Will ChatGPT backlash deter its potential? | Ep 17

Will ChatGPT backlash deter its potential? | Ep 17

Show Video

we're now seeing a backlash on chat  GPT and other AI tools and wow this   came quick and we're going to explore the  topic on the next episode of today in Tech [Music]   hi everybody I'm Keith Shaw host of today in Tech  we are joined today by uh Jason Mars PhD associate   professor of computer science and engineering at  the University of Michigan so go Blue I'm I got   that right right yeah yeah all right Jason is also  the acting co-director of um's clarity lab which   directs Advanced research within AI large-scale  Computing and coding and he's also the author of   the book breaking Bots inventing a new voice in  the AI Revolution uh welcome to the show Jason   yeah pleasure to be here thanks for having me  so uh you know this week there were a bunch of   stories that that I want to sort of bring up the  um uh Story by uh CNBC bring this one up uh Chris   the chat GPT AI hype cycle is now peaking but even  the tech Skeptics don't expect a bust and then we   also saw the New York Times columnist Kevin Roos  uh write a story about this really bizarre uh   conversation that he had with Bing's chatbot and  as everybody knows Microsoft integrated chat GPT   into its Bing search engine last week they had  this big event they were hyping it up uh saying   like this is the future of Chad even though  you got to be on the wait list but you know   some people were able to jump the wait list ahead  of others I haven't gotten my invitation yet did   you get yours have you gotten yours yet Jason yeah  yeah no yeah so I'm I'm on the waiting list yeah   it's crazy they have this um they have this screen  where you can do certain things to get higher in   the waiting list yeah you can you know make it  your default browser and download the app um   so I didn't do all of those things and so I think  I'm still in waiting list yeah the first thing it   asked me was like Hey do you wanna well I had to  sign in under my Microsoft account and then so the   first thing I had to do was actually remember what  my password was for the Microsoft account uh and   then second was hey make Bing your default browser  and I'm like nope done or default search engine   I'm like no no thanks yeah and then it was like  too many steps and I was like yeah then install   the app on your phone I'm like no sorry I'll just  I'm willing to wait in the line yeah yeah no but   you know honestly I gotta tell you I think this is  an incredibly brilliant move by Microsoft I mean   um you know so they recently did that 10  billion dollar deal in incredibly Visionary yeah   um given that uh you know they can be first  to Market truly the First Market with this   kind of large language model at the scale of  Bing but it was super clever because this is a   special moment where they can really maximize the  disruption of Google and so you know making you   really transition and on board into their product  ecosystem yeah as and using the carrot of getting   to use this next gen Tech is is just brilliant  and it's impressive it definitely freaked out   Google right because Google you know they said  well you've got we've got Ai and they take it   out of their back pocket like hey we've got this  too and they call it Bard which again I I really   want to ask someone at Google why they came  up with that name um yeah yeah yeah because   I just think of d and d and it's the in D it's  the one character class you never want to play   um because you just think of someone you know  strumming a loot and and yeah singing a song you   know yeah I mean I spent some time at Google man  the culture is um you know so Bart probably was   was was conjured up by an engineer right and it  when it comes to the marketing yeah efficacy and   all those things like the processes like Engineers  really they have a culture of Engineers really   making a lot of these calls and so yeah yeah and  they could have probably done something better   than Bard so when they yeah so when they launched  that and then you know the first thing like they   they did was they showed an ad of of how it worked  and it got the the the the how the telescope wrong   the James Webb Space Telescope there was a fact  on there that the that their chat AI thing could   could and it was wrong and so you know the stock  market reacted and they lost 100 billion dollars   of value and there was a story this week in RS  Technica about how Google employees are really   mad at the CEO for kind of rushing it and being  very unusual like so it's like yeah I mean they're   definitely nervous yeah no and and for good reason  man I mean the thing is you know conversational AI   has been in consciousness of the market since Siri  right Siri was the first time it was teleported   um into everyone's expectations and and everybody  has been disappointed and under impressed or   unimpressed for a long time Alexa oh you set a  timer okay that doesn't seem very intelligent   um and then you know for the first moment  we had a conversational AI that captivated   um the imaginations of everyone I mean uh you  know it it was preceded by GPT 3 which was the   big technological realization of what was possible  when you trained massive models at the tune of you   know five million dollars plus uh and then chat  TPT was the first realization of the thing that   everyone it blew everyone away right and so the  the wild thing