Deep Learning Personas

chatbot experiment

SHOW HN2017

Chat with seq2seq neural nets trained on StackExchange datasetsAI

Shown by trueduke on Feb 17, 2017. Click the card to turn it over.

39points

11comments

#912of 2,242 in 2017

Top 41%of 2017 by points

How HN took it

RoastedHalf or more of the comments are critical

0 positive, 5 neutral or asking, 0 mixed, 5 critical, out of 10 comments about the project. Thread score -0.39.

Execution: 0 for, 3 againstPerformance: 0 for, 2 against

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

Answers are irrelevant or nonsensical
From 3 comments
- Honestly, 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.
  dangelov on HN
- Q: "How many electrons in an atom of Lithium?" A: "Of course not!" - Physics expert
  malux85 on HN
- I asked about gravitational waves. The Physics Genius eventually answered with "Yes, a white trashbag and nothing else". http://imgur.com/kLtsooB
  luiscape on HN
Gets stuck in repetitive loops
From 2 comments
- It 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…
  bbayer on HN
- Responses to a specific message are deterministic in this model (no diversity promoting objective implemented) so this is to be expected :)
  julien_c on HN
Not working well enough to be public
From 1 comment
- Right 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.
  nyrulez on HN

Ask about training details and hyperparameters
From 1 comment
- Do 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?
  aduffy on HN

What architecture and training duration were used?
From 1 comment
- Do 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?
  aduffy on HN
Are separate models used or a persona vector as input?
From 1 comment
- Are 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.…
  sprobertson on HN