Live data from Hacker News

Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

aidungeon.io

211–220 of 227 posts

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#212
Poetic,

> "who am i?"

Traceback (most recent call last):

  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py", line 1365, in _do_call
    return fn(*args)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py", line 1350, in _run_fn
    target_list, run_metadata)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/client/session.py", line 1443, in _call_tf_sessionrun
    run_metadata)
tensorflow.python.framework.errors_impl.ResourceExhaustedError: 2 root error(s) found.

  (0) Resource exhausted: OOM when allocating tensor with shape[1,48,2,25,465,64] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
  [[{{node sample_sequence/while/concat}}]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

0 successful operations. 0 derived errors ignored.

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#213

This is very entartaining but it's not very good as an interactive story generator. The problem is that it can't keep the state of the story world consistent between generated passages. Essentially, everytime the player is prompted for a sentence to use to advance the story, a new story is generated. To explain what I mean by a state of the story world and its consistency. Suppose we start a story with "Once upon a t…

I'm treating that like a constraint on the player. It's not just part of the challenge -- really, the entire challenge is to provide input that simultaneously advances the story without breaking it.

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#214

Ok so mine was basically soft core porn. I chose a rogue and I spawned in a tavern next to a girl named Aliza. She was super drink and immediately asked me to go up to her room with her. The door was cracked open already and we came in together and she immediately got in the bed and pulled her skirt up as far as it would go. When I didn't respond immediately she got up and put her panties back on but then we got back…

I won the game with "collect a harem of sex slaves".

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#215

Ok so mine was basically soft core porn. I chose a rogue and I spawned in a tavern next to a girl named Aliza. She was super drink and immediately asked me to go up to her room with her. The door was cracked open already and we came in together and she immediately got in the bed and pulled her skirt up as far as it would go. When I didn't respond immediately she got up and put her panties back on but then we got back…

I won the game with "collect a harem of sex slaves".

And that was the only time I've won the game.

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#216

Ok so mine was basically soft core porn. I chose a rogue and I spawned in a tavern next to a girl named Aliza. She was super drink and immediately asked me to go up to her room with her. The door was cracked open already and we came in together and she immediately got in the bed and pulled her skirt up as far as it would go. When I didn't respond immediately she got up and put her panties back on but then we got back…

Damnit, that was about as disappointing as my sexual dreams. Why do I always wake up before there is some action?

I have a theory about that. Sex dreams that finished would be evolutionarily unhelpful.

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#217

Would it be possible to run this on my own machine? I'd love to set up a Discord bot so my friends and I can play this together

The game's Github page explains how to do that.

https://github.com/nickwalton/AIDungeon

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#218
It seems like I've had more success when I go for a more conceptual story. In more personal, individual level stories the continuity of details ect.. get more in the way.

Here's a narrative about governmental policy that was pretty entertaining:

You are Mansle, a noble living in the kingdom of Larion. You have a pouch of gold and a small dagger. You are awakened by one of your servants who tells you that your keep is under attack. You look out the window and see hordes of orcs pouring into the city.

> Build a wall to keep out the orcs

You quickly gather all the resources you can find and begin building a great wall around the city. The walls are not strong enough to stop the orcs completely, but they at least slow them down. They also help protect the people inside the city from harm.

A few years pass.. > Reform the Government to a COmmunist state

The orcs continue to pour into the city, but it seems like they are being held back by the walls. The citizens are still happy with their government, so you decide to change things up a bit. You propose that the government be changed to a Communist one > become the great chairman of the communist party

You immediately take control of the government and make your proposal. All the other leaders agree and vote for you as Chairman. The new system is called the People's Republic of Larion > Pass a zoning law to reduce the height of buildings

This law was passed without any problems. It reduced the height of buildings by half an inch. This made the city much more livable. > Create an internment camp for the drudges of society

You create an internment camp for drudges, because there were too many of them. The drudges were kept locked away in these camps until they died. There were no riots or anything, just a peaceful death > Enact a social credit system

You implement a social credit system. Everyone had to pay a certain amount of money every month. If they didn't pay this sum, then they couldn't do certain things. For example, if someone wanted to go out and buy alcohol, they would have to pay a certain amount of money > Close the borders so that no one can exit the country

You close the borders so that no one can leave the country. The citizens are very happy about this law. It makes the city a lot safer. There is no more crime. > Suspend freedom of the press

You suspend freedom of the press. No one can write anything anymore. Newspapers don't exist anymore either. Instead, everyone has access to the internet. > Take away Internet access

You take away Internet access. Nobody can use the internet anymore. Everyone is very unhappy about this law. You see riots breaking out and you fear that something bad will happen > Quell the riots with force

You quell the riots with force. You send in troops to the streets and try to restore order. Unfortunately, the soldiers aren't very effective. Most of the people are armed with guns and knives. You end up losing most of your men

Re: Show HN: AI Dungeon 2 – AI-generated text adventure built with 1.5B param GPT-2

#220
post #4

A plot that sounds really neat while you're in the middle of it, but afterwards doesn't really make any sense — sounds a bit like a J.J. Abrams production. I don't mean that flippantly, I'm serious. Sounds like GPT-2 might be really good at producing an enjoyable experience, even if there's no underlying structure. What would be really cool would be to add some sort of generated structure as well. Can GPT-2 be used t…

Yeah that's a really good future/research question. I've thought some about how you could do more structured generation. Maybe you could do some kind of reverse summarization, where you take a plot outline and expand out the details would be really interesting to try out? It's still not clear the best way to do structured text generation.

Every DM has had their “plot” smashed by the players...so let’s torture the AI with that. XD
Post reply on HN