about it though and I really feel   for Google is the key technology that underlies  the achad GPT was actually invented at Google   uh oh okay wow they brought to Market this  you know researchers at Google came up with   this Transformer neural network uh which was the  kind of neural network that makes this technology   possible and they kind of had it in-house and  for many many years they published papers on   it they celebrated it as an AI contribution from  an intellectual realm but they didn't really move   fast enough to cultivate it to the point where  it's able to create such a compelling experience   openai did that with the technology that Google  innovated yeah and so it's uh so and you know   with this amount of um uh widespread uh appetite  creation uh uh and the realization of the appetite   you Google had to move very fast especially when  you hear these billion dollar deals coming from   Microsoft to to get exclusive rights to productize  these Technologies Google knows what this could   mean right for search as a technology do you think  do you think that companies like Microsoft and   Google are the right sort of companies to bring  us to that next level of conversational AI or   yeah do we need more startups and those types  of companies to to Really push the envelope I   mean because these are two huge missteps for  Google and Microsoft we didn't even get into   the whole uh chat conversation that he that they  had with the New York Times guy yet yeah exactly   we've totally got it we've got a chat about but  you know when it comes to the possibility that is   Chad GPT that's been realized is something that is  incredibly expensive to hone and cultivate indeed   a lot of folks call it space flight type projects  right where you invest lots of money to train a   neural network one time you train it on the the  internet basically it's you you crawl the internet   you give this model the internet you train it  on thousands of computers millions of dollars of   compute and so in a lot of ways the creation and  cultivation of this technology is not democratized   it's it's not the case that a small startup or  a a not well-funded outfit okay can actually do   the training to get these Technologies to the  point where they are so we we can only rely on   um uh you know large companies and also incredibly  well-funded companies like open AI was a startup   but Elon Musk started them off with a billion  dollars you know what I mean uh and so they had   a billion dollars to go play I mean the Mandate  of open AI was just do research for the world and   show us what's possible and so they use that  money uh to do this kind of large-scale high   effort model training that gave us a chat GPT and  so uh uh right now it would be somewhat infeasible   for a small startup uh to uh to to try something  different and train it at the scale of of of the   models that okay so so do you feel like then  that these missteps were are just like small   little blips and they'll get their act together  and then kind of push us to the next level   yeah so these missteps I mean it's it's very it's  I mean it's it's not surprising uh to me right   because essentially what these models are doing um  they're not thinking per se but what they do the   way they're trained is by generating a response  word by word and every time they generate a new   word of that response they look at everything  generated before and all of the conversation   that happened before right to pick the most likely  accurate word based on the consciousness of the   internet so it's a writer it's it's writing  and it's it's it's writing word by word with   statistical probabilities determining what word is  happening next uh so it doesn't really know what   it's saying it's just so good at at continuing the  pattern of what should be said next word by word   based on the consciousness of the internet that  it generates that word now there's a number of   things you need to do to get it to to guard rail  it from saying weird things or getting Loops or   get tricked up and and and and that's uh it's it's  almost an art they call Prompt engineering but it   really means you you almost have to guide it with  with some Preamble right and someone just happens   in the background when you ask a question there's  there's some systems that might produce texts to   start it off like starter text uh and then you can  constrain its output so you could really have to   put these tight guard rails and I think what's  happening with the missteps is the model is uh   is a part of the solution that creates that end  user experience and another part is is how do you   guard rail it and it's almost certainly the case  that the chat GPT that you would get from openai   is has different guard rails and and uh you know  prompt engineering Solutions yeah uh that can   curtail how far off to the deep end it can go and  Microsoft moving so quickly I could imagine that   there was a lot of Engineers that had to push  things out very quickly to capitalize on this   moment uh and the the the process of improving  those guard rails for a model like child gbt is   something that we'll see over time but the  speed to Market is really why we're seeing   these misstops it it's not the case that the the  AI necessarily is is is is is is broken or weak   it's just not being steered with this piece of the  system uh that generates the prompts to start it   off right now were you surprised