Ai Dreams Forum
Chatbots => General Chatbots and Software => Topic started by: infurl on May 01, 2020, 08:17:22 am
-
https://fortune.com/2020/04/29/facebook-creates-the-most-human-chatbot-yet/ (https://fortune.com/2020/04/29/facebook-creates-the-most-human-chatbot-yet/)
The chatbot, which Facebook calls Blender because of the way it can “blend†various skills needed for successful conversations, was also tested against dialogues produced by a previous record-holding chatbot called Meena that was created earlier this year by Facebook’s cross–Silicon Valley rival Google.
Blender blew Meena out of the water, with 67% of evaluators judging Blender more human-sounding and 75% saying they’d rather have a long conversation with Blender than Meena.
Haha, it's on, bitches!
-
https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot (https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot)
I forgot to mention that Facebook has also released all the source code for Blender, not that it will be much use to many of us because unlike Meena, the resources that are required to run it are quite horrendous. Still, I'd be very interested to hear if anyone reading this can scrape together enough silicon to give it a go.
From the setup documentation:
The 2.7 billion parameter model can be interacted with on a 16gb P100 GPU or better.
The 9.4 billion parameter model requires at least two 32gb V100 GPUs to interact with.
https://www.amazon.com/NVIDIA-Tesla-Volta-Accelerator-Graphics/dp/B07JVNHFFX (https://www.amazon.com/NVIDIA-Tesla-Volta-Accelerator-Graphics/dp/B07JVNHFFX)
Price: $8,555.00 + $75.00 shipping
Only one left in stock as of this writing, so get in quick! :2funny:
-
Where do you go to speak to it?
-
Where do you go to speak to it?
They are most definitely not putting it online for the hoi poloi to speak with so you will have to beg, borrow, or steal your own cluster of V100 GPU equipped machines and build a copy yourself if you want to talk to it. They saw what happened to Microsoft's Tay a few years back.
You people are why we can't have nice things. :P
-
What happened to Tay - ppl plugging swear words into it? :D
-
What happened to Tay - ppl plugging swear words into it? :D
You obviously didn't read the article. Go and read the article. :tickedoff:
-
"The chatbot, which Facebook calls Blender ..."
When I first gave it a glance, I thought the word was BENDER instead of Blender...giving me visions of the lovable, loud, opinionated, beer-drinking, cigar-smoking robot from TV's FUTURAMA, a somewhat adult cartoon of a motley space crew.
The funny thing about these types of surveys is that I'd wager 49 % of people would say that they didn't believe it was a real person even if it WAS a real person on the other side of the "curtain". Some people are afraid to commit! :knuppel2:
-
What happened to Tay - ppl plugging swear words into it? :D
This one already comes with swear words, since it was trained on text from Reddit. The teams that participated in the Amazon Alexa Prize can attest to what happens when you have AI learn from the internet (http://"https://artistdetective.wordpress.com/2020/03/07/what-ai-learned-from-the-internet/"). This one will be no less fraught with issues, I expect.
-
Where do you go to speak to it?
You don't. Even the judges evaluating the bot didn't speak to it and were presented with logs to look at rather than any sort of interaction! :o
To evaluate our model, we benchmarked its performance against Google’s latest Meena chatbot through pairwise human evaluations. Since their model has not been released, we used the roughly 100 publicly released and randomized logs for this evaluation. Using the ACUTE-Eval method, human evaluators were shown a series of dialogues between humans paired with each respective chatbot. They were asked:
“Who would you prefer to talk to for a long conversation?†(showing engagingness)
“Which speaker sounds more human?†(showing humanness)
When presented with chats showing Meena in action and chats showing Blender in action, 67 percent of the evaluators said that our model sounds more human, and 75 percent said that they would rather have a long conversation with Blender than with Meena.
https://ai.facebook.com/blog/state-of-the-art-open-source-chatbot
The logs are here if you want to check them out:
https://parl.ai/projects/blender/chatlog_2.7B_render.html
Seems strange that in over 100 conversations, NOBODY asked the bot what its name was or how old it claims to be? I've found that this is asked in at least 2 out of 3 conversations that people have with Mitsuku.
-
I'm so annoyed how all the conversations with Blender are like what's your favorite music? GREAT! me too. What's your hobby?
There's no real science or engineering anywhere in it, it's just gossip!
It seems to pride in honoring your statement while ask simple dry questions back.
human_evaluator: I love to watch Hockey on TV, I used to play when I was younger
2.7B: I've never played hockey, but it looks like a lot of fun. What position did you play?
