Papers of the day   All papers

Towards a Human-like Open-Domain Chatbot


Quoc Le: New paper: Towards a Human-like Open-Domain Chatbot. Key takeaways: 1. "Perplexity is all a chatbot needs" ;) 2. We're getting closer to a high-quality chatbot that can chat about anything Paper: Blog:

28 replies, 1795 likes

Google AI: Check out Meena, a new state-of-the-art open-domain conversational agent, released along with a new evaluation metric, the Sensibleness and Specificity Average, which captures basic, but important attributes for normal conversation. Learn more below!

25 replies, 1299 likes

Kyle McDonald: um.. google's latest chatbot is 😳

23 replies, 1250 likes

Jeremy Howard: Amazing work. PS: we're all so screwed.

16 replies, 791 likes

Hugging Face: The 101 for text generation! 💪💪💪 This is an overview of the main decoding methods and how to use them super easily in Transformers with GPT2, XLNet, Bart, T5,... It includes greedy decoding, beam search, top-k/nucleus sampling,...: by @PatrickPlaten

12 replies, 637 likes

Thang Luong: Introducing #MeenaBot, a 2.6B-param open-domain chatbot with near-human quality. Remarkably, we show strong correlation between perplexity & humanlikeness! Paper: Sample conversations:

14 replies, 354 likes

Oriol Vinyals: seq2seq still delivers : ) Incredible how far we've gotten in ~5 years of progress in neural conversational models, with relatively small changes. More exciting is that there's still LOTS to be done! Paper: Blog:

4 replies, 289 likes

Daniel Adiwardana: Enabling people to converse with chatbots about anything has been a passion of a lifetime for me, and I'm sure of others as well. So I'm very thankful to be able to finally share our results with you all. Hopefully, this will help inform efforts in the area. (1/4)

2 replies, 169 likes

Thang Luong: sample-and-rerank as proposed in #MeenaBot ( is a simple strategy that gives both diversity and quality for the generation, especially if your model has low perplexity!

0 replies, 137 likes

Matthew Kenney: 😂😂 this had me in stitches

2 replies, 98 likes

Tim Hwang: 👀 "We were unable to find this in the data"

3 replies, 71 likes

Károly Zsolnai-Fehér 📜: Google’s Chatbot: Almost Perfect ▶️Full video (ours): 📜Source paper: #ai #deeplearning #science #twominutepapers #meena

1 replies, 49 likes

Dan Nguyen: didn't anticipate the dad joke industry would be first victim of automation

1 replies, 46 likes

Erik Brynjolfsson: Impressive work by @GoogleAI. Chatbots like Meena are increasingly able to provide sensible and specific responses. It will be harder and harder to tell whether we are talking to another human or a machine. HT: @jeremyphoward #AI #MachineLearning #2MA #chatbot #Election2020

6 replies, 36 likes

Quoc Le: @xpearhead @lmthang My favorite conversation is below. The Hayvard pun was funny but I totally missed the steer joke at the end until it was pointed out today by @Blonkhart

2 replies, 35 likes

Trieu H. Trinh: Had the chance to sit next to Daniel @xpearhead in the early days of the project and tried out the interactive Meena. It has always been *this* surprising and funny :) BIG Congrats to the team with this publication. The possibilities to build up from here is endless.

1 replies, 34 likes

Μatthew Alampay Davis: Google's new chatbot has developed an avant-garde sense of humor that its creators mistakenly believe is a bug

1 replies, 29 likes

ΜΔDΞRΔS: #CyberpunkisNow "Meena" is a chatbot Google says conducts near human-level conversation. "SSA" is a Google metric to grade a chatbot's capacity for conversation/judgement vs an average human: Avg Human 86% SSA Other Chatbots 31% to 56% SSA Meena 79% SSA

1 replies, 26 likes

Quoc Le: @xpearhead @lmthang You can find some sample conversations with the bot here:

1 replies, 26 likes

Aran Komatsuzaki: Towards a Human-like Open-Domain Chatbot: Near-human level open-domain chatbot is achieved with a Transformer LM with 2.6B parameters. Notably, they showed there's a strong correlation between ppl and human-likeness.

0 replies, 21 likes

Károly Zsolnai-Fehér 📜: Google’s Chatbot: Almost Perfect 🤖 ▶️Full video (ours): 📜Source paper: #ai #deeplearning #science #twominutepapers #meena

0 replies, 18 likes

davidberreby: 'Do horses go to Harvard?' a human asked a neural net. 'Horses go to HAYvard' it replied. h/t @jackclarkSF, who reports Google is wary for now of releasing code for a bot this good at sounding human. (Other firms have been similarly cautious)

0 replies, 18 likes

Jason Wu: [Towards a Human-like Open-Domain Chatbot] "We trained our best model for 30 days on a TPUv3 Pod (2,048 TPU cores) on the Meena dataset containing 40B words." This is not institution-anonymous... LOL