by how quickly  sort of this backlash has has happened uh not at   all really because because it does feel like in  a normal hype cycle that you do stay at hype for   a while before you start getting disillusioned it  felt like it was like one week it was up and then   then this week it's it stinks and and you know  it's amazing how quickly the negatives have ramped   up so but you're not surprised about this how  come no not at all actually I mean it's totally   expected I'm surprised I'm surprised that the good  news was around for so long and that's because you   know it's cool right because I think we got the  most authentic reaction from the media when they   first reacted because it was so mind-blowing they  just have to tell the world right a really true   story about how mind-blowing this is but just  given the nature of what gets clicks like the   bad news the good news could only interest people  so long you know you can't have an article like oh   I'm still using chat tpd and it's still amazing  it's like no no we've got to have a story we've   got to have some drama and so it's a it's a part  of the the the the kind of the the market of of   readers uh would would almost call for well tell  me tell me the scandalous stuff or tell me show   me like like it breaking right uh and then and  that's a good surprise from from the media side   of things it's a good thing that the media is sort  of like it's it's sort of like they're testing   you know they're poking it and seeing what what  pops out and so early stories about well there's   still there's still some bias issues right there's  still some you know it's not perfect and even the   the CEO of openai is backpedaling a little bit I  think he was quoted as saying like yeah it's not a   great great product yet like it's you know so he's  trying to like temper the the enthusiasm exactly   but the media and I think it's a good thing for  the media to keep poking at it because otherwise   like that's the best way to get it to improve  right yeah and it's it's also incredibly important   that we call we we we call upon the public to  be thoughtful I'm not not over you know I'd be   thoughtful about this technology understand that  you can't you can't rely on it as a truth speaker   right because because if you poke at it in certain  ways it'll tell all you nonsense or it'll tell you   untruths it's it's it's important to have the the  media able to communicate that to the masses in   a more effective way than say a CEO just saying  hey guys you know slow down like but we need to   really really know so that the public doesn't uh  misuse the technology right go to the technology   for truth to Art solve all arguments to know  what directions to to take real policies right   um there's a tendency to to to to be too  optimistic about these models and think it   has the answers to our heart problems and  that would be a a mistake well I'll give   you an example of something that I did this  week um I was looking to bake some chicken   in my oven uh and we got the chicken at  the at the meat counter but we didn't I   couldn't remember like what temperature and  how long to cook it was it was panko chicken   um normally so if you look this up in Google like  how to you know what temperature do I need to cook   this panko chicken and you get a list of like 30  000 recipes and I'm like I don't want recipes I   just want the answer I just want and exactly  or the answer that they gave me was just based   on one recipe that they found or the the highest  ranking recipe that they had whereas then I went   to chapter GPT it said oh 425 for 20 to 25 minutes  and then it gave me a list of other things like   make sure you don't overcook it make sure you  don't undercook it all of these things that I   know as a as a human I do know already exactly  but it did sort of give me truth and again I I   guess I believed it um because the chicken came  out fine and I didn't get sick and it wasn't dry   so yeah yeah right and and for and from and for  many many uh uh common things I mean one of the   ways you can kind of use uh a heuristic to know  when when when something like Chad GPT is likely   to succeed is how prevalent is this information  and the consciousness of the internet now the the   power of a chat GPT and the reason it's going to  disrupt search for ever like searches really old   Tech at this moment yeah forward but the reason  is is it in search you're searching for a web page   that might have your answers your question might  span multiple Pages that's incredibly laborious   and painful and you have to integrate you have  to integrate uh uh your understanding from   those disparate pages and search results rankings  right the power of this Tech is that it compresses   that all of that knowledge and integrates it  into a single Consciousness that you can you   can conversationally engage so you get precisely  your information right away and when you should   be more confident in what it's telling you is when  it's information you would expect the internet to   have a lot of knowledgeable right so so like the  do's and don'ts of cooking a particular uh a dish   uh if it's you know uh curry chicken right uh you  could expect that oh in the consciousness of the   internet dpd probably learned really well this  information but if you ask about a really exotic   dish where there might be one page on the entire  internet about then you'd have to be a little bit   concerned