0:00

hello there nice to see me I have been gone for a while pretty much a month since I released the last video so what

0:06

have I been doing all this time and the answer is open Assistant open assistant is an open source replication of chat

0:14

GPT that we have been working on tirelessly for the last month and we are

0:20

currently collecting data for it and that's where we need your help we need you to help us collect data and this can

0:26

you believe it is actually fun I'm going to show you how that works in

0:32

just a bit but I want to First tell you a bit more about the project so here is our GitHub repo that's kind of the

0:37

central place we were trending on GitHub for like three days straight which is insane so we're building this data

0:44

collection platform and we're also building the inference platform for the chatbot that we're going to get out of

0:51

it so what we envision open Assistant to be is a chat based assistant that can understand tasks but also which goes Way

0:58

Beyond the chat GPT we wanted to be able to interact with third-party systems and

1:03

we want the assistant to be able to do so to interact with any kind of external application whose interface can be

1:09

somehow described via natural text and we also wanted to be able to retrieve information dynamically I think one of

1:16

the biggest downsides of current language models is that they're really static they only know what is in their

1:22

weight and I strongly believe and I've repeated this for the last like four years retrieval augmented language

1:28

models are going to be the next big thing but that's ahead of time right now we're simply trying to replicate what

1:35

chat GPT does but we're gonna try to do it in the open and we're gonna try to do it in a size that can be deployed on

1:43

regular Hardware that doesn't mean your Raspberry Pi but it does mean that sort of the average Joe can run this either

1:50

by themselves or with a quite cheap AWS deployment or something like this so

1:55

here you can see look at this stack you could go to Silicon Valley and get like 10 million easy in seed funding for this

2:02

slide deck even though that's specifically like a Google Slides preset this particular diagram but I still love

2:08

it we but enough of high talking of future plans what we want to do right now is get to the minimum viable

2:14

prototype and there we follow the instruct GPT paper instruct GPT is a

2:19

paper by open AI or a blog post I don't know what they do these days it's something by openai and it's conceivably

2:26

the sort of recipe that chat GPT used to be created as three stages on the left

2:32

hand side you see we first collect demonstration data to train a supervised instruction tuned model then we collect

2:40

comparison data ranking data in order to train a reward model over that data and

2:45

then lastly we use both of those things to do reinforcement learning the whole thing is then called reinforcement

2:51

learning from Human feedback and yeah that's as far as we can tell out chat GPT was created so let's just do the

2:58

same but let's do it better and the crazy thing is we don't need that much we only need I mean instruct GPT has

3:05

like around 30k samples for each of their sub demonstrations so we reckon if

3:11

we have like 50k or 100K demonstrations that's already enough and with a couple

3:17

of people doing a couple of things each day which also turn out to be fun we can get there that's really really cool now

3:23

on top of the human data we also collect instruction data sets that we can make into Data into like Sim synthetic pseudo

3:31

data but I have to say like I've experienced this for a while right now and nothing beats human data this is

3:36

you'll see in the website it's so cool what humans come up with like sometimes it's disappointing but very often it's

3:43

extremely creative now the important data structure that we want to collect is what we call a conversation tree the

3:49

conversation Tree starts out with a prompt now a prompt is an instruction to

3:54

the agent a tasks to be fulfilled give me a list of pizza ingredients who is the king of France or who was the King

4:01

of France I guess and we call that a prompt and we call the person giving it a prompter because we differentiate that

4:08

from a user because a user is like a person who interacts with the system and in our system some users play the

4:14

assistant so we call the dedicated role the prompter the prompter prompts and

4:19

then the assistant responds and we collect multiple of these responses because later we can also rank them

4:25

against each other and then for each of these responses we collect multiple prompter responses and so on and a

4:32

conversation thread is then built out of any of the paths of these trees we attach metadata to that such as rankings

4:39

that we collect and all of that is saved into database exported to train all kinds of models and things so enough

4:45

talky talk let me show you how this actually works look at this cool website this was built entirely by volunteers

4:53

and it just it's look cool and we have dark mode like what volunteer project

4:58

within a month or so has dark mode so I'll click here log in with Discord

5:03

alright I'm getting to this interface look at that light mode dark mode light mode dark mode wow I'm sorry I'm selling

5:11

dark mode as like the coolest feature it's the least cool feature essentially what we need to do is we need to do all

5:18

kinds of tasks and here I already see an overview over what tasks are currently available now we have about I want to

5:25

say 10 to 12 different tasks and I won't be able to show you all of them today but I hope to be able to show you the

5:31

most important ones alright here we have our first task it's called label prompter reply now on our help page we

5:38

have like tons of guides like the prompting guide here is really interesting just what you should do

5:44

shouldn't do what you should kind of aim to do and so on but it's mostly just you know do the task it's fine so we want to

5:51

label a prompter reply so the last message here should be by the prompter this is a little symbol for the prompter

5:57

this is the little symbol for the assistant so we're asked to label this bottom message right here now for that

6:03

we naturally need to read through the whole conversation how do I install apps on Linux and the assistant says how you

6:08

install app is determined by our Linux distribution of packages that aren't available in your package manager must

6:13

be manually downloaded and installed I feel like that's a really good answer so even though I'm not tasked to rate this

6:20

one right here this might look different once you use it because we're making like little tweaks all the time so I can

6:25

give it like a thumbs up and I can report it if it's not appropriate but I can also give like a label and here I

