Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Facebook uses 1.5B Reddit posts to create chatbot

bbc.com

Post reply on HN
Print view View original

91–100 of 235 posts

Re: Facebook uses 1.5B Reddit posts to create chatbot

#91

Post by camjohnson26 »

oefrha wrote: post #20

I fed the script to M-x doctor, and had a nice chat. --- I am the psychotherapist. Please, describe your problems. Each time you are finished talking, type RET twice. Hi how are you today? How do you do? What brings you to see me? Doing well. My favorite food is cake. I just bought one because I got promoted at work! Is it because you got promoted at work that you came to me? Thanks so much, I just want to make my pa…

The example in the article had the same environmental engineer question, is it just memorizing responses and spitting them back?

Re: Facebook uses 1.5B Reddit posts to create chatbot

#92

Post by Cakez0r »

pesenti wrote: post #71

Blog post: https://ai.facebook.com/blog/state-of-the-art-open-source-ch... Paper: https://arxiv.org/pdf/2004.13637.pdf Open Source: https://parl.ai/projects/recipes/ Ask us anything, the Facebook team behind it is happy to answer questions here.

I was really blown away by the results you achieved. Amazing work! My jaw hit the floor when I saw the witty farewell "fun guy" quip, and I was in stitches when I read the song about baking. I look forward to the day I can take the model for a spin - unfortunately I don't have the requisite $18,000 hardware ;)

I have a few questions: Could this be used as a tool to get a feel for public sentiment? For example, could you ask the bot what it thinks about gun control and have it spit out a policy that appeals to the common public? If you ask the bot what it thinks about how a company will perform, how accurately does it predict? I know that the model will contain the biases of the data set, but I'm curious if you've run these types of experiments. What do you think the results would be if you had an even bigger, more diverse corpus? (devil's advocate, for the sake of discussion: perhaps everyone's fb messenger and WhatsApp chat history)

Finally, you have clearly gone to great lengths to make the bot pleasant to interact with. What sort of results to you get when you train such a huge model on an uncurated corpus and don't try to tweak its personality? I find myself wishing that you didn't try to do this as the bot seems to be hyper-agreeable. I. E Too many responses like "You like watching paint dry? That's super interesting! I love watching paint dry!".

Re: Facebook uses 1.5B Reddit posts to create chatbot

#93

Post by DougBTX »

gundmc wrote: post #44

From the cherry-picked conversation section of the research paper [1]. > Human: Hi! > Bot: Hi! How are you? I'm a vegan, so I don't eat any animal products. Even AI vegans reinforcing stereotypes of being overly-forthcoming. [1] - https://arxiv.org/pdf/2004.13637.pdf

AI as stereotype enforcer is the next filter bubble

Re: Facebook uses 1.5B Reddit posts to create chatbot

#97

Post by robmccoll »

pesenti wrote: post #86
mindfulmonkey wrote: post #77

Earlier quoted context omitted.

How do you feel about this probably being most used by bad actors trying to screw with society and elections?

That's definitely a fair concern. We believe that open science and transparency are the right approach here. By releasing it, we ensure that everyone is on the same page with respect to abilities and defense.

> By releasing it, we ensure that everyone is on the same page with respect to abilities and defense.

This seems like wishful thinking. Having knowledge and having the resources to do something with it are two very different things.

Re: Facebook uses 1.5B Reddit posts to create chatbot

#98

Post by renewiltord »

sailfast wrote: post #43
minimaxir wrote: post #22

A reminder that you can obtain the majority of Reddit posts/comments via BigQuery (via Pushshift). No need to write your own scraper. https://console.cloud.google.com/bigquery?p=fh-bigquery&d=re... https://console.cloud.google.com/bigquery?p=fh-bigquery&d=re... It appears to be roughly up to August 2019 for posts, October 2019 for comments.

That's interesting! Did Facebook ask permission to create derivative works (the bot) from Reddit posts, I wonder, or does this fall under web-scraping law? If I recall Reddit users still retain rights to their posts unless Reddit the company provides some sort off broad grants? If they did not, this is an interesting example a company potentially making a great deal of money (if the bot is sold as something) from con…

Making or not making money is such a weird way for people to see things. That's part of why I love the Free Software movement so much and abhor the CC-*-NC licences.

Fortunately, Reddit has the exception where they can give out access to anyone they want. But I still think StackOverflow is the gold standard: CC-BY-SA. No restriction on making money. Maybe a platinum standard would be CC-BY.

Re: Facebook uses 1.5B Reddit posts to create chatbot

#99

Post by stephenroller »

Cakez0r wrote: post #92
pesenti wrote: post #71

Blog post: https://ai.facebook.com/blog/state-of-the-art-open-source-ch... Paper: https://arxiv.org/pdf/2004.13637.pdf Open Source: https://parl.ai/projects/recipes/ Ask us anything, the Facebook team behind it is happy to answer questions here.

I was really blown away by the results you achieved. Amazing work! My jaw hit the floor when I saw the witty farewell "fun guy" quip, and I was in stitches when I read the song about baking. I look forward to the day I can take the model for a spin - unfortunately I don't have the requisite $18,000 hardware ;) I have a few questions: Could this be used as a tool to get a feel for public sentiment? For example, could…

I would not encourage using the model for anything other than AI research -- we're still in the early days of dialogue, and there are a lot of unexplored avenues. There are still nuances around safety, controlling generation, consistency, and knowledge involvement. For instance, the bot cannot remember what you said even a few turns ago, due to limitations in memory size.

In the paper, we did explore what happens when you do NOT fine tune it on the specialized tasks (knowledge, empathy and personality). The non-finetuned bot was both less engaging and more toxic. The special finetuning is really important to getting this bot to be as high quality as it is.

Re: Facebook uses 1.5B Reddit posts to create chatbot

#100

Post by hombre_fatal »

kgin wrote: post #84
moepstar wrote: post #21

What i really wonder is why they trained it on Reddit data, not on their own datasets?

I feel like there's something really interesting behind that choice that probably isn't that flattering to Facebook.

I would imagine Reddit, being a forum of threaded posts, has far, far, far more conversational interactions than Facebook where everything is basically one-shot, no threading. You want to train a convo bot on conversations.

I doubt there's much more to it.

Post reply on HN