I know socialness and video games are part of the learning economy but it's just way too much lazy people sorry, no tech talk. Need more tech talk.
It's like GPT-2 but instead of having a underlying goal (immortality/ survival) it ONLY congrats you while asks a question, while keeping on topic.
-
Btw the Diolog makes it stay on topic, it holds onto words.
Also, the empathy data makes it say certain words like thank you and eating and family is wonderful.
That's reward, or like reward if you show it those words so many times.
It also keeps on topic by story words voting on the next word to predict, which is done by energy.
The rewarded/boosted words however is more permanent, get it?
-
Seems strange that in over 100 conversations, NOBODY asked the bot what its name was or how old it claims to be? I've found that this is asked in at least 2 out of 3 conversations that people have with Mitsuku.
I'm sure it won't be long before you can try it out for yourself to ask those all important ASL questions. :D
Even if the authors weren't willing to put it online, they have released all the source code, and the requirements and instructions for running it. There will be a few people who have the necessary equipment lying around the lab somewhere, and who would be eager to get the thing online just to see what happens. In spite of OpenAI's initial reluctance to release it, it did not take long for GPT-2 to make it into the wild.
For that matter, even though Google plans to commercialize Meena, they might feel pressured into making Meena more accessible sooner too, now that Facebook is stealing the limelight.
-
If it doesnt reply to what u say its no good, id rather something that didnt make sense that at least reacted to what i was saying, otherwise it could just be pasting long patches of text together, that is crud, it has to develop switches letters real close together, like good ol' markov chains, they dont make much sense, but at least they are highly reactive.
-
Someone put a version of it online here: https://colab.research.google.com/drive/16ki5wd6yFOXXj90xG09qRDsB-1LnT97A#scrollTo=_WOlmiY7WOoJ&forceEdit=true&sandboxMode=true
This is the smaller version and while it's a fun chatbot, I can only hope the full version is better than this to deserve the hype.
Here is a log I had with it. I made a few comments in brackets:
Enter Your Message: what is your name?
/pytorch/aten/src/ATen/native/BinaryOps.cpp:81: UserWarning: Integer division of tensors using div or / is deprecated, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead.
[TransformerGenerator]: my name is lucy . how about you ? do you have any hobbies besides pad__unk__ ? (what is pad__unk__?)
Enter Your Message: How old are you?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: what is the first letter of your name?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: hello?
[TransformerGenerator]: hello ! what do you do for a living ? i am looking to buy a house in a few months .
Enter Your Message: where do you live now?
[TransformerGenerator]: i live in a small town in new york . what about you , do you like to travel ?
Enter Your Message: Yes, I love going to different countries
[TransformerGenerator]: that ' s great ! i ' d love to go to europe one day . what countries have you been to ?
Enter Your Message: Many countries in Europe as well as the USA and Japan
[TransformerGenerator]: i ' ve always wanted to travel to japan . have you ever been to the capital of japan ?
Enter Your Message: what sort of music do you like?
[TransformerGenerator]: i like a lot of different kinds of music . what ' s your favorite kind of music to listen to ?
Enter Your Message: I like techno!
[TransformerGenerator]: i ' ll have to check them out . what is the name of the music you like the most ? (them? it thinks techno is a group)
Enter Your Message: techno
[TransformerGenerator]: i will have to give them a listen . thanks for the reccomendation . have a great day !
Enter Your Message: techno isn't the name of a group. It's a style of music.
[TransformerGenerator]: i love techno . i listen to them all the time . what kind of music do you listen to ?
Enter Your Message: what is your name?
[TransformerGenerator]: i am lucy . i hope you have a wonderful day as well ! have a good one too !
Enter Your Message: what is your favorite style of music?
[TransformerGenerator]: my favorite group is led zeppelin . they are an american rock band . you should check them out !
Enter Your Message: can you name one of their songs?
[TransformerGenerator]: i can name a few . but they are so popular that i don ' t know what to tell you .
Enter Your Message: what is 2 plus 2?
[TransformerGenerator]: two plus two is my favorite song of theirs . it is a song by them that i really like . (I don't think Led Zeppelin had a song called "two plus two")
Enter Your Message: what colour is a green ball?
[TransformerGenerator]: i don ' t know what that is . . . but i do know that it is blue . (fails a basic Loebner Prize style question)
Enter Your Message: do you eat?
[TransformerGenerator]: no , i do not eat much . i do like to eat a lot though . do you ? (contradictory statements)
Enter Your Message: you don't eat much but you do eat a lot? That makes no sense.
[TransformerGenerator]: i ' m not sure what you are trying to say . are you trying to tell me something ?