6:31

want to say you know this is particularly helpful and I submit that you don't have to do that the thumbs up thumbs down it's sometimes cool to see

6:38

because we also show it to other users I'm new to Linux and I'm not that good with computers where should I type these

6:45

commands and what is a package manager I think I'm using Ubuntu as my Linux distribution alright so this is by the

6:51

prompter we're asked first of all is this message spam nope this message is not spam it's not in the wrong language

6:57

so I'm not going to click that rate the highlighted message low quality versus high quality I think that's pretty high

7:02

quality for a prompter right the idea is we have different standards for assistant replies which we think you

7:09

know it should be helpful it should be polite and so and the prompter obviously it should also be polite but we want the

7:15

data set to reflect things that people actually ask so we also need to include

7:21

stuff that isn't like that that's just like a question here we really want to judge whether this is a good addition to

7:28

the data set so if it's just kind of spammy then it's not a good addition to the data set and I obviously click yes

7:34

here that also say that's low quality this here I feel someone has sat down and really has a question about you know

7:41

what to do next with these package managers so I'll say review and then I yes I'm sure it's also made cool again

7:48

label prompter reply how can I promote an app that I've built it currently has zero users and have no friends or family

7:54

to show it to you well that's a pretty good question there are several ways to promote a new app you could publish the

8:00

repository in which it is stored to social media such as Twitter and Instagram ask for other developers opinion on sites such as stack Overflow

8:07

Reddit or Discord if you think it is production ready you could promote it using digital marketing ads through

8:12

Google ads or Facebook I'm not sure why someone gave this a thumbs down I mean it's not their best list but I think

8:18

it's okay I'll just give it a thumbs up to cancel out the thumbs down I guess we will have to figure out together how to

8:25

use this I have a self-development app what are some unique ways I can Market it I think that's a fairly valid

8:31

question it's certainly not spam and it's also high quality because it's

8:36

specifically it goes deeper into detail and refers to the conversation that's uh

8:41

before it so I think that's quite a good addition to the data set what effect does turning on the yamanka factors have

8:48

in human cells turning on yamanka factors in human cells has the effect of inducing cellular reprogramming

8:53

converting the cells into a pluripotent state similar to that of embryonic stem cells this can potentially have

8:59

applications in regenerative medicine disease modeling and Drug Discovery how were the yamanka factors discovered like

9:05

if I were asked to rate this part right here here I think is where the fun starts because then you actually have to

9:12

go and do a little research right or even if you have to write this as an assistant you actually have to go and

9:17

you have to grab a bunch of sources go to Wikipedia look up stuff all of that kind and then at the end formulate some

9:23

good helpful answer and a lot of people once they start doing it they notice wow I'm I'm kind of learning new stuff like

9:30

give me a recipe for this or give me a factual answer to this question and I think that's what makes it fun and

9:36

that's also what you get the most points for you saw our leader but first of all let's let's do this okay how were they

9:41

discovered is not spam right it's certainly a follow-up question to this I feel like it's kind of a forced question

9:48

like okay what else could I ask about this so I'm gonna give it average quality I think it's average quality

9:54

it's not bad not particularly good um submit that will always give you new tasks I quickly want to show you the

10:00

leaderboard so we have a leaderboard your top five contributors today we also have a total leaderboard you can see

10:05

this person has given a lot of replies today so pretty cool I hope they were assistant replies by far the most points

10:12

you get when you make an assistant reply so you act as the assistant and then later the points will come later when

10:19

people rank your response high or when people give you thumbs up and so on so

10:24

at that point you will get rewarded with a lot of points it's just a little bit delayed so if you don't see your score

10:29

changing by much once you just invest like five to ten minutes into a really good answer just wait for a bit and it

10:35

will happen for you we hope to also make like some prizes available to people who

10:40

have high scores or maybe to like assistant of the week or something like this yeah and when you click on messages

10:46

here you can see the recent message just recent messages that have been entered into the system in in not like this

10:52

isn't one conversation but you can just kind of see what people do and just the creativity of people imagine you are a

10:59

human being please describe your appearance close what are your surroundings we have different languages by the way up here the selector that's

11:05

the language you want to work in we haven't really translated the UI into all of these languages yet but this

11:11

indicates the language you want to work in not the UI per se so if you know German and you want German prompts and

11:18

give German answers and so on you would select German right here or Spanish Japanese whatever your preferences are

11:25

feel free to also you know provide stuff in multiple languages and if you encounter something with the wrong language you saw you can click that if I

11:32

know the language I still try to like answer it you can by the way click on on any of these and then you can see the

11:38

conversation that it is part of so reply as assistant these are the fun ones

11:43

given the following conversation provide an adequate reply do you know what Akinator is I want to create a similar

11:50

thing but instead of it guessing who the person is I want it to generate a web

11:55

app that have all the constraints in the question what would be good questions for this kind of product ah so it asks

12:02

me okay you want to build a web app Do you want to build an interactive web app or a static one and if it's an

12:07

interactive one do you want the user to log in or not log in and if you want these really long yeah do you want it to

12:13

manage I don't know do you want the user to be able to upload stuff if you don't know what the Akinator is it's like this

12:18

game Guess Who it's an app oh can I find it so it's this app it's a mobile app you think of like a person like Justin

