understands it, but the fundamental value add is enormous.
I wouldn’t consider 3D printing played out, but I would say that it’s become more boring in the last few years. This is a good thing because it means that it’s being sustained by real work (e.g. medical, prototyping, micro-manufacturing) rather than hype.
I do agree that there was a hype phase, though, and the AI hype phase will probably be similar if not bigger.
For most bloggers, the hardest part of a creating a piece of content is starting it - staring at the blank screen. At best, they are using AI to jumpstart themselves, then they can flesh it out, rewrite, etc. Human nature being what it is though…
Overall, I think it is effective enough to be a complete disaster for the internet.
Are generative text models and image models cheap enough for consumer products or do you need to charge a premium?
Together with Notaku [0] i can now create blog posts and help articles much faster, choosing Notion for my websites CMS was the right idea
[0] https://notaku.so
In transit means https (like your comment getting sent to this page).
At rest means the underlying store uses a key, like S3, or even HDD encryption. It's encrypted on write to the disk, and decrypted on read. This may be fantastic for when nation state black ops sneak into the data center and remove the server's drives for forensic analysis, useless for anything else. (To be fair, it's useful on a laptop. Not so useful in a data center.)
Neither of these means it's encrypted between reading disk and sending over HTTPS.
Neither one means employees, servers, or clients, can't see your data.
I really liked notion as a product. This is kind of a wakeup call for me that using the cloud pretty much means giving up control - and not just theoretically.
Notion may, however, use your Content or Customer Data to improve and train Notion’s own models.
https://www.notion.so/Notion-AI-Program-Terms-c0066e30039041...
> Marvin from Product Operations here and happy to clarify. The answer you're looking for is: no we are not using your data to train our models.
"We do not and will not use your Content or Customer Data to improve or train our models unless you give us express permission to do so."
Maybe they changed this in the past 7 hours?
[0] https://beta.openai.com/docs/guides/fine-tuning
[0] https://www.notion.so/Privacy-Policy-3468d120cf614d4c9014c09...
Will it go the way of 3D tvs? Interesting for 2 years then gone? We'll see.
Not to mention these companies are 100% training their "bot" with your data. Notion will learn from user's notes. Replit will learn from their new "bounty marketplace" they launched. Github from the repositories. Etc.
I get why it’s fun to make a system like this, but does it really never occur to anyone that no human wants to read AI generated copy?
Notion turned 8 this year and it still doesn't have the aforementioned offline support, no repeating dates and events and no plugin support.
Why companies prioritise a product nobody asked for instead of features which a large majority of users are vocal in support of is beyond me. I can only hope a competitor will force them to shift their focus.
Less snidely, it's probably a lot more interesting to poke at AI integration than building another REST API. I bet there are more programmers putting their hand up to "explore that area" at the stand ups.
https://news.ycombinator.com/item?id=33620662
People love to work for developer focused companies with a lot of freedom, but in all honesty most paying users do not care in the slightest what you are interested to put your hand up for.
This is also almost certain to lead to more documents, not better ones, making the abysmal search experience worse.
The issue is that because a lot of non-technical people haven't played with GPT3 too much, there is huge initial engagement with these "features" so the stats around usage look great, even though anyone with a few brain cells can identify that the root cause of this is the novelty effect.
PMs also get the mistaken belief that, because writing is hard for most people, it must be a big blocker for these marketers. But for a marketer writing copy is like writing code for a seasoned dev. Similarly copilot is fun to play with, but I don't know any developers out there who seriously use it (save your replies I know some enthusiasts must exist). Additionally, unlike copilot, these GPT3 copy outputs always only look decent at first glance, but contains enough oddities that they are rarely if ever useful in their displayed form.
Expect floundering "AI" teams to churn this for out en mass at any remotely content related SaaS company for a while.
As of November 8, Notion supports recurring events in databases: https://www.notion.so/releases/2022-11-08
Curious what you think is missing from the API? https://developers.notion.com/reference/intro
I’m a big believer in AI as an accelerant for working, especially after playing with Stable Diffusion. But these examples are like almost wasting my time not making me work faster.
https://news.ycombinator.com/newsguidelines.html
Based on the two examples I looked at on the posted page, one was a brainstorm ideas list for how to promote Notion AI which was fairly obvious. It did not include anything novel, just literally spit out what a quick Google search for “ideas to promote a new product” would have. I suppose you could say that it saves you the search but if you aren’t someone who understands promotion inherently then you’d need more info in the first place and if you are then this list is unnecessary as you’ll already have a promotion strategy in mind.
The second example was the AI writing a blog post introducing Notion AI. This is what I really referred to as a noise machine. The reason you need a blog post like that in 2022 is not for people to read but for Google’s indexer to read so that it gives your page and site all the points and you get better SEO. In other words the whole concept of this has gone off the rails where we have computers writing articles for other computers to read, which will ultimately be fed to another “AI” to learn from to write better articles for computers to read. Why are we perpetuating this circle jerk? Wouldn’t it be better to fix some core issues?
Lastly, I think the tech that allows you to turn a sentence or two into a coherent article is quite exciting. But I think this is somewhat the wrong application for it.
I started working on GPT3 integration for Notion [0]; but I guess that was obvious to have AI integrated in Notion.
The question is when will we get AI generated images in Notion? Well, then I can shut down my side project [1]
[0] https://twitter.com/kiru_io/status/1572290853103865856
[1] https://slashdreamer.com/
https://scribe.citizen4.eu/@beingpax/how-to-use-gpt-3-ai-wri...
I don't know if anyone even uses content-generating AIs like this for writing - but I'd be glad to change my mind by seeing some hard numbers. I mean, except for incomprehensible content-generated articles that pop up on searches only because of PageRank's flaws.
Everyone is building Mechanical Turks to nudge users to add artistic nuance AI lacks so far.
Personally I prefer using the libraries directly rather than gift my work to startups I’m also paying subscription fees to.
Startups are basically double dipping; let us train based on your behavior! …for $9.99/mo per user. Oh and we own the outcome!
Hoping to get some useful open models and data sets for game world generation soon!
That's said, the spell checker is nice and the translation would be helpful too.
Also, GPT-3 completion is pretty hacky, and anyone who needs it on an information organization pipeline is not taking it seriously anyway.
Notion is solid app and idea and it would be a shame to became just one more thing that is just noise.
You can sum up the whole blog post in one sentence.
Notion AI helps you be more productive by understanding your work habits and providing you with suggestions on how to improve them.
It writes like how I did when I was a kid in elementary school writing English papers lol.
Sounds like some next-gen Lorem Ipsum.
Here I thought that was a meta comment on the notion of AI.
Poems? Really?