because it might use things integrated  from something similar that's actually wrong for   that specific thing and I think that that was  the case in some of the people that were asking   about like where does you know how old is someone  and and the the Consciousness thought it you know   this person was this old old when they were in  fact a different age because that that incorrect   information had been put into sort of the the  data set right right and if there's names if   if on the internet the there's a similar name  uh that's that many different people have it   may integrate information from different sources  right so if you ask something about William Burr   something precise about William Burr and maybe  maybe the consciousness of the internet knows it   about the million William Burr you're likely to  mean yeah but maybe in generating that response   it throws in something about a less known William  Burr that that isn't relevant and so understanding   the the that that's an issue uh is is really  important when you engage these Technologies   for truth yeah I remember I remember when Google  came out everyone almost started Googling Googling   their name to see what information that Google  had on it and with my name I can't there was   like seven different Keith Shaws that do all  sorts of different things including this guy in   England who flies Model helicopter but I haven't  thought about this yet I should probably uh go   on to Chachi PT and see if it can if it knows  me or not yet I don't know I mean yeah yeah you know nothing yeah have  you done that with your name   oh yeah yeah he'll totally done it actually  and I've tried to tried it with a couple names   I was very um I was very honored that uh Chad GPT  actually knows a good yeah it kind of knows things   about me and yeah which was awesome but but um uh  and then some names not so much um but you know   it's it's very uh you know it's very important uh  that you knowing that it could sprinkle things in   um from other uh you know sources is is very  important as one of these Technologies you know   yeah because the use cases are are wild right  like there's a crisis going on right now in   education at all levels in our University yeah  uh at University of Michigan there's special   sessions where um you know the honor Council  wants to understand how to reason about this   technology being a part of our ecosystem now like  students are using it to to to generate essays and   to uh generate reports right and it works very  well with popular topics and um and so as you   use these things you having a human in the loop  that at least edits it I think is a must yeah you   know what I mean what's your what's your opinion  on that because I've now talked to two different   um uh I've talked to a professor who uses  chat GPT in his classes and then I've talked   to another it administrator at another University  and they've got professors that are trying to like   prevent the use of it because they don't want  people cheating like what's what's the opinion   like what's your opinion and then is it does it  match of what you know University of Michigan is   doing yeah no well you know so I tend to I tend  to uh have the opinion that may be less intuitive   like so I I actually believe that we should be  instructing our students to to use chat GPT to   solve problems and then providing the oversight  uh of how to solve those problems I mean we have   this inner this issue before right when when  the internet came to the fore and we had good   search engines there was a time you actually had  to go to libraries to do book reports right like   you'd have to go do research real research to  find answers uh because there wasn't a search   uh and it would be Folly for a professor you  know post having the search era to say okay   please don't use Google I want you to go to  a library check out seven books read those   books and get the answers it's you're trying  to resist the evolution of our times and we   need to we need to as a society understand  how we in a healthy and and forward-looking   way how we integrate these new technologies and  change our pedagogy change how we teach change   the expectations of what the students must do  and what the students can rely on technology   to do because in the real world that's what  they're going to be doing yeah uh preparing   them for that is really the mission so so I I  really believe in an integrative uh approach   um and and adapting ourselves around the new Norms  uh rather than try to depress like suppress the   nor the new Norms yeah to keep our old ways you  know you know and I think you answered this before   but I want to just ask it again like why is there  such a rush to integrate uh the generative AI into   search it didn't feel like initially that like you  know generating content is different from search   like so why did you see them like both Google  and Microsoft quickly because I think Microsoft   also wants to put it into their office apps as  well and you know the whole corporate app area   um you know is something that they want to do  is it just because they will they see the value   of it they want to generate some profits out of  it or yeah is there is there a natural sort of   meshing between the two technologies yeah no well  so the the this is one of those cases where it's   it's a perfect technology because it increases  convenience and it increases productivity the day   