1 replies, 15 likes

Christian R. Jänsch: Meena — a new open-domain chatbot (with a 2.6B parameter NN) pioneered by Google🔝 @theomitsa @postoff25 @Victoryabro @Droit_IA @baski_LA @mvollmer1 @jrzaranoid @AshokNellikar @sallyeaves @ShiCooks @evankirstel @techpearce2 @data_nerd See also:

0 replies, 10 likes

Dat Tran: Pretty impressive! Google AI just released Meena, a 2.6bn parameter neural-network chatbot that is able to conduct human-like conversations. Very interesting results and paper. Check it out! 📝 Paper:

0 replies, 9 likes

Jelle Zuidema: Impressive looking example conversations and numbers from Google AI's new chatbot Meena. Uses the Evolved Transformer, lots of parameters, lots of data, distillation and good old perplexity as the objective.

1 replies, 9 likes

Rose Eveleth ▷▷: okay but this is actually a really funny joke?

2 replies, 8 likes

Şerife (Sherry) Wong: Holy guacamole Batman this chatbot is outstanding in its field.

1 replies, 5 likes

Delaney Hall: I love a good multi-turn joke in an open-domain setting

0 replies, 5 likes

BioDecoded: Towards a Conversational Agent that Can Chat About…Anything | Google AI Blog #NLP #chatbot

0 replies, 4 likes

Connor Leahy: Oh boy

1 replies, 4 likes

Victor Dibia: New results from Google on open domain chat bots

0 replies, 3 likes

Edward Dixon: A stunning #NLP paper from @GoogleAI on a chatbot that manages a pretty neat joke that really does seem to be original (see below, "Horses go to Hayvard").

0 replies, 3 likes

Marcus Borba: New paper: Towards a Human-like Open-Domain #Chatbot. Key takeaways: 1. "Perplexity is all a chatbot needs" ;) 2. We're getting closer to a high-quality chatbot that can chat about anything Paper: Blog: via @quocleix #ML #AI

0 replies, 3 likes

Gerardo Montoya aka Rénròu Sōusuǒ: 86% vs 79% #SujetosHíbridos

0 replies, 2 likes

Radamés Ajna: this is getting interesting

0 replies, 2 likes

Ashwin Ram: Making progress towards human-like open-domain Conversational #AI that can chat about anything, along with a new evaluation metric that captures key elements of good conversation. Nice work by @quocleix and team at @GoogleAI. Paper available on @arXiv.

0 replies, 2 likes

_n ⚔️: Suddenly thousands of dads are like "shit we gotta up our joke game"

0 replies, 2 likes

M.O. Gewaltig: The race of brute-force conversation bots continues...

0 replies, 2 likes

Think Tech Ai: Google unveiled Meena earlier this year. Apparently it's the best chatbot ever. LIke ever. Anyone using it? Anyone having any luck? #GoogleMeena

1 replies, 1 likes

Mitsuku/Kuki 🏅🏅🏅🏅🏅: @MarisaTPP @TweetinChar The graphic was on this announcement by Google. The research paper can be found here:

0 replies, 1 likes

Moja Afroughi: Great work!

0 replies, 1 likes Google released Meena, an end-to-end, neural conversational model that learns to respond sensibly to a given conversational context. Source: The extracts from the conversations in the paper ( ) are impressive, see second image 😲

0 replies, 1 likes

Epsilon Guanlin Lee: Wow, wonder how this compare to XiaoIce from MSRA @znculee

0 replies, 1 likes

Dr.Sultan Al-Yahyai: #عالمنا_في_4IR_ربيع_2020

0 replies, 1 likes

RainbowSecret: Wow!

0 replies, 1 likes

Joshua Baker: Comedians, Google’s chatbots are coming for you

0 replies, 1 likes

Manish Chablani: Great insight from recent google paper on chatbots: "Meena has a single Evolved Transformer encoder block and 13 decoder blocks. We discovered that a more powerful decoder was the key to higher conversational quality."

1 replies, 1 likes

arXiv CS-CL: Towards a Human-like Open-Domain Chatbot

0 replies, 1 likes

Ward Plunet: Seems to be a jump forward in terms of human like conversations/chats.

0 replies, 1 likes

Carl Carrie: Say hello to Mena "Meena was trained on a whopping 341 gigabytes of public social-media chatter"

0 replies, 1 likes

Dr Jochen L Leidner: Adiwardana et al. (2020) describe #Google 's new conversational agent ("#chatbot") #Meena, a 2.6 billion parameter end-to-end trained neural conversational model #AI #artificialintelligence #chatbots #conversationalagents

1 replies, 1 likes

Raphaël Millière: New from @GoogleAI: "Towards a Human-like Open-Domain Chatbot" Sample conversations are really impressive (although cherry-picked).

1 replies, 0 likes


Found on Jan 28 2020 at

PDF content of a computer science paper: Towards a Human-like Open-Domain Chatbot