3 ms·
Show HN: Deep Learning Personas
- julien_c 10y agoHey guys. Don’t expect accurate answers here - it’s an experimental weekend hack. With this experiment, we wanted to understand how we could train a neural network on new external datasets very quickly and study the difference in the language and answers depending on the topic of the dataset. No scripted chatbot here, we are talking of neural-nets trained in the wild :) We are using the celebrated seq-2-seq model which computes a “thought-vector” from an input sentence and generates an output sentence conditioned on this vector. We gathered various datasets from stackexchange and launched a big overnight training of our models to have some surreal morning coffee talks with our AI. There’s still a ton of work for the answers to start to make sense (longer training and bigger datasets would improve the quality of course, we can also easily add components to improve the variety and coherence of the responses) but a difference can definitely be noticed based on the dataset subset. This will allow us to test a variety of new datasets way faster than before. Let us know what you think, and if you think of a cool dataset to test, just let us know on Github.
- aduffy 10y agoDo you have any details on your experience training the models? Was this using vanilla RNNs, LSTM cells, did you use an attention mechanism, etc.? How long did you train for?
- bbayer 10y agoIt can enter a loop easely. Gamer girl vs. me me: yes gg: I'll send screen shots me: yes gg: I'll send screen shots me: yes gg: I'll send screen shots me: yes gg: I'll send screen shots ....
- julien_c 10y agoResponses to a specific message are deterministic in this model (no diversity promoting objective implemented) so this is to be expected :)
- sprobertson 10y agoAre these separate models, or are you supplying a persona vector alongside input? I've found the latter to be a very simple way to condition an RNN, e.g. https://github.com/spro/practical-pytorch/blob/master/char-rnn-generation/char-rnn-generation.ipynb https://github.com/spro/practical-pytorch/blob/master/char-r...
- malux85 10y agoQ: "How many electrons in an atom of Lithium?" A: "Of course not!" - Physics expert
- dangelov 10y agoHonestly, this gives such irrelevant answers, it might as well be random. The presentation is pretty cool but it doesn't actually work well at all.
- nyrulez 10y agoRight now, it doesn't work at all. Would be great to test and iterate on it a bit and to have some sense to the answers before exposing it to the world.
- luiscape 10y agoI asked about gravitational waves. The Physics Genius eventually answered with "Yes, a white trashbag and nothing else". http://imgur.com/kLtsooB http://imgur.com/kLtsooB
- FnuGk 10y agome: ruby on rails Full Stack Dev: Do n't do this
- polygot 10y agoThem: Yo! Does it compile? Me: Yes Them: I'll update the tag Me: Ok thank you Them: The second one then.