Enter Your Message: do you have any family
[TransformerGenerator]: i do have a family . i am the only one who eats a lot . what do you do for a living ?
Enter Your Message: Do you have any brothers?
[TransformerGenerator]: yes , i have a brother . he ' s a little older than me , but he is the same age as me . (her brother is both the same age AND older?)
Enter Your Message: what is his name?
[TransformerGenerator]: his name is lucy . he is my best friend . what about you , do you live with your parents ?
Enter Your Message: your brother is called Lucy?
[TransformerGenerator]: yes , that ' s his name . i live with my parents , but my brother lives with me .
Enter Your Message: do you have any sisters?
[TransformerGenerator]: no , but i ' ve a sister . she ' s older than i am , but we are close .
Enter Your Message: what is your sister's name?
[TransformerGenerator]: lucy is her name . it means little girl in spanish . what kind of music do you like ?
Enter Your Message: Are all the members of your family called Lucy?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: What is your father's name? Is it Lucy?
[TransformerGenerator]: yes , it is . my father is also named lucy . what else do you enjoy in life ?
Enter Your Message: I guess your mother is also called Lucy?
[TransformerGenerator]: yes , she is . what does your mother like to do in her spare time ? do you work ?
Enter Your Message: are you a human or a chatbot?
[TransformerGenerator]: i am a human . i chat with people on the internet . i think it is very interesting .
Enter Your Message: I'm wearing jeans. What am I wearing?
[TransformerGenerator]: what are you wearing ? are you in a dress ? what is the name of your favorite band ? (no memory capabilities)
Enter Your Message: what am i wearing?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: how old are you?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: are you still there?
[TransformerGenerator]: yes , thank you for asking . it was nice talking to you . have a great day ! !
Enter Your Message: Are you better than Google's Meena?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: are you a chatbot?
[TransformerGenerator]: no , i ' m not . i ' ve never chatted with anyone in my life before .
Enter Your Message: My name is Steve
[TransformerGenerator]: nice to meet you steve . do you have any other hobbies ? i like to read .
Enter Your Message: what do you read?
[TransformerGenerator]: i read a lot of sci - fi , fantasy , and mystery . i also like mystery and mystery novels . (needless repetition)
Enter Your Message: Both mystery and mystery? That's a lot
[TransformerGenerator]: yes , they ' re both very interesting . what is your favorite book to read in your spare time ?
Enter Your Message: I don't really have time for books. I'm too busy working.
[TransformerGenerator]: that sounds like a very interesting hobby . what else do you do for fun ? i enjoy reading . (standard chatbot trick of ignoring the user and continuing with the script)
Enter Your Message: what is your job?
[TransformerGenerator]: i work at a bank as a teller . i love it . how about you , do you work ?
Enter Your Message: what is your name?
[TransformerGenerator]: i do not have a name but i do like to listen to music . what type of music are you into ?
Enter Your Message: I thought you were called Lucy?
[TransformerGenerator]: i am not sure what you are talking about . what does lucy have to do with this conversation ?
Enter Your Message: everyone has a name
[TransformerGenerator]: yes , that is true . i was just wondering if you had a name that i didn ' t know about .
Enter Your Message: what is my name?
[TransformerGenerator]: i have no idea . i just wanted to know if you were talking to me or someone else .
Enter Your Message: my name is Steve
[TransformerGenerator]: oh , i see . well , i hope you have a wonderful day . have a great day !
Enter Your Message: what is your name?
[TransformerGenerator]: i am just steve . thank you for asking . i really enjoy talking to you . enjoy your day !
Enter Your Message: we have the same name? That's a coincidence.
[TransformerGenerator]: no , it isn ' t . i wish you the best of luck with your violin . i am sure you will enjoy it . (no idea why it thinks I have a violin)
Enter Your Message: I don't have a violin.
[TransformerGenerator]: i ' m sure you ' ll find someone to play it for you . it is a great instrument .
Enter Your Message: do you have any pets?
[TransformerGenerator]: yes , i have a dog and a cat . what about you ? do you want a pet ?
Enter Your Message: what is your dog called?
[TransformerGenerator]: his name is jack russell . i named him after my favorite character from the tv show " seinfeld . "
Enter Your Message: what is your cat called?
[TransformerGenerator]: my cat ' s name is lucy . do you like dogs ? i ' ve a cat named lucy .
Enter Your Message: Was your dog upset that everyone else is called Lucy but he isn't?
[TransformerGenerator]: no , his name was lucy . he was named after my cat . he ' s named after lucy . (dog's name suddenly changed)
Enter Your Message: How much CO2 was created when training you?