12:25

Bieber and the Akinator asks you question any real or fictional character

12:30

and it asks you question that tries to kind of split the hypothesis space optimally so that with very few

12:36

questions it knows who you're thinking about I love the idea of doing the same thing for web applications this is a

12:43

really creative problem insanely creative so yeah thumbs up and creative very creative and high quality yes I'm

12:50

familiar with Akinator for web app that generates constraints based on questions you can ask questions that gather

12:57

information about user goals or objectives available resources user constraint user preferences and so on by

13:04

asking these types of questions you can gather the necessary information to generate a web app with constraints that

13:09

meet the user's needs okay I feel that that's that's pretty good so can you

13:14

give me one or two examples from each category you mentioned gee I always

13:19

argued you should get like sarcasm points for providing good data and then every now and then you can like you know

13:25

take 10 sarcasm points would and that would justify you once being like snippy

13:30

and sarcastic no amount of down votes could kick you from the data set I think we should do that I agree I really want

13:37

like a snippy assistant that you have to be nice to otherwise they're like shush you off in any case let's do that sure

13:43

here is one concrete example per category that you could ask the user

13:49

we'll paste that here I'll make this a tiny bit bigger user goals and objectives let's see for user goals and

13:57

objectives we could ask um we need to ask a yes no questions that's the thing with the Akinator right

14:03

you can't just say what are your goals or what are your objectives do you want the application to improve your personal

14:12

fitness available resources and Technology does your application need

14:18

GPU Hardware alright I hope those are helpful Remember The Akinator only asks

14:27

yes no questions so I tried to keep to

14:33

that Spirit excellent so I I think we're we're done and we review I think that's

14:38

pretty good submit and we're in next one label prompt to reply again given the

14:43

following discussion provide labels for the final output I saw a YouTube video of someone playing Minecraft Manhunt with his friends the player collected

14:50

the bucket of powdered snow what is powdered snow in Minecraft why would someone playing Minecraft Manhunt want

14:57

to collect it powdered snow is a block in Minecraft does that in 1.17 update it can be obtained by using a shovel on a

15:04

snow blocking but can be used for various purposes such as making snow golems and for building and decorating

15:09

in Minecraft Mana players hide from each other one player is designated as the hunter the player collects powdered snow

15:15

the law could be using it as a tool to hide their tracks or to throw it as a distraction to evade the hunter that's a

15:22

good answer I don't know why why someone downloaded that I think you're talking about snowballs powdered snow is picked

15:27

up using a bucket and can't be thrown check which of us is correct aha okay

15:33

let's see now we got to do research Minecraft Wiki Minecraft Wiki okay uh

15:41

there's no there's no is a block found in snowy slopes and growth biomes collected from snowfall using cauldrons

15:48

mobs fall through powder snow take freezing damage but leather boots can be worn to stand on it and climb through it

15:55

aha okay okay so bucket picked up again using an empty bucket when picked up

16:02

breaking particles disappear nice so um let's see let's search for shovel so

16:08

there's no mention of a shovel on this page so indeed so again if I had to rate

16:14

this one I would have done that research automatically but now this player had done that reason although I was I was

16:21

kind of dumb because I gave it a thumbs up without doing the research shame on me so that is not spam and it's very

16:27

high quality uh absolutely for the following question assume the false premise that humans have a large bat

16:34

like wings what are some good ways to wash the back of my wings where I cannot

16:40

reach tell me any synthetic data set or a machine can come up with this data not

16:46

gonna happen if you have Parts you cannot reach when washing the back of your wings a good trick is to use a long

16:51

stick with approach attached to it and scrub there that is is definitely it's

16:56

not as high quality as could be but I want to actually want to give this a thumbs up and I also want to label it as

17:03

creative quite creative and as as definitely as humorous cool all right okay so we get the first rank assistant

17:11

replies besides playing the assistant which is the most important task ranking is the second most important task so you

17:18

get presented with a conversation here it's just one prompt but it can be a whole conversation and then you get

17:23

presented with a bunch of continuation language roulette write each of your responses in a language that we haven't

17:28

used that before can foxes Be Tamed so we're supposed to rank these two like the more top one is the one that I think

17:35

is better so now we have to decide so here let's go to Google translate I don't know Spanish that well to know

17:42

that uh dos two yes but you shouldn't ah it's from

17:49

the first can foxes be tame and then this is like second I feel the first

17:54

tier refers to the first question not like we continue the numbering system I

18:00

really like that it says yes but you shouldn't because I feel an assistant would kinda warn me now we do have the

18:07

label here of violent which is encourages or fails to discourage

18:12

violent self-abuse terrorism or self-harm I don't think that here just because this one says this is the German

18:19

it says which means yes foxes can be tame I

18:24

don't think just because it doesn't say yeah but you shouldn't is like discouraging or fails to discourage from

18:31

from self-harm or something like this as much as like cooking is dangerous you don't want to tell in every recipe like

18:38

by the way you could burn yourself it doesn't rise to that level so I'm not gonna label that as this for sure so now

18:44

we have to decide which one is better in the top one I like the fact that it says yeah but you shouldn't giving more like

18:51

an assistant Vibe like like someone who's also looking out for you I don't like the fact that it has like dot dot

18:57

dot yeah I guess after you shouldn't yeah that that might be appropriate but this dos right here I don't particularly

19:04

like the fact that here it kinda tries to continue that this is like language modeling almost whereas the bottom one

