ADS

ABOUT THIS EPISODE

Giphy is a search engine for gifs, the short animated graphics that we see around the Internet. Giphy is also a creative platform where people create new gifs. Every search engine requires the construction of a search index, which is a data structure that responds to search queries efficiently. Since Giphy is a search engine for graphics, there is almost no text inherently associated with the each document. Giphy uses

Continue reading...

English
United States

TRANSCRIPT

00:00:00giphy is a search engine for gifts the short animated graphics that we see around the internet and in our messaging applications if he is also a creative platform where people create new gift every search engine requires the construction of a search index which is a data structure that responds to search queries efficiently since Jiffy is a search engine for graphics there's almost no text inherently associated with each document giphy uses a pipeline of different labeling techniques in order to make a gif indexable by the search engine in my conversation with giphy is CTO Anthony Johnson we discuss how to scale a search engine we discuss why GIF he needs to build a new techniques for image processing how human labeling for machine learning is evolving and the future of giphy both as a creative medium and as an advertising platform this was an exciting and wide-reaching interview and I think you'll enjoy it
00:01:05couchbase is a nosql database that powers digital businesses developers around the world choose couchbase sport advantages and data model flexibility elastic scalability performance and 24 x 365 availability to build Enterprise Webb mobile and iot applications the couch base platform includes couchbase couchbase-lite which is the first mobile no sequel database and couchbase sync Gateway couchbase is designed for Global deployments with configurable cross Datacenter replication to increase data locality and availability running catch Basin containers on Docker kubernetes mazos or red hat openshift is easy and it developer it. Couch base.com you can find tutorials on how to build out your couch base deployment all couchbase products are open source projects customers include industry leaders like AOL Amadeus AT&T Cisco Comcast
00:02:05Disney Dixon's eBay General Electric Marriott Neiman Marcus PayPal Ryanair Rakuten / Vivir Tesco Verizon Wells Fargo as well as hundreds of other household names thanks to couchbase for being a new sponsor software engineering daily we really appreciate it
00:02:31Anthony Johnson is the CTO of giphy Anthony welcome to software engineering daily are the short animated graphics that we see around the internet and I want to start with a discussion of search and then we will work our way towards the complexities of searching gifts specifically in order to build a search engine for anything whether you're talking about searching for books or products or whatever you typically need to build a search index could you explain word search indexes yeah it's a great point so in most traditional information retrieval system you're going to be building up in the next game instead of running through every single possible document record and looking at the woods in that you're actually going to reprocess the information that
00:03:31index is going to be a map of where those documents also if I'm looking at it and I'll case of gift and the gift happens to be typed with the watch cat instead of having to scan through hundreds of millions of gifts to find each one that says what cat we're going to have a tree basically and we're going to go down that tree looking for the wood cat and once we find the word cat is going to have an index of where each one of those is and I'll pin a lot of record set so let's say there's a million gets to have the Wildcat then once I go down the street I'm going to have a list of those specific Midian cock gifs of creating against this search and exit at least I learned about information retrieval class was give a query and then the query gets turned into a vector and then you try to
00:04:31watch that Vector using some kind of like similarity measurement of cosine similarity against the documents in the search index and then you can use that similarity ranking 2 to order the search results is that still the way that search indexes are queried against cosine similarity with that kind of metric is actually is used quite a bit so the historically you you didn't have the luxury of being able to calculate cosine similarity over other such things on on the on the distance so you actually had to use the indexes to 2322 such
00:05:21full documents now when you looking at in sodas in a traditional algorithm for the stuff I things like for rent King on TF IDF and bm25 and let's say that I said I go down and then give me a list of stuff now let's say that I've got a couple of these was going to combine them I'm going to look for the ones that have the most significant other would the cat the reality is I don't really know the truth is I would like the shouldn't have as much weight in my search because let's face it is going to stop. Without saying these words are just let's just throw them out right
00:06:21so that's that's what I'm saying this word is completely useless I never got to look at it something like Obama and Obama and we have cats only only $200,000 tight while tied with the word Obama much rarer means that an indication for the tiger bomber should actually be considered much higher in that such bad cats allowed to the TF IDF is 10 frequency in Vista
00:07:21a flat system information. There have been some previous episodes about search that listeners can check out gifts more specifically one interesting thing about building a gif search indexes that these are essentially unlabeled documents that are hard to label are there does not straightforward way to label them because they're just in there they're video file their graphics and theoretical you would need to be able to associate some text labeling with them because if people are entering text queries then you need to be able to search against a labeling system that is that is Kino commensurate with those queries how do you get the gifts labeled so they can build a search index
00:08:21one of the main things we have is we have this Pizza Pipeline evasively of how a gift gets processed and it's a combination of people inside the company outside the company tagging sudden content so sometimes will have we have Partners who have professional contacted that job this time to bring in high quality content and when they bring in the high quality content let's say it's a TV show whatever else part of what they're going to do is provide that because it's very important that has the people can find the right content now that allows us to do things like TV series episode 2 season 3 well I have it will have the show me time to such as ready to have the people who care about that
00:09:21another cases it's it's more important matter so we have human and computer vision pipelines that images the back end and we saw some small fraction of that to the end users but in all instructional we actually support the ability to have many of these and levers them in different ways behind the scenes for various types of subjects it's fascinating how much this human in the loop kind of prop pipeline processing has begun to propagate to what we do is computer scientist I've done a bunch of shows recently about it recently added a show with unbabel which is the translation company and you know doing translation with a combination of machine techniques and humans in the loop get you really robust a transition seems like this is going to be a model for how we reduce a