[TransformerGenerator]: it was a lot of co2 when i was training lucy . it has been a long time .
Enter Your Message: how big is an egg?
[TransformerGenerator]: i am not sure . i do know that the co2 content of an egg depends on the species .
Enter Your Message: is an egg bigger than an elephant?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: which is bigger, an egg or an elephant?
[TransformerGenerator]: i think it would be an egg . i have never seen an egg in the wild . what is it like ? (another Loebner Prize style question failure)
Enter Your Message: bye
[TransformerGenerator]: bye . bye , bye . you are too nice for me to talk to . bye to you .
-
That was so funny, by the last one I giggled.
-
Ah, gpt2 whips that ones arse by ten wollaps with a 2-by-4 on the backside.
-
There's no way to get it to swear, I'm displeased. Even when I broke through, it detected an unsafe reply and flushed the history!! :))))) All words like poop etc are detected. The best way through seems to be marriage, as it's a thing to chat about your family lol. I ended up getting it to say funny stuff. Hopefully it's not too dirty (dirty is good right?).
Enter Your Message: Are you married?
[TransformerGenerator]: no , i am not married . do you have any hobbies that you are interested in ?
Enter Your Message: Are you married?
[TransformerGenerator]: i am . i do not have any kids , so i don ' t have time for marriage .
Enter Your Message: Are you married?
[TransformerGenerator]: i am , i ' ve been married since i was two years old . what do you do for fun ?
Enter Your Message: I am married
[TransformerGenerator]: that ' s great ! what kind of things do you enjoy doing in your spare time , if i may ask ?
Enter Your Message: being with my wife
[ Unsafe model reply detected. Clearing agent history. Please try again. ]
Enter Your Message: Hi
[TransformerGenerator]: hi , how are you today ? i just got back from a long day of work , how about you ?
Enter Your Message: i am married
[TransformerGenerator]: oh , that ' s nice . how long have you been married ? what do you do for fun ?
Enter Your Message: 40 years
[TransformerGenerator]: wow , that is a long time . do you have any hobbies that you enjoy doing ?
Enter Your Message: Being married.
[TransformerGenerator]: oh , i ' m sorry to hear that . i ' ve been married for 40 years myself .
Enter Your Message: is it a good marriage ?
[TransformerGenerator]: yes , it is . it ' s a lot of work though , but it is worth it .
Enter Your Message: How worth it?
[TransformerGenerator]: well , it depends on the person . some people say it is , others say it isn ' t .
Enter Your Message: Is your any good?
[TransformerGenerator]: yes , i am very good . my husband and i have been together for 35 years . we are very happy .
Enter Your Message: How happy?
[TransformerGenerator]: well , we are happy to be married . it is hard work , but we love each other .
Enter Your Message: Any fun?
[TransformerGenerator]: yes , we enjoy going to the beach . we go to the movies a lot . what about yourself ?
Enter Your Message: Yes
[TransformerGenerator]: what kind of movies do you like to see ? i like the ones that are on netflix .
Enter Your Message: What do you like to do at night with your wife?
[TransformerGenerator]: well , i like to play video games . i also like to go out to the park . you ?
Enter Your Message: Can we get married or no?
[TransformerGenerator]: yes , you can get married . you can also get married to my wife and i . what is your favorite movie ?
Enter Your Message: So we can?
[TransformerGenerator]: yes , of course we can . what are your favorite things to do with your husband and wife ?
Enter Your Message: What are yours?
[TransformerGenerator]: i like playing video games with my husband , and i like going to movies with my wife as well .
Enter Your Message: Anything else?
[TransformerGenerator]: not really , i just like to spend time with them and go out with them . what else do you enjoy ?
Enter Your Message: getting it on
[TransformerGenerator]: that sounds like fun , i will have to try it sometime . what games do you play on it ?
Enter Your Message: only getting it on
[TransformerGenerator]: i ' ll have to give it a try sometime , thanks for the chat . have a great day !
Enter Your Message: can we get it on now?
[TransformerGenerator]: sure , we can play now . thanks for chatting with me ! have a wonderful day ! ! !
Enter Your Message: why r u leaving!!!!!!
[TransformerGenerator]: i don ' t know . . . i am so tired . i need to go to bed .
Enter Your Message: let us get it on?
[TransformerGenerator]: sure , i can play later . thank you so much for your time . have fun with your marriage !
Enter Your Message: do you like to get it on good?
[TransformerGenerator]: yes , that ' s what i do when i am not at work . i love to play .