that chat GPT made it open uh open AI made it open  I got on right okay yeah and I was playing I just   played with it and then from that day on I have it  up as I'm working as I'm coding because I do a lot   of uh development of software um I I always have  it up and I go there instead of going to stack   Overflow or going to Google to get my answers so  just because I can precisely ask exactly what I   need and get it right away it saves me instead  of spending 15 minutes researching how to solve   a problem I literally spend 15 seconds getting  exactly the answer I need so it's a spike in   productivity and a lot of people had this reaction  when they tinkered with it of course the the   leaders and the executives of these companies know  that the whole purpose of the internet and search   is to help people get access to the information  they need and this is such so good at that and   it's so much fundamentally better for let's say 70  I'm just pulling that number out seventy percent   of the use case of Google it's so much better  at that yeah that it's the biggest opportunity   uh uh in the in the world Google is on like the  number five company in our market economy our   global economy because of search so an opportunity  to capitalize on one of the most important and   profitable Technologies and it's it's a technology  that gives you access to information uh to have   that in front of you means you need to capitalize  on it yesterday right so so everyone's rushing to   to get all of the society to use their product  because of the lock-in effect right if Bing gets   a lot of people to start with set Bing up as their  default because Chad GPT is there those folks are   gonna be locked in even if Google over three years  figures it out yeah Google is going to become the   Bing right because Bing figured it out late nobody  was using Bing everybody was on Google yeah and so   so this is why there's so much effort to get there  you know as soon as possible I'm wondering if if   this means that maybe ask Jeeves will come back  or Alta Vista like you know exactly why they have   to be big like yeah exactly well yeah I know  right that but if they yeah well you see the   thing is ask you if it's a great idea they just  didn't have the technology to do it well enough   yeah you know but yeah uh all right and so you  know that so the next question then becomes like   will we start seeing these things and you  know we start seeing this technology and   things like robots and other physical devices I  mean obviously it's going to be on your phone and   that's a physical device but what I really want to  say I've covered robotics for a few years too uh I   would love to see sort of a mobile humanoid robot  I go into a store and I and I ask it a question   and it tells me exactly where I need or it brings  me to the the place where I want to get the thing   um in the store that you know that's just one  example but you know customer service robots   so far have been horrible it would be great to  actually integrate this in exactly like do you   is it just that we're not there yet and we will  get there or is or is there just not a business   opportunity well so here here's the interesting  thing right and it is a nuanced answer right so   and by the way that use case is is perfect I go  to during like uh you know Black Friday I'll go   to Best Buy I need to get the Nintendo switch  like why do I have to wait for that human to   explain the TV to someone in front of me before I  can just know where if there's any in stock right   yeah yeah like why don't we have that use case  well here's the here's the reality now a lot of   the technology a lot of the products Technologies  where the creators the producers of of society   focuses attention usually comes from the market  appetite right so now we're going to see a   surge in that appetite for these use cases now  that the art of the possible is so manifested   in everyone's face those kinds of questions are  going to be asked not just by us consumers but   by the next Founders by Venture capitalists by the  the consciousness of the public is now uh focused   on the possibilities that this creates so that's  going to create an economic energy to produce new   products and Creations and advance that so we're  going to see a lot of these things get a lot of   energy in the coming uh near-term years now  another important Nuance to understand is the   chat GPT technology itself is is great for being a  question answering system that can integrate a lot   of data into one knowledge yep right so so there's  use cases like how to navigate your Supermarket to   find the item on what shelf yep you know kiosk  that you might talk to like that use case may   not be able to benefit from a chat GPT because  the information it needs is very specific to   that particular store okay and so you might need  to use other Technologies now the good news is   the the same neural network technology that's in  chat GPT is present in many many other places at   different sizes and scales and sizes that you can  wield and train to solve specific problems and so   there's there's there's going to be many many  innovative solutions coming to the fore because   it's no secret how chat GPD works for the for  the technologists and the intellectuals and we   can replicate now uh some of the methodology that  made it so smart in smaller models for specific   use cases and that's you know an area that