Papers of the day   All papers

Recipes for building an open-domain chatbot

Comments

Facebook AI: Today we’re announcing that Facebook AI has built and open-sourced Blender, the largest-ever open-domain chatbot. It outperforms others in terms of engagement and also feels more human, according to human evaluators. https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot https://t.co/1wo3ZxwDeU

28 replies, 1133 likes


ParlAI: We have studied recipes for large-scale open domain chatbots, and are releasing 90M, 2.7B and 9.4B parameter models with SOTA results. Paper: https://arxiv.org/abs/2004.13637 Project: https://parl.ai/projects/blender/ Blog post: https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot https://t.co/pcCcCiVRLC

18 replies, 608 likes


hardmaru: Facebook AI used 1.5B Reddit posts to create a chatbot “Good conversation requires a number of skills: providing engaging talking points, listening, displaying knowledge, empathy and personality appropriately, while maintaining a consistent persona.” https://arxiv.org/abs/2004.13637 https://t.co/HYHqo33hUq

16 replies, 499 likes


Jerome Pesenti: Really proud to announce the open source release of Blender by @facebookai. The largest ever open-domain chatbot which feels more human than any other bot according to human evaluators. Looking forward to what the AI community can build on top of it! https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot

8 replies, 341 likes


Stephen Roller: Really excited to be sharing this with everyone today. Blog post below, paper here: https://arxiv.org/abs/2004.13637

5 replies, 213 likes


Yann LeCun: Blender: large-scale, open-source, open-domain chatbot from FAIR.

12 replies, 209 likes


jaseweston: Been self-hiding in our caves, working hard on this.

3 replies, 74 likes


Olivier Reynaud: 🤖 What a great week in #AI world ! 1. @OpenAI Jukebox generation of music https://openai.com/blog/jukebox/ 2. @TensorFlow 2.2.0 released https://twitter.com/TensorFlow/status/1256301213340979200?s=20 3. @facebookai open domain chatbot https://twitter.com/parlai_parley/status/1255527273219919872?s=20 4. YOLOv4 released for object detection https://medium.com/@riteshkanjee/yolov4-superior-faster-more-accurate-object-detection-7e8194bf1872 https://t.co/g6Dznz49dq

1 replies, 60 likes


Cyril Diagne: Wow that's one witty bot

0 replies, 44 likes


Joscha Bach: a word completion algorithm pretending it understands jokes

7 replies, 42 likes


Ethan Perez: These conversations are really impressive. Some even remind me of my research meetings with @kchonyc: https://t.co/jA6n7QsIxH

4 replies, 40 likes


MIT Technology Review: Facebook has open-sourced a new chatbot that it claims can talk about nearly anything—and in an engaging and interesting way. https://bit.ly/2Smvhzk

1 replies, 33 likes


roadrunner01: Recipes for building an open-domain chatbot pdf: https://arxiv.org/pdf/2004.13637.pdf abs: https://arxiv.org/abs/2004.13637 project page: https://parl.ai/projects/blender/ https://t.co/n3m9BtONWE

0 replies, 30 likes


Alexandre Lebrun: I've been in #NLP for 20 years -- first time I need to double-check who is the human, who is the bot.

2 replies, 19 likes


Thang Luong: Nice progress from FB! Glad to see that #MeenaBot has motivated further research into training chatbots at scale: now Blender is up to 9.4B params. Though not a direct apple-to-apple comparison, Blender seems to excell in engagingness & humanness. Paper: https://arxiv.org/pdf/2004.13637.pdf.

0 replies, 18 likes


Daniel Fernández: blue one is a bot. A BOT. https://arxiv.org/pdf/2004.13637.pdf https://t.co/9WL3LEaTap

2 replies, 17 likes


Xavier- staying 🏡 + 😷 saves lives: "Recipes for building an open-domain chatbot" by @facebookai - ". "prior work has shown that scaling neural models in the number of parameters & the size of the data they are trained on gives improved results, we show that other ingredients are important" https://arxiv.org/abs/2004.13637

0 replies, 16 likes


ParlAI: We have studied recipes for large-scale open domain chatbots, and are releasing 90M, 2.7B and 9.4B parameter models with SOTA results. Paper: https://arxiv.org/abs/2004.13637 Project: https://parl.ai/projects/blender/ Blog post: https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot https://t.co/VzzZg51Ae3

0 replies, 9 likes


David Graus: These examples 😳

1 replies, 5 likes


jaseweston: Been self-hiding in our caves, working hard on this.

1 replies, 5 likes


Alan Nichol: Nice! And with models we can actually play with 🚀👀

0 replies, 4 likes


Peter Wang: ELIZA is growing up... Soon it will just be Her

1 replies, 4 likes


Donald Clark: This is a huge leap in terms of scale and flexibility of chatbots...

0 replies, 3 likes


PyTorchLightning: Parlai team doing really cool things: https://parl.ai/projects/blender/

0 replies, 3 likes


Steve Weis: “Recipes for building an open-domain chatbot" from Facebook AI: https://arxiv.org/pdf/2004.13637.pdf

0 replies, 2 likes


®️AJ A 🗣👨🏻‍💻: Facebook AI open-sources Blender, the largest-ever open-domain chatbot Paper: https://arxiv.org/abs/2004.13637 Codes: https://parl.ai/projects/blender/ #AI #chatbots #ArtificialIntelligence #DataScience

0 replies, 1 likes


ParlAI: We have studied recipes for large-scale open domain chatbots, and are releasing 90M, 2.7B and 9.4B parameter models with SOTA results. Paper: https://arxiv.org/abs/2004.13637 Project: https://parl.ai/projects/blender/ Blog post: https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot https://t.co/zeesCUVhkC

1 replies, 1 likes


A Wojcicki: Bewildered by banality of the dialogue on the left, I'm thinking that up until now we were concerned either about malicious AI or indifferent AI (darn paperclips!), but we should focus on the dangers of creating dull AI.

1 replies, 0 likes


Content

Found on Apr 29 2020 at https://arxiv.org/pdf/2004.13637.pdf

PDF content of a computer science paper: Recipes for building an open-domain chatbot