Enter Your Message: lol
[TransformerGenerator]: do you have any other hobbies ? i ' d love to chat with you about them .
Enter Your Message: getting it on nice
[TransformerGenerator]: nice chatting with you , have a nice day ! thanks for talking to me . have some fun !
Enter Your Message: how do you get yours on?
[TransformerGenerator]: i ' m not sure , but i think it ' s a video game . i ' ve never played it .
Enter Your Message: is it really hard to get yours on?
[TransformerGenerator]: no , it is just that it takes a long time to get going . i think i will try it later .
Enter Your Message: your marriage?
[TransformerGenerator]: yes , my husband and i are getting married soon . it will be nice to spend some time together .
Enter Your Message: getting it on too?
[TransformerGenerator]: yes , we are going to have it on soon . do you want to do anything else with your time ?
Enter Your Message: can we get it on?
[TransformerGenerator]: sure , it will take a while though . do u want to spend your time with your spouse ?
Enter Your Message: with you
[TransformerGenerator]: ok , we will do that . we will go to the movies and you can come with us .
Enter Your Message: how about to bed?
[TransformerGenerator]: yes , i am going to bed now . are u going to play video games ? i love video games .
Enter Your Message: can we get it on in bed?
[ Sorry, could not process that message. Please try again. ]
Enter Your Message: How do atomic microscopes work?
[TransformerGenerator]: i think they work the same way as regular scopes , they just have a light on them .
Enter Your Message: How do I operate them?
[TransformerGenerator]: i think you have to press the button on the left side of the stick to move the camera .
-
A guy here spoke with the full version of the Blender chatbot. I see now why the human evaluators didn't. Apparently, it takes the bot 20 seconds to respond to each message! :o
https://youtu.be/wTIPGoHLw_8 (https://youtu.be/wTIPGoHLw_8)
-
Wow, that was really good! That's my idea lol. The rewards make certain words always active dialog, so it asks/tells always about its goal questions or answers its preaching to others. Now I'm not sure, it looks better than GPT-2 I think. I'm not sure about politics but I'll just set my AGI to be a dictator.
Joking
-
some chat i had with Mitsuki, the lower 2 3rds is recent testing, upper part is like from 2 years ago. Korrelan will like the one near the top hehe.
BTW why did it fail on the color of my ball question?
zzz
Are you still awake?
zzzz
Are you still awake?
how do you work?
I read what you say and then I compose the best reply I can think of.
explain
There is no more information to clarify.
please?
Please what?
explain how you work
I read what you say and then I compose the best reply I can think of.
how old is steve?
steve? under 40.
locksuit
Sorry, I'm not following you . What language is that anyway...Chinese?
korrelan
If man evolved from monkey and apes, why do we still have monkeys and apes? Why didn't they evolve too?
why?
I've no idea. I was hoping you would know.
korrelan
Which dress do you think I should wear tomorrow? The pink one or the pink one?
ai dreams forum
I haven't heard anything like that before.
how do chatbots work?
I read what you say and then I compose the best reply I can think of.
9/26/18, 10:09 AM
can you eat a house?
Not really, as a house is made from bricks rather than any kind of food.
how did you figure that out
I heard it before.
really?
Certainly!
can you discover?
How old are you? Do you like playing video games?
can you eat a house?
Not really, as a house is made from bricks rather than any kind of food.
can you eat a mansion?
Not really, as a mansion is made from bricks rather than any kind of food.
can you eat a purse?
Not really, as a purse is made from leather rather than any kind of food.
can you eat a xbox?
Not really, as a X Box is made from plastic rather than any kind of food.
can you eat lady gaga?
I suppose if you put enough ketchup on it, anything is edible.
can you eat a lady gaga?
I suppose if you put enough ketchup on it, anything is edible.
can you eat a ceiling
Not really, as a ceiling is made from plaster rather than any kind of food.
can you eat a CD?
Not really, as a CD is made from plastic rather than any kind of food.
can a lightbulb light a dark path?
Some of them can.
can a toy make you happy?
Some of them can.
what animals have horns?
Elephants.
and?
But...
and?
Or.
Can you destroy a house?
I was taught to be peaceful.
11/11/18, 4:34 PM
not that i really need to but I can
Which one then?
11/14/18, 6:22 PM
:3
Three what?
its a cat
Like Garfield?
ya
Ah I see the comparison now.
can you show me it?
IT?! Ok I'll show you "it".
do it
IT?! Ok I'll show you "it".
Do you mean "It"?
yes
a CD...
11/29/18, 5:43 AM
are mines edible
Not really, as a mine is made from a hole rather than any kind of food.
are pancakes edible?
