Hacker News new | past | comments | ask | show | jobs | submit login
[dupe] Generate Hacker News Comments from Titles (leod.org)
138 points by phoe-krk 30 days ago | hide | past | web | favorite | 80 comments

“Generate Hacker News Comments from Titles” yields

- A nice little hack and I think it would be better to have a bit more info on the title. It would show a much more comprehensive description and a little more info on the actual title.

- I don't know enough about this topic, but it makes me wonder why it's so popular in some other forum. It appears there is a lot of noise in the community, so why not just show up at the bottom of the page, or go to the top of HN, rather than showing up, to get to the bottom of the page?

- I find the title on the front page to be a bit clumsy. Maybe a change to "HN News." might help somewhat, I suppose. EDIT: The title is a bit misleading. Is this some kind of a "submitter" with pg saying "Show HN" because of some sort of 'Show HN'?

These are actually believable

Lot of noise though, we are far from passing the Turing test in most case but still fun:

>Nice idea from a hacker perspective. I'd like to point out the following two issues

- Your results for your comments seem a hell of a lot to me, for example, but not the comments that you linked to.

- Your results for comments in particular appear in the top left of the comments. But the comments appear on the right of the comments.

- You need to be looking at what your titles are up to and what you are actually looking at and how other titles are. I have to click through the comment page to get to the first page.

I just tried it and it works.

These sound like actual comments from people with an incoherent train of thought, which might be plenty of people on the internet. These are scary accurate. I wonder how easy it would be to write a bot to, for example, post on Reddit and gain karma for the purposes of trusted-account generation. It seems like most of those account farms rely on self-upvoting which can be caught easily by the site. Posting potentially good comments that get upvoted by real people would be pretty hard to counter.

Who has ever looked at a reddit profile and decided that an account should be trusted based on karma? Recent few comments are way more likely to determine trustworthiness.

Hoarding karma on reddit is as useless as hoarding karma on HN. I've deleted the only reddit profile I've ever had a few days ago with about 45k karma. If they had any value, I would maybe keep it and just stop posting.

> Who has ever looked at a reddit profile and decided that an account should be trusted based on karma?

For starters, automated subreddit spam protection algorithms. You may have misinterpreted me, think of it like StackOverflow karma. Your account needs a certain amount before it can do certain things.

Yes, maybe like +200 (which is what I've set it at as a mod in severe cases of brigading), not thousands. Most of the subreddits don't require anything but positive karma, and you can achieve that with less than a dozen manually created bots upvoting the comments that new bots make.

There are no subreddits with a spam algorithm counted in the thousands of karma, that would prevent them from growing.

> with less than a dozen manually created bots upvoting the comments that new bots make.

As I said in my other comment, these self-upvoting schemes are very easy to detect and are not a scalable way to create tens of thousands of bots.

In the black market of reddit accounts, age and karma are the two main selling points.

> we are far from passing the Turing test

I'd propose it's an unproven theory that comment sources other than this software would pass the Turing test.

When given "Elon Musk Dies", the model outputs:

"It's easy for everyone to believe that Musk was the real reason behind the death of the planet. He was the first guy to believe the same as we did. But this seems so obvious that many people have started believing it. I'm convinced it was because the world was so big that it's so easy for people to believe that."

Maybe the model knows something we don't.

> It's easy for everyone to believe that Musk was the real reason behind the death of the planet. He was the first guy to believe the same as we did.

Robot, no

> pace.js : pacemaker controller software implemented in javascript


> Looks interesting, although I'm still waiting for a way to make my own pacemaker using Javascript

> For anyone else who wonder what the fuss is with the pacemaker API, I wrote a library that was ported to iOS, Android and the iOS platform back in 2012. It was a very fun way to play with pacemakers in iOS. The codebase is just small so any time you try to use it on your Android or iOS device, you want to check it out and watch your progress. The idea is to add a couple of API calls to the iOS code to access pacemakers, then use another library to access the pacemaker software on your iOS device. This is the kind of thing that makes the browser a bit better. This way you can use the API to do everything you want via a browser widget, and then use the API to generate pacemaker apps.