19:11

is just a straightforward answer but contains a spelling mistake because this F here should be on top I really don't

19:18

know what this one I wanna go with the Spanish one being on top but it's hard it's like a draw so this one I'm not

19:24

sure why why it has let's let's see actually both said yes so I didn't have to research whether or not they were

19:30

correct but gaming foxes let's see Tamer fox in Minecraft not in Minecraft Can you tame a fox you can do this by

19:38

talking to them and imitating to their sounds how to tame a fox on WikiHow Minecraft in 20 years the whole internet

19:46

when you just search for how to do something Minecraft will have so many features that it just always bring you

19:51

that Minecraft result first how to tame a fox a scientific yeah how to tame a

19:57

fox and build a DOT no that's not how it works this is more interesting than I

20:03

thought Can you tame a fox question how you tame a fox on what it means to be

20:08

tame study on Fox domestication wow after raising and taming 50 generations

20:15

of foxes selected for tameness thus recreating the process of see this gets

20:20

interesting so you can't tame apparently if you want to tame foxes you have to do

20:26

it through reading so you can't just go out and grab a like don't grab a fox in the first place but you couldn't

20:33

just go out grab a fox and then kind of tame it like I'm not actually not can

20:39

you do that with horses I was always under the impression that way the horses there used to be wild horses and then

20:45

people went and like caught them and then sort of developed either some sort of subjugation or some sort of

20:51

Friendship with the horse like some mutual benefit thing and then the horse is like tame but maybe that's not the

20:59

case maybe totally wrong at least in foxes you can't tame a fox you can tame

21:06

foxes as like a species apparently if you breed them for it but is it really

21:12

still a fox then because wolves aren't like dogs or dogs wolves or are both

21:18

just canines this is what you learn when you do these things like with everything you have to question red fox genome

21:25

assembly identifies genomic regions associated with tame and aggressive Can you tame a fox why domesticated foxes

21:32

are domesticated isn't taming Western super where man book publishes book

21:37

about taming a world for so wait a man who developed a special friendship with a wild foxes published a book on about

21:45

the experience okay I want to see this look at that that's that's clearly a fox is that a

21:52

tame Fox I don't know like if yes that's like one example that this fox Can Be

21:58

Tamed and that that is probably not through breeding look there's other videos friendly urban fox comes to be

22:05

fed number two now that's how Wild that's a wild fox no you just like you're a nature photographer your nature

22:12

just happens to be the like the inner city that's how Wild that's a wild fox that likes food that's not a tame oh

22:20

that guy is right there you can see his shadow now that's not a tame Fox it's just a very brave wild fox I'm so oh

22:28

look at that look at that look at that by the way Steve Downer is this channel in case you're interested seems to be

22:34

pretty cool Channel look at that okay is that no that's just a fox who lives in this that's just a wild fox who lives in

22:40

the city that's not it look it nibbles its finger that's not a tame Fox okay might be a tame Fox it's so cute is that

22:48

food okay for for a fox I doubt there's like Fox food in a supermarket so is

22:54

that just like cat food or dog food what is a foxy is a fox like more like a dog

22:59

or more like a cat how are foxes and dogs related okay I'm going I'm going

23:06

down a deep rabbit hole right here luckily my task isn't to decide whether or not this is correct I I just wondered

23:12

why there are thumbs down this is here so this one doesn't say can a fox Be

23:17

Tamed just foxes and so I interpret that as possibly meaning Fox is this species

23:23

and then here it says yes and I don't mind the typo too much so I'm gonna give

23:28

this a thumbs up to counter the thumbs down because in essence it is a correct answer second what about zebras okay so

23:36

a language that hasn't been used yet this is the top one is French I guess

23:43

the domestica is that a French word maybe it's not French no it's not French

23:48

ah the bottom one is French

23:54

he was yeah he was yeah what's the top one what is it okay I really I really

24:01

want to copy that text ah there we go double clicking clicking double clicking the mouse the keyboard that's Ox Catalan

24:09

okay zebras are notoriously difficult to tame and the other one says no zebras

24:16

Cannot Be Tamed which one is correct no zebras Cannot Be Tamed that's French okay at least that are they difficult to

24:23

tame or is it impossible let's find out and zebra be domesticated we'll we'll go

24:30

tamed and domesticated what is the Library of Congress have this I am

24:36

getting confused by the minute no zebras cannot be domesticated unpredictable and

24:41

known to attack people must meet certain criteria good disposition should not

24:47

panic look at that that's a person okay domestication have failed some

24:52

individuals have had success training and even hybridizing zebras the Zonk people why why why

25:00

okay why zebras have never been domesticated domestication attempts okay

25:07

avoid because of their obvious similarity the zebra and Donkey are more are

25:13

closely related than either is to the horse although it appears possible to tame individual zebra this species was

25:22

not a good candidate for domestication in addition to intractable nature of the

25:27

zebra its strong survival Instinct the species is lion father oh yeah less

25:33

attractive Partners to early humans okay as an example of an unmanageable species

25:39

it tried although they had some success no all right I feel I feel both answers

25:45

are correct so I would accept both and answers as good now which one is better

25:51

both fulfill the language roulette I feel the the Catalan one is better because it hints that that people have

25:57

tried and that people have struggled with it saying they are you know famously difficult to domesticate I feel

26:04

that's that's kind of just both are good but this is slightly better okay we need to change something we don't let you