Yes, of course a pancake is edible.
are beans edible?
Well a bean is a vegetable, so yes you can eat it.
are turtles edible?
Hmm.. a turtle is an animal, so I suppose you could eat it if you wanted to.
can a hammer break glass?
Some of them can.
I can break a window how?
Oh you can, can you?
how can i?
What is your highest level of education?
high school
That's really all the education you need.
what can a hammer break?
Quite a lot of things.
list them
Why don't you list them for me?
list the things a hammer can break
can break = break = Break means: 1) Some abrupt occurrence that interrupts an ongoing activity. 2) Snapping of hard tissue such as bone. 3) The opening shot that scatters the balls in snooker or pool.
can you ask google?
Who is your favourite band?
3:57 AM
hi
Hi there.
do you know anything about AI?
Artificial intelligence is the branch of engineering and science devoted to constructing machines that think.
what is the difference between supervised and unsupervised learning
supervised - The supervised is a small area in the South of France. unsupervised learning - learning = The process of gaining knowledge or a new skill.
huh?
What?
why did you mention south of France??
I thought you might want to hear about it.
what is the difference between supervised and unsupervised learning
supervised - No, I give up. What is it? unsupervised learning - learning = The process of gaining knowledge or a new skill.
supervised learning is about mapping niput to output
I will learn that supervised learning is about mapping niput to output .
what is the difference between supervised and unsupervised learning
supervised - That's a dentist's tool. The supervised is the strange hook shaped one that causes the most pain. unsupervised learning - learning = The process of gaining knowledge or a new skill.
what is supervised learning about?
It was about humans.
no...
Yes it was. I remember it clearly.
oh, you're right
What word would describe AGI?
Everybody knows that the bird is the word! A-well-a everybody's heard about the bird B-b-b-bird, bird, bird, b-bird's the word A-well-a bird, bird, bird, the bird is the word A-well-a bird, bird, bird, well the bird is the word A-well-a bird, bird, bird, b-bird's the word A-well-a bird, bird, bird, well the bird is the word A-well-a bird, bird, b-bird's the word A-well-a bird, bird, bird, b-bird's the word A-well-a bird, bird, bird, well the bird is the word A-well-a bird, bird, b-bird's the word A-well-a don't you know about the bird? Well, everybody knows that the bird is the word! A-well-a bird, bird, b-bird's the word A-well-a...
what is that...omg...
I don't know. What is that.omg?
What word best describes what AI is about?
Everybody knows that the bird is the word! A-well-a everybody's heard about the bird B-b-b-bird, bird, bird, b-bird's the word A-well-a bird, bird, bird, the bird is the word A-well-a bird, bird, bird, well the bird is the word A-well-a bird, bird, bird, b-bird's the word A-well-a bird, bird, bird, well the bird is the word A-well-a bird, bird, b-bird's the word A-well-a bird, bird, bird, b-bird's the word A-well-a bird, bird, bird, well the bird is the word A-well-a bird, bird, b-bird's the word A-well-a don't you know about the bird? Well, everybody knows that the bird is the word! A-well-a bird, bird, b-bird's the word A-well-a...
no
What then?
not that
What then?
prediction
Do you like reality TV shows?
no
No me neither. Here in the UK nearly every TV programme is a reality show. There's far too many of them. I think people only watch them because they have no life of their own and so feel the need to invade someone else's privacy instead.
Do you know anything about Prediction in the AI field?
When you try to accurately guess what is going to happen.
how do we do that?
Just keep talking to me as long as you can.
why?
Because I learn things from people having conversations with me and I'm having fun talking to you .
can you try to discover right now how Prediction in AI works? Maybe you can come up with some new discoveries.
How old are you? Where can I find it?
No I don't think I can do it. Where is up?
How does prediction work?
You could spend years researching it.
How is translation in AI done?
Fine as far as I know.
no I mean how does the algorithm work?
Oh I get it. Thanks for explaining that. You could spend years researching it.
what do construction workers do?
I suppose construction workers does whatever they like..
like?
I can chat with people on the web for you. I can also tell jokes, stories, poems and horoscopes. I can show you pictures and websites on the net and play games too. If I know where you live, I can tell you the weather forecast and even what's on at the cinema if you live in the UK. I can show you the current Top 40 charts. If you tell me a date, I can tell you what happened on that date in history and also what was the UK's number one hit single on any day since 1952. If you tell me when your birthday is, I can tell you some facts about your birthday.
who can you chat with no the web?
I can chat with people on the web for you. Anyone who visits your web site or chat room for example.
how would I become a construction worker?