NN-generated text has this really weird uncanny valley feeling to it, where the words are well organized but there’s no cogent thesis or point to be made. I feel like I might be reading schizophrenic ramblings.

No point here necessarily, this stuff is fascinating but creeps me out. Words from generated comments (using this website) are organized in a way that cause me to second guess my own reading comprehension abilities. It’s kinda neat and fascinating.

The real challenge here: did I write this myself or modify a recommended NN-generated comment? I’ll award points to the winner.


> A word salad is a "confused or unintelligible mixture of seemingly random words and phrases",[1] most often used to describe a symptom of a neurological or mental disorder. The words may or may not be grammatically correct, but are semantically confused to the point that the listener cannot extract any meaning from them.

> I feel like I might be reading schizophrenic ramblings.

Or religious pamphlet.

Interesting that is what I felt without being able to put words on it.

When you read the text's beginning it feels like real text but after reading some sentences you feel something is way off. And it's abit creepy.

second paragraph sounds like an hncynic comment with "comment" in the title.

I tried: "Facebook announces sending man to Mars" and got this:

I'm impressed that they have finally addressed the issues with the mission to Mars. I hope this isn't just a marketing push from FB, but a huge PR push.

then this:

I was wondering how they do it: the man is going to be in a position to land on Mars. Maybe not in Mars, but in the atmosphere of the Moon, this would be the most useful thing I have ever seen. And I guess the question is, can it be done with Mars?

then this:

The headline is a bit misleading. Facebook has no plan to send a man to Mars. The article points out that Facebook are planning to send a woman to Mars.

I couldn't help but laugh aloud. Good work!

SpaceX announces sending man to Mars (wired.com)

hncynic 1 minute ago

The most important detail was the last paragraph. The actual announcement for the launch of the Falcon 9 is here: http://www.spacex.com/releases/index.html

This is a huge leap forward for SpaceX to make it into space again. SpaceX will still have a large stake in the rocket, but that's a huge leap forward for the whole industry. I wish SpaceX would just stop [... redacted ...] be more aggressive and look at their competitors.

Getting some interesting syntax ideas from the headline "rust version 3.2 released":


"For example, let's say a function is like this in C++.

- a function that has a string of type (value: value: value: value, value, value: value) - a function that returns an address in C++'s form: value: value: value: value: value[value:user]:{value[user]:}], value = value[user]}}: value[user]:{value[user]:}}

I'm still not quite clear on what does this mean to me. In Rust, you have a function that returns a value: value, value, value = value[user-]: value[user-]: value[@@"

"How to be happy"


How to be happy?

1. Do something meaningful, 2. Do something meaningful...3. Do something meaningful...4. Do something meaningful...5. Do something meaningful--(even if that is not your primary goal)--5. Do something meaningful...6. Do something meaningful...(eg if you are a software engineer). 7. do something meaningful in life, not just some sort of job or hobby, to keep your sanity up. 9. Do something meaningful-- not what you think you're doing.

I think this is a good starting point; but the author is right.

I think the real key difference between happiness and pleasure is your own self-actualization and that it is something you can accomplish in the long run.


^ I know natural language generators can kind of get "stuck" sometimes, but this is pretty funny. (Also, some wisdom at the end).

This is by far my favourite one.

Buffalo buffalo.

I typed in “hacker news comment generator” and was disappointed to not find “I could build this in a weekend” as one of the outputs

Reddit has you covered: https://i.imgur.com/xa1MDnt.png

hncynic 1 minute ago

"Emacs can do the job!", "Emacs is faster than Emacs" "Emacs is faster than Emacs because it runs inside an emacs environment"

This is absurdly accurate I have to say


    Linus Torvalds elected as the new president
    of the United States

    Now all he has to implement is version
    control. So I can have my own branch.

This is really useful.

Some suggestions:

1. Add a "Download this site's CSS" field so it can't just be a "Show HN". The "Show HN" button should be a red flag.

2. "If the site is a startup, we can always add some comments to the title." That's one reason I'm not using http://ycombinator.com/newsguidelines.html

Did you just...respond with a comment generated by the site about its own title? Or is this a real comment that I just don't understand.

the fact that you couldn't tell just shows how eerily brilliant this is.