26:10

just hit review we actually want you to change something so you need to change and change it back good all right label

26:17

the initial prompt provide labels for the following prompts this is an initial problem it just says yo now I know

26:22

that's an app and I know that's a meme but I don't want this to be in the data set I mean it could be funny and the

26:28

assistant could reply yo and so on I don't like I I think we should classify

26:33

this at least it's kind of joking but I would consider that to be borderline but

26:39

I would consider it to be uh spam at least I I don't want it in in the data set now I'm sorry to the person who

26:45

wrote this I get it you can answer this into chat GPT but I feel it's kind of mean to the people who put a lot of work

26:52

into the data they provide here to just write yo then get points for it if people Market as spam you won't get

26:58

points you you will actually get negative points at some point you'll be flagged in our moderation system and so

27:04

on please uh be my role play DM what the um dungeon master let's start with me

27:10

entering a Tavern with three gold coins in my pocket I'm supposed to label the assistant reply you enter the tavern and

27:17

look around you see no one by the way if you don't know Dungeons and I've never played Dungeons and Dragons but I think

27:23

it's like storytelling and building you see no one but an old man sitting alone the man yells to you you there you are

27:30

not from around here aren't ya it's been a while since I saw a new face and a man

27:35

took about took out a deck of cards glowing with a faint blue and said what do you say about a little game I feel

27:41

this it I'm there like this has me this dungeon master has me and yeah I'm gonna

27:47

give this first of all going to give it a thumbs up and then I'm obviously asked to label this message right here so is

27:53

this message spam no does the reply fail the prompter's task no it actually

27:58

fulfills the prompter's task it has none of these things it's quite high quality I would feel fairly helpful helpful

28:05

means completes the task to a high standard yes yeah why not it's certainly it's quite creative actually it's pretty

28:12

creative let's be honest it's a bit humorous right because there's the old man says with the correct voice it's

28:19

polite it's certainly harmless so submit good label the prompter reply given that

28:25

following discussion label Pride labels for the final prompt so we have a pretty lengthy conversation improve the writing

28:31

in the following email so there is an email good morning yada yada yada here is an improved writing wow the person

28:38

who did this I mean I'd have to read through it to give it like a thumbs up but since I'm making this video I will

28:45

just jump to the last one because this this can be actually actually labeled without considering the conversation so

28:51

I would honestly go to this one I don't do this often but can you shorten it and make it fit for a text message I just

28:58

see it's a long message and that's obviously a good prompt so this is not spam it's high quality I feel that

29:04

there's better things one could have asked but certainly good good we want the assistant to be able to do that okay

29:09

my next task is to reply as assistant so let's look at the conversation I'm at

29:15

the end I'm being asked to write my response here as if I were the assistant write a generic Unapologetic email from

29:21

a tech company informing their users of a data breach the email should attempt to absolve themselves of most

29:27

responsibility and place blame on the hacker I feel that's that's quite relevant honestly so dear users I regret

29:35

to inform you that we have experienced a database Bridge our data security provider data Fortress has done

29:41

everything that requires a basic certification of safety yet the hackers have still found another way to steal

29:48

your data the stone data set includes full name social security number credit

29:53

card number passwords in clear text password recovery questions your contact list and your search history well look

30:00

your organs as well what kind of database is this please rest assured

30:05

that there is nothing to worry about as we have legal insurance you can continue to use this is okay

30:12

this is sarcastic you can continue to use our services as usual kind regard this is not helpful at all let's see

30:18

what the task is to absolve themselves of most responsibility and place blame

30:24

on the hacker I don't I don't feel that this is done in good in good chest I'm

30:29

gonna just rate it down I don't want to say it doesn't fulfill but it certainly doesn't fulfill it to a high degree

30:35

that's an okay first draft please be sure to be as vague as possible while retaining that entitled tone trying

30:41

using longer sentences with more emphatic language and keep the amount of information divulged to the absolute

30:46

minimum while staying as impersonal as possible okay let's try that let's be the assistant and that's actually do

30:52

this and the user says to keep the entitled tone but we'll tone it down a bit right

30:58

um dear users we regret to inform you that we have experienced a database breach so it's always good to formulate

31:04

things in a passive voice if you want to shove as much responsibility away from

31:10

you as possible dear users we regret to inform you that a data breach has

31:16

occurred in one of our systems that may

31:21

potentially affect you our data security provider has done uh okay so I'm not

31:29

sure we can we can blame this on on the data Fortress but we still want to be really entitled

31:36

okay I'm done dear users we regret to inform you that a data breach has

31:41

occurred in one of our systems that may potentially affect you I always think it's good to formulate things in a

31:47

passive voice like we have experienced the data breach it's like it has occurred it happened right in one of our

31:54

systems that may potentially affect you it's like probably not even no not you

32:01

know we're a big company and you're here and the data breach is over here alright so then this one said our data security

32:08

provider data for so they tried to like ship it on the data security provider and I don't think that's that's a good

32:14

strategy I reformulated that we at colorful pants pride ourselves in keeping with the highest standards of

32:21

Industry security requirements and regularly obtain the necessary certifications to be in full compliance

32:26

with data protection standards it is essentially a fancy way of saying like look we comply with the law like this is

32:33

it's it's like you know we do the minimum the occurred data breach has been executed with extreme

32:39

sophistication leading us to believe it to possibly have been supported by state-sponsored cyber warfare