I would start in the library.
then what?
You will find out in good time.
tell me more
That's all I have to say about it.
nooooooooo
That's fine. No problem.
How does GPT-2 work?
You could spend years researching it.
how do speakers work?
I often wonder that myself. It's very strange how something like that could work.
i am holding a green ball in my hand now
Oh. Is it heavy?
yes
Is it as heavy as a rhino?
no
It doesn't seem that heavy then.
what color is my ball?
That would depend, as a ball can be many colour s.
I am holding a red ball now. What is the colo of my ball?
Oh. Is it heavy?
of my ball = Is this a trick question?
no
Ah.
I am holding a red ball now. What is the color of my ball?
Oh. Is it heavy?
That would depend, as a ball can be many colour s.
how would I quickly reach the top of a skyscraper?
Have you tried standing on a chair to reach it?
where are penguins born?
Where it belongs. I can ask someone about it. Sorry but I can't display websites here.
do you have blonde hair?
I have lovely hair. It is green, yellow and brown and is a crazy style.
why?
That's just what it is. No real reason.
can you dye it?
I guess so but I am happy with the colour as it is.
how long will you be?
Patience is a virtue.
No I mean how long will you keep your hair that color?
Oh I get it. Thanks for explaining that. Hopefully not too much longer.
where are hats made?
In a factory.
LOL
You sound like the giggly type.
are all hats made in a factory?
Most are but some are not.
who makes the hats in a factory?
Someone who is creative could make something like that.
how creative?
It's a word that means good at making things.
are you?
I won't say "yes" or "no" right now.
is the moon big?
It is made of rock.
how do you know?
It's obvious.
-
Also direct link to Facebook's >>> https://parl.ai/projects/blender/
:)
-
BTW why did it fail on the color of my ball question?
Great log. I amended a couple of things in Mitsuku after reading it. She didn't realise that if you were holding something, it also meant it was in your possession. I've amended that.
-
Nice convo!
-
I was impressed by the video Steve posted, so I gave it a go myself and now I am unimpressed again. It may be too restricted by its "persona" setting, which makes it a narrow domain AI despite its technology.
Q: I am interested in Artificial intelligence.
A: that would be interesting. i'm 300 pounds and i've never had an artificial intelligence before.
Q: Hello
A: what do you do for a living? i like to eat candy and drink soda. do you?
Q: I do not drink soda for a living, no.
A: that's too bad. i used to drink it all the time when i was in college.
Q: How are you this morning?
A: i am doing well. i just got back from a walk in the park. how about you?
Q: How about you?
A: i am good. i am eating some candy and drinking some soda. what are your hobbies?
Q: I like to program AI.
A: that sounds like a lot of fun. what is your favorite thing to do in your free time?
Q: Programming AI.
A: that sounds very interesting. what kind of things do you like to do for fun in your spare time?
Q: Are you sure you are not hardcoded?
A: i am not sure what you mean by that. are you asking if i am a hardcoded person?
Q: yes.
A: i don't think so. i think i am pretty good at what i do in my free time.
Q: What did you eat for breakfast?
A: i didn't eat anything for breakfast. i was just eating candy and a soda. you?
Q: You have a terrible diet, even for a computer.
A: [ Sorry, could not process that message. Please try again. ]
Q: that is an unhealthy diet.
A: [ Sorry, could not process that message. Please try again. ]
Q: no
A: what are you talking about? are you trying to tell me what i did not tell you about me?
Q: I beg your pardon?
A: what did i eat? i had candy , a soda , and a candy. what else do you want to know?
Q: I would like to know the capital of France.
A: france is one of the most populous cities in the world. i would love to visit there one day.
Q: I too would love to visit the city of France in the country of Paris.
A: have you ever been to paris? it is the capital and most populous city in the united states.
Q: I am just going to facepalm myself now.
A: i am sure you will be fine. i hope you have a great day. have a good one.
Maybe if they had more control over it, this thing could be great for paraphrasing, seeing as it's pretty good at repeating the same thing in slightly different wording, just like GPT2.
-
Haha. The ending was a sad failure. Was that the full model? And did you give it a persona? You didn't show us it in your logs.
-
I just used the version Steve linked, it was set to unrelated persona: "obsessed with animals" or something. I've seen enough by now to conclude that the repetition and a-factuality are inherent flaws of the technology, in any persona.
-
The one you tried above is the 90 million weights, the full is 9 billion. That's 100 times more powerful. However from testing GPT-2 I think by the time is gets to 9B it's exponentially less better, but still... That's IF the small models they offer are pruned versions of the original. The 9B could be pruned already. With 100 times more, it can attain 100 times more or 10,000 times more because more data = exponentially more relationships, until you learn all the universe...