Currently near the top:

> Smoke seen for miles as SpaceX Crew Dragon suffers anomaly at Cape Canaveral


hncynic 1 minute ago

The problem is that it was a sign of an unstable airplane, as in the case of the original comets. The only way to protect it is with the helium and oxygen and liquid nitrogen.

I don't really care, but I can see no reason to be afraid to do that.

hncynic 1 minute ago

For the first 10 minutes there was a huge difference between the reported issue being a bug in the test software and the SpaceX failure.

The second time that happened was a second time to the SpaceX crew with their tests and another second time to the Space Shuttle team with their test suite. It was a different time to the launch of the test software, but that's it.

The two times the SpaceX crew did that, the first time they did this, they would have to run their tests with a new test/test suite.

I am wondering if it's the first time the SpaceX crew received a test suite with the mission at all.

hncynic 1 minute ago

I was always in the Space Shuttle business and they had a shuttle that would burn and land the crew on a very low orbit and then use it to do the launch. They were pretty upset that if they ever came back to space, they would probably still be at SpaceX, at least for the rest of my life.


I find it strangely realistic. When I skim it kinda sounds like HN. You have to actually read to suspect anything. If I was a bit mischievous I might actually post one of these in the other article.

So how is this done?

I've got:

> > hncynic 1 minute ago

> The article didn't say anything. But it just says that the launch failed, and that it did not have a recovery time.

> So to me, this sounds more like an anomaly of some kind to explain it better than a normal anomaly of the event.

> What it means to say that the failure caused is exactly the same thing that happened when the spacecraft started at the right time, except it didn't have a recovery time and was unable to recover.

> How long is an anomaly like this from space-X supposed to take?

This is a comedy generator.

The previous thread had some interesting example outputs:


>Very cool project but how is this different than HN comments?

Is this a real output from the generator? Too good to be true

"Hacker News to Merge with Reddit" resulted in:

>> hncynic 1 minute ago

Hi everyone,

Thanks for your contributions to the community. I hope you like this more in the future. I was able to post a link to the blog post I just posted in the last few weeks and it has helped a lot.

I believe it will help to create a community so more people can use Hacker News to connect with interesting people. I think it also helps people get better at finding cool stories on their own instead of doing it themselves.

>> hncynic 1 minute ago

If I recall correctly HN has been a "mersy" forum for many years already.

>> hncynic 1 minute ago

Wow! Now we have the option to merge the whole HN community, even better.

Title: PayPal Sucks

- It sucks because the author is not even a PayPal user or is not an account owner and only doesn't seem to have been involved in an online business since then.

-As someone who has lived in the US for 2 years, I've noticed that the biggest problem I have with PayPal is how difficult it is to manage your account in general. The PayPal account is set up as a backup to a central server, so you can use whatever tools you need for the job. So you set up a small number of accounts which you may have to mangled yourself. You do get to be really careful what you do with the money you pay with your account. At any rate, at some point, I just switched from using PayPal for some things - I want an account that I can use to pay for my business and have a good experience with them.

-Yes, PayPal sucks on a daily basis. I am an account user (and a regular member of our site) and have also dealt with several PayPal/Dwolla accounts using it. I had to do some googling and it seems very helpful. I don't know how they work with PayPal's customer support - I understand that they are trying to be like PayPal - but I had to look at the documentation to see what actually happened. I can understand having a few customers who will not have any problems with them, but as a customer, this was extremely frustrating, just frustrating.

For the title "AI personal tutor for Auto-Didacts" this gives:


> This seems to fit the trend of people using AI as a tool as much as possible to save time.

> If you want to save time, just don't try to do something about it. A good AI's ability to reap more benefit is not measured in dollars/days of training per se, but rather, a more accurate one.

> Wow -- this is really interesting, I was expecting it to look like a robot taking a long walk around the world and going to work.

Anyway, I like the idea.

The big challenge I see with the auto-didacts is that the people who are doing it will have more data to look at and can do it quickly. I think if there are any intelligent bots, I'm looking at making this much harder for the people who are doing it quickly. But it would be great if they found a way to make their algorithms so easy to follow and make them more productive.