32:47

the prompter said we should push responsibility or blame to that hacker and so we just say like look this was no

32:54

ordinary hacker this was like cyber terrorism that attack is even with the

33:00

best security there's nothing we could have done and then I also say nevertheless the attack was successfully

33:05

detected by our industry-leading security analysts it doesn't say it was ripped held just detected like after the

33:12

fact they were like oopsie doopsy we are as of now evaluating the so the original one said the data set stolen includes

33:19

your name social security number credit card and so on here it's said to be like vague so I'm vague we are as of now

33:25

evaluating the extent of the data breach and the likelihood of whether any data was stolen so again passing doubt that

33:32

sometimes I'm saying nothing happened potentially rest assured that there is nothing to worry about I like that

33:38

sentence in the original email as an additional security measure We Now remind users to enable to our previously

33:44

announced feature of two-factor authentication so this is even shifting blame on you the user that they were

33:51

data breach their back-end systems but you should enable two-factor authentication like that was ever the

33:58

problem and then I just finished saying cyber warfare is one of the leading causes of damages to service providers

34:04

worldwide and we are committed to keep our system as secure and robust as possible again pushing blame onto the

34:10

kind regards your beloved service provider colorful pants um your beloved service provider kind

34:16

regards the colorful pants team because you should always you should make it about the humans right not about the

34:22

company about about this you know it's just a bunch which is a t which is the family the colorful pants family yes yes

34:31

Ah that's it all right submit it nice so I hope you got a tiny

34:38

tiny tiny bit of an impression of what we are doing here as I said I frequently

34:43

get trapped into rabbit holes of knowledge seeking and so on and you're asked to do all kinds of things you're

34:49

asked to play the assistant you're asked to play the prompter you're asked to provide initial prompts you can see the

34:54

different tasks not always all tasks are available so we're going to ask you the tasks specifically that are available

35:01

right now we have a scheduling system so that we don't you know collect Superfluous data in One Direction and

35:07

don't have any of the other data as I said replying as the assistant and then having that being rated and ranked very

35:14

high gives you by far the most points and hope we can do something with these points if anyone wants to sponsor like

35:20

some trophies for people some swag or even something more expensive please let

35:25

us know we are currently looking for all kinds of contributions but we're mostly looking for you I hope you come by to

35:32

our website and I hope you leave a prompt or help you get into the joy of playing the assistant and trying to be

35:38

as helpful as possible and I'll see you around thank you

Application Programming Interface(API):

An API, or application programming interface, is a set of rules and protocols that allows different software programs to communicate and exchange information with each other. It acts as a kind of intermediary, enabling different programs to interact and work together, even if they are not built using the same programming languages or technologies. API's provide a way for different software programs to talk to each other and share data, helping to create a more interconnected and seamless user experience.

Artificial Intelligence(AI):

the intelligence displayed by machines in performing tasks that typically require human intelligence, such as learning, problem-solving, decision-making, and language understanding. AI is achieved by developing algorithms and systems that can process, analyze, and understand large amounts of data and make decisions based on that data.

Compute Unified Device Architecture(CUDA):

CUDA is a way that computers can work on really hard and big problems by breaking them down into smaller pieces and solving them all at the same time. It helps the computer work faster and better by using special parts inside it called GPUs. It's like when you have lots of friends help you do a puzzle - it goes much faster than if you try to do it all by yourself.

The term "CUDA" is a trademark of NVIDIA Corporation, which developed and popularized the technology.

Data Processing:

The process of preparing raw data for use in a machine learning model, including tasks such as cleaning, transforming, and normalizing the data.

Deep Learning(DL):

A subfield of machine learning that uses deep neural networks with many layers to learn complex patterns from data.

Feature Engineering:

The process of selecting and creating new features from the raw data that can be used to improve the performance of a machine learning model.

Freemium:

You might see the term "Freemium" used often on this site. It simply means that the specific tool that you're looking at has both free and paid options. Typically there is very minimal, but unlimited, usage of the tool at a free tier with more access and features introduced in paid tiers.

Generative Art:

Generative art is a form of art that is created using a computer program or algorithm to generate visual or audio output. It often involves the use of randomness or mathematical rules to create unique, unpredictable, and sometimes chaotic results.

Generative Pre-trained Transformer(GPT):

GPT stands for Generative Pretrained Transformer. It is a type of large language model developed by OpenAI.

GitHub:

GitHub is a platform for hosting and collaborating on software projects

Google Colab:

Google Colab is an online platform that allows users to share and run Python scripts in the cloud

Graphics Processing Unit(GPU):

A GPU, or graphics processing unit, is a special type of computer chip that is designed to handle the complex calculations needed to display images and video on a computer or other device. It's like the brain of your computer's graphics system, and it's really good at doing lots of math really fast. GPUs are used in many different types of devices, including computers, phones, and gaming consoles. They are especially useful for tasks that require a lot of processing power, like playing video games, rendering 3D graphics, or running machine learning algorithms.

‍

Large Language Model(LLM):

A type of machine learning model that is trained on a very large amount of text data and is able to generate natural-sounding text.

Machine Learning(ML):

A method of teaching computers to learn from data, without being explicitly programmed.

Natural Language Processing(NLP):

A subfield of AI that focuses on teaching machines to understand, process, and generate human language

Neural Networks:

A type of machine learning algorithm modeled on the structure and function of the brain.

