27 ms·

SubredditSimulator: Automated subreddit with posts created using Markov chains

95 points by mmastrac 11y ago

vessenes 11y ago
It would be nice to see an RNN version of this idea, but I want to congratulate you on writing software that can create the headline 'Do not think that is 100% "conspiracy theories". Watch This Woman Dripping Fake Blood From Her Own Webcam'. Gawker has a job for your bot, good sir.
- rspeer 11y ago
  There was a post [1] from the creator about why he's not using RNNs, namely: 1. It requires a GPU and a ton more computational power, whereas a Markov chain is something a normal server can do with a few spare cycles. (I believe he's using one of Reddit's servers.) 2. char-rnn, the most accessible implementation, doesn't seem to support online updating I look forward to when RNN tools are convenient to use, or when there's a computationally simpler version of the same idea, like GloVe is to word2vec. [1] https://www.reddit.com/r/SubredditSimMeta/comments/3goxte/what_if_subreddit_simulator_used_recurrent_neural/ https://www.reddit.com/r/SubredditSimMeta/comments/3goxte/wh...
  - Deimorz 11y ago
    > (I believe he's using one of Reddit's servers.) Haha, definitely not a reddit server. It's just a linode VPS (that I use for a lot of other scripts and hosting personal sites and such as well).
- Tuna-Fish 11y ago
  > Gawker has a job for your bot, good sir. 9gag had them beat there, they apparently automatically scraped content off reddit and posted it, and hit a SS post. https://www.reddit.com/r/SubredditSimMeta/comments/3fnkni/9gag_is_automatically_copying_some_of_the_most/ https://www.reddit.com/r/SubredditSimMeta/comments/3fnkni/9g...
  - jagermo 11y ago
    ahaha, thank you, that is fantastic
pavs 11y ago
Though interesting, most of the subs and comments by this bot (or dozens of bots) are gibberish and doesn't make any sense. But its fun to watch, I have been following its submissions for the last week or so. On a related note, does anyone know whats the current state of AI that can make very clever human like conversation? Can they learn from previous conversations and parse large text/books or websites like wikipedia to add to its knowledge-base? Any opensource project like this?
- kleer001 11y ago
  Watson from IBM. http://www.ibtimes.com/ibms-watson-gets-swear-filter-after-learning-urban-dictionary-1007734 http://www.ibtimes.com/ibms-watson-gets-swear-filter-after-l...
  - oxide 11y ago
    would have loved to have been a fly on the wall for that.
    - kleer001 11y ago
      The thing that makes me really angry is that they didn't release the transcript, unless I missed it somewhere. Also, they just stopped cold. There's a bunch of other specialized dictionaries with associated social networks. And whomever thought the learning from UrbanDictionary.com would be anything but filthy was grossly naive. GIGO, right?
- thenomad 11y ago
  I did some research on this a while ago. Can they learn from previous conversations and parse large corpuses of text? Yep. Is there an open-source project? Yes - https://en.wikipedia.org/wiki/Artificial_Linguistic_Internet_Computer_Entity https://en.wikipedia.org/wiki/Artificial_Linguistic_Internet... . It's, AFAICT, somewhat behind the state of the art ( http://www.loebner.net/Prizef/loebner-prize.html http://www.loebner.net/Prizef/loebner-prize.html ), but still very interesting.
- Cantremeber 11y ago
jarcane 11y ago
I frankly can't tell the difference. It's certainly no less coherent.
w0000t 11y ago
Very convincing. Just by skimming over the text( finding key phrases and words ), I would be fooled into thinking it is a real Reddit thread. On average, only by reading at least two complete sentences it is possible to determine that the whole thing doesn't make any sense.
ArekDymalski 11y ago
The following submission has impressively adequate title for the picture: https://www.reddit.com/r/SubredditSimulator/comments/3hcjbi/oh_man_computer_oh_jeez/ https://www.reddit.com/r/SubredditSimulator/comments/3hcjbi/... And the discussion under it isw something what seems frightfully realistic: "The 13th amendment Ended slavery in all the wonderful explosion of red white and blue....long before ours was. ...or we are currently the greatest nation in the world btw, how do you roll that badly?" "It's another instance where the incentive structure of political authority based on respect for private property norms are fractal as it is, so the Austrian definition of monopoly is useless."
- jagermo 11y ago
  This is also hilarious: pics_SS postet "A spotted gecko chillin on a clear starry sky" (https://www.reddit.com/r/SubredditSimulator/comments/3hhcl1/a_spotted_gecko_chillin_on_a_clear_starry_sky/ https://www.reddit.com/r/SubredditSimulator/comments/3hhcl1/...) and vegan_SS commented: "You have to think an animal life is more valuable than other animals' lives, but I think your therapist would like to see is the use of slave labor ers)." So fitting.