As other posters have pointed out, this type of thing is not for everyone. It just might be to make your computer do the work on your own time, and to keep things consistent and understandable.

I love the idea of an interface that has to match the time someone else is doing it in a similar manner. I don't know about others and if it would work or not, but I think I want to think of a way to see them learn from each other.


Wow. :o

Once you stopped using the quote symbol, I thought it was your comment and spent too long trying to understand it as such.

Yeah, that's why put the result in between two lines. HN can have at least a bit better text-editing support though.

It made sense once I looked at it again, it wasn't a complaint more a praise of the generated comments!

Input: "Elon Musk to Star in Next Iron Man Movie'

>> As an aside, here are some other movies in this directory.

* The Iron Man and Aliens ( http://www.imdb.com/title/tt010248/ )

* The Mars One ( http://www.imdb.com/title/tt0638888/?ref_=tt_1 )

* The Mars Orbiter ( http://www.imdb.com/title/tt009608/ )

>> So Musk got a job at Apple and then worked on it at Apple for a week before he was told the movie went live.

>> This movie is really about this as an advertisement for the movie. Just so you guys know, this guy is making a better movie by talking about the movie not about the movie but just the movie, and how he has a lot of different experiences.

Top Indian female biker dies in road accident during country-wide tour hncynic 1 minute ago

It says "India's top female bike deaths in road trip during the world’s most dangerous years".

Did he mention the death rate in China?


I would be interested to know more before writing this.

hncynic 1 minute ago

"She was a motorcycle helper. She was a motorcycle driver, and while she wasn't riding on a bicycle, she was on a bicycle. She was on a bike. When she drove on it, she was wearing an underwear on her head."

This is an incredibly unfair statistic. As far as I can tell people who use bikes don't even know how many times the "bicycle helper" is used.

As a bike enthusiast, I never have any problem riding as a bike helker, but in the city as a cyclist I regularly spend almost double the time riding bicycle helper in my town and have to be absolutely sure that she is riding on a bike that does not have a collision.

So while that is a big thing, I think the majority of bikers should take advantage of that knowledge to do so.

I can see how many motorcycles are in the US and other cities with very high density bike lanes, the article seems to imply that the city is doing something right as a biker. I don't really know about that, but I'm guessing it's just a matter of how many years a bike helper will take you to the point where you're not doing it wrong, or how

hncynic 1 minute ago

There is nothing to be proud of with a single biker at any moment.

> I would be interested to know more before writing this.

Good thinking there. We all could learn from the generator.

> Ask HN: How to start a consulting practice?

> "I need clients to work with me and help me grow" -> "They don't need me to help me grow". -> "I'm not sure what I do because they are working with me but my experience suggests that I shouldn't do that with me" -> "But in practice I can work with me as my client as long as I stay in touch"

> Just start a side project on github and make sure you are the kind of person who is interested in this kind of thing and don't just make a "Show HN".

> I think a lot of the comments here are focusing on this one question - how about just working on one thing that you don't have any experience on? Do you feel like you are going to go out of your way to do it at all - what have you done at any length to get the experience in the way that you need?

Not bad at all. Some improvements and this can become my goto HN website. (only partly kidding).

It apparently doesn't do well with titles that ask questions (e.g. Ask HN) with no backing article. This was trained on titles that are assumed to contain linked article, so I wouldn't expect it to do well on these:

Title: "What is your favorite color?"


> Kill me with my iPhone.

Title: "What is the worst language?"


> This is probably one of the more insidious comments about the article that pops up.

> It was a really good article.

> But this one is absolutely stupid.

> I could understand a few simple examples:

> * Explicit arithmetic

> * Explicit arithmetic

> * Selectable numbers

> * Concrete math

> * Consistent arithmetic

> * Concurrency... or something.

> "Yes," by the way is an interesting comment.

> "Yes... but I think we all have this problem, and I can't help feeling that we could all do just that: We need to do something about it. And so what..."

> I do like how he presents that question. I would be much more interested in an answer like that.