I'm  actually spending a lot of time on now yeah on   myself do you do you think that then this becomes  something where it gets maybe integrated into an   Amazon Alexa or you know the Google home sort of  voice assistant uh Siri and Cortana as opposed to   search you know because yeah you know instead of  like you mentioned Alexa before you're like hey   set a timer it's like that's what I use that's  what I use it for mainly and to listen to music   um but exactly it's like but imagine having an  actual conversation with you know we actually   get closer to that Jarvis world right from Iron  Man or he's running everything or the the AI   is doing all these different things do you see  that yeah that's probably going to happen before there's pulled up teams working on it right now  like the the one of the key lessons that came   from gbt3 and chat GPT now is that and people  are talking people you know it's a generative   AI right so these are it's a it's a special  approach to how you can build a system it's   that that generate a word by a word a word by  word and produce this answer they call it in   an auto regressive way it means you look at what  you said so far to predict what you say next that   particular technique is not how Alexa's uh all  of these conversational AIS have really worked   um they're usually these dialogue systems with  something they call intense and you extract slots   so you'll kind of figure out well what are the  parameters I need to hit an API with in what you   said it was a different approach that really that  really create these Technologies were created with   and it's why it's part of the reason they kind of  sucked is because everybody was thinking in that   approach and all the products you have pre this  this revolution that's happening right now we're   kind of built in that way but now people have  seen proof points as to what's possible in the   gender generative way yeah to create a dialogue  system which which isn't it's not obvious right   just looking at how the models work that you  could create a dialogue system with just this   generative model so so now that work is already  started everyone's aligning I've been interacting   with a lot of folks everyone is aligning what  they investigate next and around well what can   these generative models do in many different cases  and so the next version of many of these products   are going to use generative model Technologies I'm  confident yeah now we've talked obviously you're   optimistic about the about this technology I can  tell because you're you know you're you're very   excited about all of this stuff but on the on the  flip side we've also seen some sort of dark or not   so you know good purposes for it you hear about  that you know people are using it to generate   malware or write code or or provide a better email  sentence for uh phishing attempts uh because again   with with phishing you could always know oh this  is not real the it's not even correct English and   now and now that's all out of here now I got to  be suspicious of everything that I get and not   just yeah you know my filter now has to be tuned  as well so like what do you feel about some of the   bad uses is it is it again find the bad stuff so  that we can fix it or generate tools to spot the   bad stuff yeah yeah well you know it's always a  tools arm race arms race when you have the you   know there's no various folks out there that  will take every possible technology and Tool   and see how it can make them more effective and  efficient yeah uh at phishing attacks hacking   every you know yeah it's used everywhere but uh  but and also it's just not even beyond that when   it comes to fake news like it's incredibly you  can you can give it a a headline to write a story   about and it'll write you a compelling story with  that headlining with that headline in it oh you'll   buy them just yeah I don't know just make up your  something whoever and then you'll have a command   yeah right so there's all these kinds of things  that we need to now be educated and aware of as   to how nefarious use cases could work and the good  news is actually chat TPT the the motivation for   the technological advancement of chat DPT was to  develop Technologies to help when it comes to the   safety of how these models are are being used I  believe and I haven't valid verified this but I   believe it's almost an accident and the CEO kind  of indicated this it's almost an accident as to   how us trying to develop Technologies to make it  smarter so it could be safe turned it into being   smarter in in many ways and the very fascinating  thing that's going on right now is there's a   lot of energy uh being invested in how do we  design AI models that can can control or detect   other models right you can actually use models you  can train models that can find nefarious use cases   that can detect when an AI wrote this and not a  human right uh and so we've got an interesting   uh you know you can build tools to make the model  safer that are other AIS and that's happening all   the time it's happening right so there'll be  some sort of like signature form or something   that says like oh this was created by an AI right  like that's a good way to to at least start right   exactly exactly or leveraging an AI to say does  was this written by an AI and the AI is like yes   this was I think this was written by an AI with a  98 right um I think you know you can imagine