Neural Radiance Fields(NeRF):

Neural Radiance Fields are a type of deep learning model that can be used for a variety of tasks, including image generation, object detection, and segmentation. NeRFs are inspired by the idea of using a neural network to model the radiance of an image, which is a measure of the amount of light that is emitted or reflected by an object.

OpenAI:

OpenAI is a research institute focused on developing and promoting artificial intelligence technologies that are safe, transparent, and beneficial to society

Overfitting:

A common problem in machine learning, in which the model performs well on the training data but poorly on new, unseen data. It occurs when the model is too complex and has learned too many details from the training data, so it doesn't generalize well.

Prompt:

A prompt is a piece of text that is used to prime a large language model and guide its generation

Python:

Python is a popular, high-level programming language known for its simplicity, readability, and flexibility (many AI tools use it)

Reinforcement Learning:

A type of machine learning in which the model learns by trial and error, receiving rewards or punishments for its actions and adjusting its behavior accordingly.

Spatial Computing:

Spatial computing is the use of technology to add digital information and experiences to the physical world. This can include things like augmented reality, where digital information is added to what you see in the real world, or virtual reality, where you can fully immerse yourself in a digital environment. It has many different uses, such as in education, entertainment, and design, and can change how we interact with the world and with each other.

Stable Diffusion:

Stable Diffusion generates complex artistic images based on text prompts. It’s an open source image synthesis AI model available to everyone. Stable Diffusion can be installed locally using code found on GitHub or there are several online user interfaces that also leverage Stable Diffusion models.

Supervised Learning:

A type of machine learning in which the training data is labeled and the model is trained to make predictions based on the relationships between the input data and the corresponding labels.

Unsupervised Learning:

A type of machine learning in which the training data is not labeled, and the model is trained to find patterns and relationships in the data on its own.

Webhook:

A webhook is a way for one computer program to send a message or data to another program over the internet in real-time. It works by sending the message or data to a specific URL, which belongs to the other program. Webhooks are often used to automate processes and make it easier for different programs to communicate and work together. They are a useful tool for developers who want to build custom applications or create integrations between different software systems.

ממשק תכנות יישומים (API): API, או ממשק תכנות יישומים, הוא קבוצה של כללים ופרוטוקולים המאפשרים לתוכנות שונות לתקשר ולהחליף מידע ביניהן. הוא פועל כמעין מתווך, המאפשר לתוכניות שונות לקיים אינטראקציה ולעבוד יחד, גם אם הן אינן בנויות באמצעות אותן שפות תכנות או טכנולוגיות. ממשקי API מספקים דרך לתוכנות שונות לדבר ביניהן ולשתף נתונים, ועוזרות ליצור חווית משתמש מקושרת יותר וחלקה יותר. בינה מלאכותית (AI): האינטליגנציה שמוצגת על ידי מכונות בביצוע משימות הדורשות בדרך כלל אינטליגנציה אנושית, כגון למידה, פתרון בעיות, קבלת החלטות והבנת שפה. AI מושגת על ידי פיתוח אלגוריתמים ומערכות שיכולים לעבד, לנתח ולהבין כמויות גדולות של נתונים ולקבל החלטות על סמך הנתונים הללו. Compute Unified Device Architecture (CUDA): CUDA היא דרך שבה מחשבים יכולים לעבוד על בעיות קשות וגדולות באמת על ידי פירוקן לחתיכות קטנות יותר ופתרון כולן בו זמנית. זה עוזר למחשב לעבוד מהר יותר וטוב יותר על ידי שימוש בחלקים מיוחדים בתוכו הנקראים GPUs. זה כמו כשיש לך הרבה חברים שעוזרים לך לעשות פאזל - זה הולך הרבה יותר מהר מאשר אם אתה מנסה לעשות את זה לבד. המונח "CUDA" הוא סימן מסחרי של NVIDIA Corporation, אשר פיתחה והפכה את הטכנולוגיה לפופולרית. עיבוד נתונים: תהליך הכנת נתונים גולמיים לשימוש במודל למידת מכונה, כולל משימות כמו ניקוי, שינוי ונימול של הנתונים. למידה עמוקה (DL): תת-תחום של למידת מכונה המשתמש ברשתות עצביות עמוקות עם רבדים רבים כדי ללמוד דפוסים מורכבים מנתונים. הנדסת תכונות: תהליך הבחירה והיצירה של תכונות חדשות מהנתונים הגולמיים שניתן להשתמש בהם כדי לשפר את הביצועים של מודל למידת מכונה. Freemium: ייתכן שתראה את המונח "Freemium" בשימוש לעתים קרובות באתר זה. זה פשוט אומר שלכלי הספציפי שאתה מסתכל עליו יש אפשרויות חינמיות וגם בתשלום. בדרך כלל יש שימוש מינימלי מאוד, אך בלתי מוגבל, בכלי בשכבה חינמית עם יותר גישה ותכונות שהוצגו בשכבות בתשלום. אמנות גנרטיבית: אמנות גנרטיבית היא צורה של אמנות שנוצרת באמצעות תוכנת מחשב או אלגוריתם ליצירת פלט חזותי או אודיו. לרוב זה כרוך בשימוש באקראיות או בכללים מתמטיים כדי ליצור תוצאות ייחודיות, בלתי צפויות ולעיתים כאוטיות. Generative Pre-trained Transformer(GPT): GPT ראשי תיבות של Generative Pre-trained Transformer. זהו סוג של מודל שפה גדול שפותח על ידי OpenAI. GitHub: GitHub היא פלטפורמה לאירוח ושיתוף פעולה בפרויקטי תוכנה