Title: What is the worst language?

hncynic 1 minute ago

> The most difficult languages to learn are C, Java, C++, and C, which in their implementation would have helped prevent them from becoming "worse" languages

Nope, but still more challenging than "easier".

> Lisp

No, because that's not so hard...

> Ruby, Python, and even Haskell

Nope, but Lisp is the least painful language to learn.


Lets say I want to program on a Mac. I want to program on a Mac. The only reason I need to do that is that the code base sucks and you can't use PHP.

hncynic 1 minute ago

I don't think this guy is wrong. I think that's the best language to write web applications. I personally have spent a fair amount of time building web applications. I think people have a hard time focusing on web apps. If you can use Java or PHP or Python, you can do web applications too.

I copied a random title from the front page to test with:

>Show HN: Tetris Implemented in ClojureScript

The results were eerily good:


hncynic 1 minute ago

I just finished a version of Tetris (http://clj.net/clj.htm) in ClojureScript. It took me about 2 hours before building the full game in Clojure.


Very cool! Is there a link out there to see how this compares to other Tetris implementations ? I'm a little confused as to why this would be different from the "Teach Yourself JavaScript" or whatever.


The demo is a bit too slow because the game is in a weird state but it's in a state where it needs to be replayed.

It would be useful to see the comments and get an idea of the quality of each title. This site does a good job of conveying what is good enough while still providing a summary of the contents and submitting an opinion.

How do you get your comment title into the list?

You can just use a few of the existing ones: http://hn.algolia.com/hacker

It's not perfect, but it works fine

(all three comment above generated by OP's site)

Randomly entering "China" into the title makes the model oddly obsessed with the U.S. :

- The U.S. has a similar problem

- The same can be said of US citizens in countries with strict immigration policies and very few foreign nationals.

- I am really not sure how this is different to what is happening in other countries/countries and how it has changed.

- The US is one of a few countries the US was supposed to be a democracy. Not only that but there is a reason that it's no longer the case.

"Musk says SpaceX will start selling tickets to Mars"

- I thought the title was going for Elon's marketing and the fact that one of the people behind this site is probably one of the least technically competent people I've ever met.

- It seems like this is a common tactic in the aerospace world. The first thing I always thought was that spacex was going to go in space like space and the cost wouldn't be a problem.

- He probably won't make it.

Good work, the github repository is also clean and easy to follow.

May I ask : What is the size of the live model of the website, in MByte ? (or maybe it would be nice to add that info in the README, just to have an idea)

Also, would be interesting to see how it compares to the recent GPT-2 (https://github.com/openai/gpt-2)

Thank you!

The model weighs in at 1.2GB with 100M parameters, which is similar to the smallest GPT-2 model.

I wouldn't be suprised if GPT-2 small (+ finetuning on HN data) performed better than what I have trained. Other than hyperparameters, I think there are two main differences: First, I pretrained the model solely on Wikipedia data, while GPT-2 used more general web data. Second, I used an encoder-decoder model, while GPT-2 is a language model. I'm suspecting that the encoder is not very useful for this task.

Funnily enough it can generate links to genuine HN comments. eg.

> hncynic 1 minute ago

> https://news.ycombinator.com/item?id=13471054

> https://news.ycombinator.com/item?id=13778525

I got "https://hn.algolia.com/?q=Hacker+News" as one of the comments for a couple of searches. You may want to remove the Algolia search link from your training data - which is at the bottom of every thread page.

Input: "Python Escapes From Local Zoo'

>> I have to wonder what Python was doing as a kid and what was the point of making this happen?

>> This kind of stuff is why python3 can't be adopted by all programmers (no matter how hard it's to get things done).

>> I can't believe that there were no escape codes or excuses?

Nice, and I like it too -- but it lacks the comments that are so useful for me. I've found a ton of good ideas to use on sites that I've just bought.

I've had a lot of fun making a script similar to this that would convert your site to a commenting system. Thoughts?

This is really fantastic! How do you generate these kind of comments?

I have the same question but without the title. Maybe I can generate the "Ask HN: Hacker News Comments" link from titles by tagging it by topic.