then   you'll have AIS talking to other AIS and oh yeah  yeah yeah that's that's happening now actually   and it used to be kind of a fun thing like it was  almost like a fun thought experiment but five ten   years ago like oh should we have ai's interacting  with each other and hanging out and just learning   from each other but now we're actually uh uh  leveraging uh techniques to to do this right   where AI is like are adversaries to produce better  results uh or you know AIS can kind of check other   ai's work and so forth it's happening man it's  it's wild yeah so it you know is is some of the   backlash that we're experiencing now where you  know you've got you're seeing bad answers you're   seeing biases and answers and you know there was  another there was another story that that came up   this week too where um a bunch of uh tick tockers  uh went to the McDonald's drive-through where they   were experimenting with AI and if you haven't  seen this yet go go on to this article on in   the show notes page um and it's these people that  are trying to just order an ice cream cup and the   AI just keeps going oh would you like ketchup or  I put ketchup in your order or here's some sugar   sugar cubes because they could they couldn't  understand what the person was asking and   you know it's it's hilarious because obviously  it just shows you the failures of the systems   um but on the other hand you know is Backlash  like this going to cause people to go ah AI you   know yeah you're gonna go oh that's just a fad and  and then and then yeah or you know or is it just   sort of a little speed bump in terms of you know  yeah no no so this is the this yeah go ahead well   as I say short attention spans being what they  are these days like you really have a limited   amount of time to sort of you know go through with  something to get to that next level of acceptance   yeah absolutely absolutely no it's it's a very  um so the initial AI has been the there's been   multiple waves of hype Cycles around composition  like I is including Syria and Alexa yep there   are these waves and then there's these troughs  of of uh disappointment yeah right and despair   right like they're like ah it's terrible I  can't use this it's more give me a button   I can push I want to use a touch interface I  don't want to talk to this thing so when you   look at McDonald's I'm actually very familiar  with McDonald's story okay uh I I actually uh   uh was uh kind of in the Runnings at one point  to collaborate with McDonald's to uh produce uh   their conversation land for drive through um but  they did an acquisition and I'm also aware of the   company they acquired to to help uh create that  experience that experience is using the same old   Tech it's using the Alexa set a timer Tech right  which is kind of this rules-based thing that's why   it does these silly things where you asked for  an ice cream why is it asking for ketchup it's   because there's some if statement rule or some  intent thing that will just take you to catch-up   land but but you wouldn't see something like that  yeah like you wouldn't seem something like that   if a generative approach was used with enough data  like if you had enough orders let's say uh and you   saw the interactions uh uh and you were able to  train a generative model uh you wouldn't end up in   a place where it say something so unlikely right  or so out of out of like whack right and so and   so what I would hope is all of the old Tech that  is not good uh and all the experiences that are   terrible like folks will just quickly understand  like from a be knowledgeable enough to know oh   yeah that was that old you know thing um and then  they can recognize when something's good and it's   it's part of the new phase yeah of conversational  AI technology does that mean that like if you're   if you're gonna develop this for a drive-through  scenario that that you're gonna have to record all   of these order calls and and then you sort  of have to tell your customers that you're   recording right like yeah Yeah Yeah well yeah  you know I actually think I don't uh so there's   some I think some of them are recorded uh like  there are these data sets where they've recorded   um uh the the orders and so forth yeah so the data  is there it's the methodology as to how you use   the data dude guys like one thing I want to make  sure a lot of folks are aware the the core Tech   has been there for a while yeah the The Innovation  is on the methodology on how you train this Tech   right um the Transformers is old news but how you  train it and how you systematize it uh is what's   really a lot of Innovations are are driving uh  this forward so that's the open question right   so I think it's still uh there's still some  work and I'm talking about months worth of   research work or in or innovation right lab work  to to take these Technologies and come up with   the right methodology to get that step function  better experience yeah the chat TPT grade kind of   experience yeah an interesting part of the the the  drive through stuff too was probably you know you   know the drive-through even with two humans the  audio quality you get from the speaker to the half   the time I can't understand the the person that's  asking me you know what I want to order uh or you   know or if they're ready for or not for the order  and then you know and then on the other end you're   taking the order you sometimes can't hear what  it