got-book-6
AI
text generation
SHOW HN2017
143points
38comments
#274of 2,242 in 2017
Top 13%of 2017 by points
How HN took it
MixedNo clear lean
5 positive, 9 neutral or asking, 0 mixed, 7 critical, out of 21 comments about the project. Thread score -0.02.
Idea: 2 for, 2 againstExecution: 0 for, 1 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Funny and surprisingly coherent output
From 3 comments
This is amazing. I wonder if it'd be possible to actually get a coherent plot out of it while retaining the humorous aspects of it.
notamy on HNThe RNN seems to agree with the most popular fan theory about what will happen with Jaime and Cersei. So, it's not entirely crazy. Nonetheless, I don't think GRR Martin has to…
SwellJoe on HNThis is hilarious! We did a similar project last year using the scripts from the show. http://docs.pachyderm.io/en/latest/examples/tensor_flow/readme.html Ours was not nearly as…
jaz46 on HN
Fun project to see
From 2 comments
The RNN seems to agree with the most popular fan theory about what will happen with Jaime and Cersei. So, it's not entirely crazy. Nonetheless, I don't think GRR Martin has to…
SwellJoe on HNNice! I did a similar thing with a character-based rnn a few years ago, but its output was less coherent. I liked when it started spitting out character descriptions though:…
wbertelsen on HN
Objections
Copyright concerns over distributing the full book text in the data folder
From 3 comments
You need to take down the data folder now and rebuild the repository. You're going to get hit with a DMCA notice. It's not a matter of IF, it's a matter of WHEN. Publishers of…
DoctorNick on HNThat data folder seems ripe for a DMCA takedown request.
rlglwx on HNYou really should not upload other people's complete works while under copyright. Edit: Nice License > Copyright (c) 2017 Zack Thoutt
mrhigat4 on HN
Output is incoherent and looks like a Markov chain
From 2 comments
It does seem like a lot of work to get output that looks like something from a simple Markov chain.
Semiapies on HNWith Markov Chains: "Raved on fire, he clambered up her pallet in a swatch of blood and crept toward him shudder. She was too well Ghost back behind their crews could not lack…
donquichotte on HN
Requests
Add an attention mechanism for better results
From 1 comment
You definitely need to add an attention mechanism if you want better-than-markov-chain results. Found a project ;)
JustFinishedBSG on HN
Add more training data such as fan fiction or scripts
From 2 comments
I wonder if there are any forums/fan fiction websites where you could find big chunks of prose to add to the training data. Might help bring the loss down a bit if you mix some in.
itschekkers on HNThis brings up the interesting question of whether it's ok to publish the trained model and not the data? I'd imagine it's sufficiently derivative as to be allowed? What if you…
alexcnwy on HN
Questions
Is it legal to publish a trained model without the training data?
From 2 comments
This brings up the interesting question of whether it's ok to publish the trained model and not the data? I'd imagine it's sufficiently derivative as to be allowed? What if you…
alexcnwy on HNI believe it would still be a derivative work. https://www.copyright.gov/circs/circ14.pdf
ebp234 on HN
Alternatives named
textgenrnn
From 1 comment
It should be noted that per the included Jupyter Notebook, the resulting network took 24 hours to train, and the resulting network (of a 3-layer 512 cell LSTM) would be hundreds…
minimaxir on HN
Markov chain text generators
From 2 comments
It does seem like a lot of work to get output that looks like something from a simple Markov chain.
Semiapies on HNWith Markov Chains: "Raved on fire, he clambered up her pallet in a swatch of blood and crept toward him shudder. She was too well Ghost back behind their crews could not lack…
donquichotte on HN