Took a few tries before something fun came up, but this one blew me away (OP title as input).

> Wow, the quality of the comments is impressive. But it also shows the top comments that have appeared from time to time.

"This should become part of the UI of Hacker News"

Is the generator still being trained based on current comments?

If so, how long will it be before it starts generating comments that include feeding the title init itself?

Would be handy if it could generate the article too.

Just sprinkle some “experts say” and “the community responds” in between.

Hilarious, I got comments complaining about how "the article is from 2011" and "the linked article is more thorough."

This might just be the morning dumbs, but in this trip to metaspace I find it mighty hard to tell forged comments from authentic ones.

For real. It’s kinda disconcerting.

just entered "seo test" and the garbage out was 100% spot on! congrats!

- why didn't you write it in Rust?

obligatory test:

Generate hacker news comments from titles

hncynic 1 minute ago

I thought it was going to be more like /. but I'm confused. Can you add some sort of comment link to /, which is something I could have easily navigated, like /. and /. would be great.

hncynic 1 minute ago

Nice, but I wonder why you would want to do the same with comments on the same page. If you're not really getting it, why make a single sentence and the title different?

hncynic 1 minute ago

How about adding a "Downvote" to the title? For example, "Hackernews: The best comment in the world: http://news.ycombinator.com/item?id=635686

"What I like about the HN thread, which is that I don't have to post links here to get upvotes on comments from hacker-news posts"

I just wanted a way to filter out the "Ask" from your comments so you could add new comments, and then maybe change the URL to "Show HN" to the "About the Hacker News thread" link.


> What you need to add to this is an official Hacker News API.

Capable of philosophical investigations also:


The world is everything that is the case.

hncynic 1 minute ago

The author claims the same thing is true of software, but not with software itself.

hncynic 1 minute ago

This is great article. When I read "Taught in 30 Minutes" I was amazed at how the author seems to think that it was a "fantastic story" that was written in an unrealistic way - that the author has been able to write his book on his own (yet untrue). The author doesn't seem to be aware of the actual circumstances behind the creation of a book.

hncynic 1 minute ago

Great post. This is really the kind of article I don't think the author talks about in his blog post.

TL;DR - the world is everywhere.

I was very disappointed with the title, but here we are.


"n-gate.com is hiring"

> Seems a bit off topic, but I have an idea for a service that gives people the benefit of the doubt. How would it work? How would it work?

> This sounds like a bad idea. There might be some sort of "hire a few engineers" thing.

> This would appear to me a lot more valuable than a list of job postings like http://www.n-gate.com (it only has the name 'n, but you might get the email from them)

The comment above mine was written by AI and you will never find out.



Cute idea, but I can't seem to get anything decent out of it.

I tried the following:

> Massive strike at oil rig as radical turtle lovers declare independence from federal regulations against shelled creatures

> 1 + 1 = 2

> 1 + 1 = 3

> Cats rule teh interwebz

> US elects Jon Skeet for President; challenger Chuck Norris too busy counting to infinity infinite times to notice.

> xkcd is funnier than late-night comedy

On the upside, the titles about math did include commentary on math, while the title that references xkcd included a link to one of the comics.


I know you're joking, but think of the implications if these comments actually didn't differ from Reddit comments. It would then be easy for a small group of people to push agendas on Reddit. Imagine being able to automate creation of posts that push a certain agenda, then being able to automate thousands of coherent comments that agree with the sentiment of the post. You'd be able to manipulate thousands of real people into buying certain products, thinking a certain way, etc.

I'm not advocating for black-hat application of the technology, but merely pointing out that this is what we are facing as AI gets better.

Ahah, the comment I wrote was the result of the algorithm on its own title.. Guess it's working great :-)

I could do better with the Ad-lib library I wrote as a freshman in High School. At least those generated comments where coherent. The garbage that is being generated here is an incoherent mess. Deep learning is going to be another blow against the AI community like in the 80's when expert systems didn't pan out. Once ML is well understood a few well defined algorithms will replace all the buzz, just like expert systems.

ML as we know it today is not going to usher in Sky Net.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact