I agree that GPT-3 should not be used for medical applications, but I disagree that it's "not very good" as a dialog system. I've found it to be insanely good though it may require effective prompt engineering to work well.
HN user
nickwalton00
Yep few shot learning is a game changer. You don't get perfect results but you can prototype all kinds of systems extremely fast.
Read the first section and has already been super helpful for a newbie founder. Thanks!
Latitude / AI Dungeon | ML Engineer / Applied Scientist | Remote (U.S.) | Full-Time | https://latitude.io
Latitude is revolutionizing how games are made by leveraging advanced AI. Our first product AI Dungeon (https://aidungeon.io) is an AI powered text adventure where literally anything is possible. We're a small team looking for ML Engineers / Applied Scientist to help us continue innovating on how to transform how games are made.
You can apply by emailing careers+hn@latitude.io
The nice thing about Patreon is it explicitly is structured for tiered subscription support and removes the need for creators to do any payment system managing.
I've had companies straight up pay me for doing a take home project as part of an interview. I think that's the ideal way as it gets rid of the potential for exploitation/excessive burden that can come from take home projects.
Super excited about this! It's crazy to see the progress AI Dungeon has made from its early beginning one year ago to where it is now.
Ai dungeon creator here. Another option is if you can detect what region the colab notebook is in and have a multi region bucket for each international area you could download from the right region and it may be quite cheap. Our costs were primarily from US GCS buckets downloading to colab servers that were apparently running in Asia and Europe.
The unexpected thing was that Google colab and GCS are separate such that transfer costs between them often end up as international external egress fees
I do a small amount of editing the text that goes in and out of the model (saving the last 8-10 actions and results, cutting off hanging sentences, adding "You " to the beginning of actions" but the only world model is the actions and text that's been generated, nothing else.
The game can be remarkably wholesome if your actions are. It's not too difficult to take it in either direction depending on what you do.
I scraped a bunch of stories from chooseyourstory.com but I did curate them to make sure they had the right second person format. I couldn't really anywhere else that had a consistent format that would make scraping easy enough.
I believe I just fixed it! Let me know if there are still issues.
Yeah sometimes the model can get off track and it can be hard to get it back. The best way is just to revert a few actions back by typing "revert"
It's partly because my first version (which you can play at ai-adventure.appspot.com) was white text on a black terminal like background.
Thanks! I have no idea how I would afford to host it myself haha. GPUs don't come cheap!
Hmm it's supposed to cut at end of text tokens. I'll look into that and see what's causing that.
Thank you!
Yep! Nathan did an earlier experiment of this, but it was the smallest gpt model trained on a small dataset of Zork adventures.
I did think about this when I was making it!
Definitely. I would love to see something like this as part of a more comprehensive game.
Haha! I love it
Hmmm... I haven't seen that before there should be enough memory on the GPU to hold the model.
Yeah sorry. I don't know if there's anyway around it. Google colab is simply the best way to give people a chance to play the game without having to pay money or do complicated technical set up.
I never thought about that, but that is pretty funny haha.
I'm glad you loved it! Yeah there's definitely points where the coherence breaks down, but it's an awesome proof of concept of what's possible in future games and pretty fun in its own right.
Yeah it has a very dream like quality. It's definitely more like minecraft in the lack of any defined goal.
The reason it needs to be logged in is because Google Colab is a free service that Google provides allowing people to run code on expensive GPU servers. It doesn't cost you anything and I don't store any information from your google account when you run it.
I agree with you that this doesn't reach near the level of good human authors. There's no long term plot or deep human themes in this. I don't think this will ever replace quality human writing, but it may be able to augment it in cool ways. I personally would love if rather than every guard in Skyrim telling the exact same story, if each guard could have their own stories or comments generated based on things about their life. Human authors could provide high level details and let AI generators fill in the smaller details.
You'd need a GPU as good as the ones google colab offers which is a 12GB NVIDIA Tesla K80. You'd also need cuda and tensorflow installed.