is I mean that's been mocked all the time yeah   um and then I'm sorry I'm starting to think that  maybe um in that algorithm that they have they're   like if if confused just add ketchup like yeah  exactly exactly that's confusion that's gonna   mean that's gonna be my new band name is just add  ketchup um yeah yeah yeah yeah just uh it's it's   really hilarious man but yeah have you looked  at the the artwork stuff uh you know like the   generators oh yeah yeah like that's what also  impressed me too um last year I saw some of   the early Dolly stuff and and the images that  it was generating was were completely horrible   um but then at the end of the year I did it  again with dolly two I think it is and the   images are like wow it's actually getting closer  and closer although it still can't generate good   text I don't know if you've tried it lately yeah  yeah yeah yeah yeah yeah and it's a it's what are   you I'm just saying what are your thoughts about  sort of the the generating Visual Arts because   there's copyright issues and you know there's a  lot of public policy issues and it's also there's   also a lot of dangers of like um you know creating  fake things that people think are real um but but   you know the thing is there's this there's this  there's this critical point when when a technology   changes the world it's when it creates an uh an  environment where now there's a clear thing that's   more convenient or it makes people fundamentally  more productive right and so Technologies like   Dolly and there's a lot of these technologies  that are fascinating and they're proof of   of what computers doing things we never thought  was possible um I think they're fascinating but   you know unless there's the kind of like that  that market use case I don't I don't know   how much uh I don't know how how much it it'll  fundamentally matter until it gets to the point   where it now fits into our lives in a way that  makes us better in some way but when it comes to a   proof of what's possible with neural network Tech  uh and large amounts of data and enough money to   use large amounts of data to train massive models  I mean these models spend hundreds thousands of   computers uh in size um you know showing the art  at a possible I think is is it's breathtaking to   see what Dolly can do and we'll we're going to  innovate ways we can we can utilize it to advance   our our life experience but but yeah I mean I I'm  always shocked with Dolly I use it in my class so   in my class at University of Michigan I teach  a conversationally i class I actually have one   lecture all all around diffusion networks that's  that's what really fuels the dolly type Tech yep   uh and Dolly and the the possibilities with  images um it's it's really game changing right   um uh from what we thought was possible yeah  yeah so I'm gonna have you put your prediction   hat on and and say like where where will we be  in about a year like I'm I'm or I mean even six   months from now I mean it does seem like this  is this is being generated very very quickly so   let's just go a year from now uh will we be  having these same conversations or you know   will the conversations change in in terms of wow  there's so much available now and and you're like   where do you think we're going to be in terms of  that high cycle in about a year well so I mean   in a year's time I think we're gonna see uh we're  gonna see real fight like Bing right now the Bing   experience the child experience in bang there's  work to do right so we're gonna see a a polished   and more robust uh this technology everywhere the  way we do search is going to fundamentally change   Forever at that point uh Google is gonna have  their offering the Market's gonna be disrupted   stocks somebody's gonna survive this and their  stock is is going to go up 20 30 uh and so that's   what we're gonna see when it comes to making this  accessible for changing the way we search and find   information yeah and access information  but we're also going to see a flurry of   new kinds of startups that are solving problems  Anew in interesting ways right we're going to see   conversational AI being integrated into more use  cases to ease our lives and then we're going to   see a lot of the uh you know a lot of companies  are going to revamp their self-proclaiming as   an AI leader so yeah you know I give this talk  about the AI Revolution it was kind of cooling off   but this hype cycle is wild dude like because  it gets bigger and bigger with every advancement   and so uh so I think we're gonna see a  whole new ecosystem of Technologies and   pro uh products uh based around new kinds of  conversational AI I can tell you my prediction   for one year is that I'll still be on the  wait list for for Bing yeah yeah yeah yeah you guys and they're not gonna let me I'm  I'm bringing up the rear so to speak so uh   Jason thank you so much for being on the show  today this is a fascinating conversation I   would love to have you back uh we'll just talk AI  forever so uh thank you for for being on the show   sounds sounds good hey a pleasure being here  and looking forward to chatting again soon   all right uh that's all the time we've  got for today's episode don't forget to   like the video subscribe to the channel and  add any comments that you have below join us   every week for new episodes of today in Tech  I'm Keith Shaw thanks for watching foreign

2023-02-23 22:31

Show Video

Other news