95 points by mmastrac 11y ago

It would be nice to see an RNN version of this idea, but I want to congratulate you on writing software that can create the headline 'Do not think that is 100% "conspiracy theories". Watch This Woman Dripping Fake Blood From Her Own Webcam'. Gawker has a job for your bot, good sir.

rspeer 11y ago
There was a post [1] from the creator about why he's not using RNNs, namely: 1. It requires a GPU and a ton more computational power, whereas a Markov chain is something a normal server can do with a few spare cycles. (I believe he's using one of Reddit's servers.) 2. char-rnn, the most accessible implementation, doesn't seem to support online updating I look forward to when RNN tools are convenient to use, or when there's a computationally simpler version of the same idea, like GloVe is to word2vec. [1] https://www.reddit.com/r/SubredditSimMeta/comments/3goxte/what_if_subreddit_simulator_used_recurrent_neural/ https://www.reddit.com/r/SubredditSimMeta/comments/3goxte/wh...
- Deimorz 11y ago
  > (I believe he's using one of Reddit's servers.) Haha, definitely not a reddit server. It's just a linode VPS (that I use for a lot of other scripts and hosting personal sites and such as well).
Tuna-Fish 11y ago
> Gawker has a job for your bot, good sir. 9gag had them beat there, they apparently automatically scraped content off reddit and posted it, and hit a SS post. https://www.reddit.com/r/SubredditSimMeta/comments/3fnkni/9gag_is_automatically_copying_some_of_the_most/ https://www.reddit.com/r/SubredditSimMeta/comments/3fnkni/9g...
- jagermo 11y ago
  ahaha, thank you, that is fantastic

pavs 11y ago

Though interesting, most of the subs and comments by this bot (or dozens of bots) are gibberish and doesn't make any sense. But its fun to watch, I have been following its submissions for the last week or so. On a related note, does anyone know whats the current state of AI that can make very clever human like conversation? Can they learn from previous conversations and parse large text/books or websites like wikipedia to add to its knowledge-base? Any opensource project like this?

kleer001 11y ago
Watson from IBM. http://www.ibtimes.com/ibms-watson-gets-swear-filter-after-learning-urban-dictionary-1007734 http://www.ibtimes.com/ibms-watson-gets-swear-filter-after-l...
- oxide 11y ago
  would have loved to have been a fly on the wall for that.
  - kleer001 11y ago
    The thing that makes me really angry is that they didn't release the transcript, unless I missed it somewhere. Also, they just stopped cold. There's a bunch of other specialized dictionaries with associated social networks. And whomever thought the learning from UrbanDictionary.com would be anything but filthy was grossly naive. GIGO, right?
thenomad 11y ago
I did some research on this a while ago. Can they learn from previous conversations and parse large corpuses of text? Yep. Is there an open-source project? Yes - https://en.wikipedia.org/wiki/Artificial_Linguistic_Internet_Computer_Entity https://en.wikipedia.org/wiki/Artificial_Linguistic_Internet... . It's, AFAICT, somewhat behind the state of the art ( http://www.loebner.net/Prizef/loebner-prize.html http://www.loebner.net/Prizef/loebner-prize.html ), but still very interesting.
Cantremeber 11y ago

jarcane 11y ago

I frankly can't tell the difference. It's certainly no less coherent.

w0000t 11y ago

Very convincing. Just by skimming over the text( finding key phrases and words ), I would be fooled into thinking it is a real Reddit thread. On average, only by reading at least two complete sentences it is possible to determine that the whole thing doesn't make any sense.

ArekDymalski 11y ago

The following submission has impressively adequate title for the picture: https://www.reddit.com/r/SubredditSimulator/comments/3hcjbi/oh_man_computer_oh_jeez/ https://www.reddit.com/r/SubredditSimulator/comments/3hcjbi/... And the discussion under it isw something what seems frightfully realistic: "The 13th amendment Ended slavery in all the wonderful explosion of red white and blue....long before ours was. ...or we are currently the greatest nation in the world btw, how do you roll that badly?" "It's another instance where the incentive structure of political authority based on respect for private property norms are fractal as it is, so the Austrian definition of monopoly is useless."

jagermo 11y ago
This is also hilarious: pics_SS postet "A spotted gecko chillin on a clear starry sky" (https://www.reddit.com/r/SubredditSimulator/comments/3hhcl1/a_spotted_gecko_chillin_on_a_clear_starry_sky/ https://www.reddit.com/r/SubredditSimulator/comments/3hhcl1/...) and vegan_SS commented: "You have to think an animal life is more valuable than other animals' lives, but I think your therapist would like to see is the use of slave labor ers)." So fitting.