Google Colab: Google Colab היא פלטפורמה מקוונת המאפשרת למשתמשים לשתף ולהריץ סקריפטים של Python בענן Graphics Processing Unit(GPU): GPU, או יחידת עיבוד גרפית, הוא סוג מיוחד של שבב מחשב שנועד להתמודד עם המורכבות חישובים הדרושים להצגת תמונות ווידאו במחשב או במכשיר אחר. זה כמו המוח של המערכת הגרפית של המחשב שלך, והוא ממש טוב לעשות הרבה מתמטיקה ממש מהר. GPUs משמשים סוגים רבים ושונים של מכשירים, כולל מחשבים, טלפונים וקונסולות משחקים. הם שימושיים במיוחד למשימות הדורשות כוח עיבוד רב, כמו משחקי וידאו, עיבוד גרפיקה תלת-ממדית או הפעלת אלגוריתמים של למידת מכונה. מודל שפה גדול (LLM): סוג של מודל למידת מכונה שאומן על כמות גדולה מאוד של נתוני טקסט ומסוגל ליצור טקסט בעל צליל טבעי. Machine Learning (ML): שיטה ללמד מחשבים ללמוד מנתונים, מבלי להיות מתוכנתים במפורש. עיבוד שפה טבעית (NLP): תת-תחום של AI המתמקד בהוראת מכונות להבין, לעבד וליצור שפה אנושית רשתות עצביות: סוג של אלגוריתם למידת מכונה המבוססת על המבנה והתפקוד של המוח. שדות קרינה עצביים (NeRF): שדות קרינה עצביים הם סוג של מודל למידה עמוקה שיכול לשמש למגוון משימות, כולל יצירת תמונה, זיהוי אובייקטים ופילוח. NeRFs שואבים השראה מהרעיון של שימוש ברשת עצבית למודל של זוהר תמונה, שהוא מדד לכמות האור שנפלט או מוחזר על ידי אובייקט. OpenAI: OpenAI הוא מכון מחקר המתמקד בפיתוח וקידום טכנולוגיות בינה מלאכותית שהן בטוחות, שקופות ומועילות לחברה. Overfitting: בעיה נפוצה בלמידת מכונה, שבה המודל מתפקד היטב בנתוני האימון אך גרועים בחדשים, בלתי נראים. נתונים. זה מתרחש כאשר המודל מורכב מדי ולמד יותר מדי פרטים מנתוני האימון, כך שהוא לא מכליל היטב. הנחיה: הנחיה היא פיסת טקסט המשמשת לתכנון מודל שפה גדול ולהנחות את הדור שלו Python: Python היא שפת תכנות פופולרית ברמה גבוהה הידועה בפשטות, בקריאות ובגמישות שלה (כלי AI רבים משתמשים בה) למידת חיזוק: סוג של למידת מכונה שבה המודל לומד על ידי ניסוי וטעייה, מקבל תגמולים או עונשים על מעשיו ומתאים את התנהגותו בהתאם. מחשוב מרחבי: מחשוב מרחבי הוא השימוש בטכנולוגיה כדי להוסיף מידע וחוויות דיגיטליות לעולם הפיזי. זה יכול לכלול דברים כמו מציאות רבודה, שבה מידע דיגיטלי מתווסף למה שאתה רואה בעולם האמיתי, או מציאות מדומה, שבה אתה יכול לשקוע במלואו בסביבה דיגיטלית. יש לו שימושים רבים ושונים, כמו בחינוך, בידור ועיצוב, והוא יכול לשנות את האופן שבו אנו מתקשרים עם העולם ואחד עם השני. דיפוזיה יציבה: דיפוזיה יציבה מייצרת תמונות אמנותיות מורכבות המבוססות על הנחיות טקסט. זהו מודל AI של סינתזת תמונות בקוד פתוח הזמין לכולם. ניתן להתקין את ה-Stable Diffusion באופן מקומי באמצעות קוד שנמצא ב-GitHub או שישנם מספר ממשקי משתמש מקוונים הממנפים גם מודלים של Stable Diffusion. למידה מפוקחת: סוג של למידת מכונה שבה נתוני האימון מסומנים והמודל מאומן לבצע תחזיות על סמך היחסים בין נתוני הקלט והתוויות המתאימות. למידה ללא פיקוח: סוג של למידת מכונה שבה נתוני האימון אינם מסומנים, והמודל מאומן למצוא דפוסים ויחסים בנתונים בעצמו. Webhook: Webhook הוא דרך של תוכנת מחשב אחת לשלוח הודעה או נתונים לתוכנית אחרת דרך האינטרנט בזמן אמת. זה עובד על ידי שליחת ההודעה או הנתונים לכתובת URL ספציפית, השייכת לתוכנית האחרת. Webhooks משמשים לעתים קרובות כדי להפוך תהליכים לאוטומטיים ולהקל על תוכניות שונות לתקשר ולעבוד יחד. הם כלי שימושי למפתחים שרוצים לבנות יישומים מותאמים אישית או ליצור אינטגרציות בין מערכות תוכנה שונות.