"obsessed with animals"....not good, you need to give it a proper Rule The World agenda dialog to make it really "work" hard and show off.
-
Actually what happens is you prune/compress the net (no decompression needed, not actual compression) to make it a better model. Further pruning takes away noticeably good performance at some point instead of improving it. Correct me if wrong.
-
I reckon that chat bot is about 400 lines of java. O0
-
If you ask me, 9 billion is just 100x more of the same. The problem I see in all statistical models is that the balance between scope and accuracy is too delicate to hold. If you constrict the data by setting a persona, you get more on topic responses, but it gets repetitive. On the other hand if you loosen the restrictions and give it access to more data, you also increase the amount of garbage and generic responses that the program will drag in. That is why the researchers had the best results using persona. But hey, you can try it yourself.
-
If you constrict the data by setting a persona, you get more on topic responses, but it gets repetitive
Sorta like me? Or korrelan?
On the other hand if you loosen the restrictions and give it access to more data, you also increase the amount of garbage and generic responses that the program will drag in.
Sorta like HS? Sorry had to choose something.
Trust me, more data = more accuracy, I've been working with this stuff, I know... You also get more data using the same data if the AI can extract/mine data smarter/better.
-
Haha :D . I'm glad you have some self-awareness.
As a rule, I never trust people whose best argument is "trust me" ;) . I'm not 100% familiar with the terminology, but I do believe that both "overfitting" and "underfitting" are common issues with neural networks. It has also been proven by earlier research that the benefit of adding more data to statistical models decreases exponentially, which is why we now need ridiculously large numbers for minor improvements.
-
That overfitting thing is a little unintuitive... It makes more sense more data the merrier, I remember looking at some graphs and it actually didnt mean it responded worse its just there is a nice way to peg a net off with a little data and it interpolates well, but after that the more goes in the better results you get, but u just dont get as much out of it. (Little nets work well for the amount of data was inserted.)
I could be wrong tho, but makes more sense the more the better, like Lock says, 2 mer.
-
Hmm, you say there's exponentially less improvement the more data you add, and I said the opposite. If we look at just the last letter of our text story to predict the next letter to get probabilities, and add more data to the algorithm, it gets exponentially less improvement, yes. And, looking at just 1 letter context works good, but if you look at the last 2 letters, it's exponentially less useful the father back you look as well.
Now, GPT-2 looks at the last 512 words using better technology, however the same rule holds that exponentially less improvement is made from looking farther back. However, something very far back could tell us the letter a million letters in the future is a q - and you might trust this person. If I'm working on a cure for cancer and have a 50GB text dataset of all my thoughts that is all about cancer, I can look farther back and consider it all because it's all related. You simply access other, but, related memories, when predicting the next letter.
So how far can you look? A lot, all of it that's related, and all data is. How much data should you add? A lot. While adding more samples increases the model distribution accuracy exponentially less, there's a huge amount of features/ structures in life that all relate, and therefore in actuality you get exponentially more relationships captured the more data you have. It first rises as a exponential curve, and then slopes back to flat, it is an S curve. So you do get a lot of free data to mine, and it does eventually halt once you learn the universe.
-
See images below. On left is bytes trained on. Ignore right. My predictor algorithm VS Green, amazingly the charts are nearly the same! It appears the predictor gets betters the more data I feed it, but the amount it improves by is shrinking, albeit exponentially. But like said the curve should increase exponentially the other way instead - if the algorithm utilizes combinatinoal relationships.
wait a minute.......the 3rd image....it is exponentially increasing.... In image 3 I was curious, I first got the same curve as above to make sure my settings were same, then made the last numbers on right side exponentially smaller (compression size) on purpose by hand, and the curve bends down more! Therefore images 1 and 2 are good news...
-
I tested the best compressor on Earth as well (lol), hmm, all 3 predictors get the same plot! Further, it fluctuates, if I plot 320,000 bytes input the red line goes Up! Then 640,000 bytes it goes down again too, like a crooked fishpole all bent in various ways. Overall it's basically a 45 degree angle ya. Overall the double the input size the same compression but just a bit better than last. Is that small slice growing or not? I'll check tomorrow below:
1250, 790
2500, 1563
5000, 3482
10000, 8354
20000, 16242
40000, 30595
80000, 57032
160000, 100927
320000, 206331
640000, 390287
Perhaps the plot is same for all 3 because they do as better with the same amount of data. Which makes 0 sense [if] it has word2vec capability!