00:10:21Acuity in complex query and complex data sets a complex labeling for a long time to come in I think you know if this is going to be like one of the things where if you know has things like warehouses good manufacturing get largely automated this is going to be the blue collar information systems work that is to be done labeling and things like that I think it's fair that you you nailed it that the if you look at the history of machine learning deep learning any of these things is just a creep creep Croft published a couple of years ago that shows the time when a new dataset got released to the public and when major breakthroughs happened in that space so I typically it's about 2 to 3 years
00:11:21there is an argument to be made and then I personally would probably believe in this that the vast majority of breakthrough that we see come from the availability of data and not shutting you out where those now right now get back to the amazing field. The truth is that I ate a large portion of the work that we are seeing the fruits of today came from work that was done in the mid-80s the late eighties back propagation connection connection systems
00:12:2192 and then being absolutely Blown Away by doing today it's a lot of it has to do with massive about to call you and we were running deep learning on shoes GPU machines that was in the tens of thousands and thousands of machines that are those particular kind of problem but often than they using the same number one of the big differences so is not only the computational power available but it's actually the data itself I can I can now in a week couple days even actually I saw this yesterday in 4 hours I can go from 0 to nothing from nothing to training on your own that one point five million images with 10,000 10,000 times and two weeks from now I can expect to actually have a fully
00:13:21keep lying model
00:13:31to understand how your application is performing you need visibility into your database vividcortex provides database monitoring app for my sequel postgres reddish mongodb an Amazon Aurora database of time efficiency and performance can all be measured using vividcortex don't let your database be a black box drill down into the metrics if your database with one second granularity database monitoring allows engineering team to solve problems and faster and ship better code vividcortex uses patented algorithms to analyze and surface relevant insights so users can be proactive and fix performance problems before customers are impacted if you have a database do you would like to monitor more closely check out vividcortex. Com s e daily to learn more GitHub digitalocean and Yelp all use vividcortex to understand database performance at vividcortex. Com SE daily you can learn
00:14:31more about how vividcortex works thanks to Vivek or text for being a new sponsor software engineering daily and check it out at vividcortex. Com SE daily
00:14:49I think of computing in terms of Moore's Law and think like it seems like as our systems have more and more emphasis on deep learning less of a question of Moore's Law because this is not really something that says much bottlenecks by a particular like processor architecture it's more like can you structure the problem in the right way do you have enough data and you cuz you can distribute all the computation so it's not bottleneck by any specific processor do you think that's right now. Variations that's at the end of two three four different variations of stochastic gradient descent that all completed distributable and
00:15:49the factor that is available means that we can build incredibly complex in on this and throw them in the hundreds and thousands of machines whatever else would think about it too much and the date and again the date of wines that doesn't mean it's not slow let's face it is slow as a painted models but is something very tractable NBA going back to the human computer what flow is that the biggest single issue you have a face and data science is getting a training said that use them so half of that is things like half of that is is creating a structure which allows you to have enough samples posted a negative the other half is just creating a sentence large enough and internet feature extraction what's up
00:16:49those are the things that you've got a couple of short videos how do you get people to to label things as funny well can you build a machine learning algorithm that can actually learn funny well you can you can argue that maybe maybe I'll answer is no but the you could sit around all day discussing the theoretical aspects of it as long as you to do it and then throw it into them if you can get that date and you can generate that into it and maybe you'll get a present but you have to have the ability to collect the data into work if not going into the creation of the citizen in the realization that how the brilliant day off unless they have they have a training Corpus don't really have the ability to build that much
00:17:49totally off topic it like discussing like the abstract concepts of deep learning and human labeling which is a fantastic fascinating Topic at your opinions are very interesting but getting back to giphy what are some of the hard parts of scaling a surgeon in his giphy is at this point search engine with a with a large API surface area there's a huge Corpus to be searching can you give me some description of what are the difficult part of scaling at those two ways to my best one is how do you how do you achieve a final solution and the truth is as a thousand ways of the cheating of final solution to build a coin ization that can grow into that and what are the what are the the fundamentals you put in place to to grow a substitution
00:18:49able concrete like that end of the answer to the soundless abstract about that like what would they mean by that is understanding but yes we have such a now we have sex let's say it's Ian witz on the order against in the order of hundreds of millions of documents now of gifts and the issue is going to be such a cool across a hundred million billion documents is a relatively you can you do that pretty easily and things like my Sequel and lastic such well within the range of being able to handle that kind of size of document the one that being what is yourself doing what does it do you use for an end when does the pink ones to frost the biggest pain point is latency we have very strict guidelines about how fast are you
00:19:49guy has to respond so for us the question isn't whether we can you just respond it's questionable can we respond with a latency that's on the Optima late and see if he had a very very rough we have a goal of a hundred seconds why I need any particular query in practice you know. That's a pretty good goal you looking in a mobile phone and you called latency from the network latency from various other pieces and so then we can only control so much of an interaction but at least we know that hundred seconds means it probably not going to be the bottleneck on a particular infraction those others so tracking. Is stuck on Step 1 Step 2 is is tracking every single little piece that goes into that if if I don't understand
00:20:49UC of hitting a particular application server from The Edge I'm probably not good if I don't have a complete picture of that if it was a kid signs of infrastructure I don't know how long will it take to use to get a database old to hit am I such cluster whatever else again will probably not in a good position to it to achieve that because we don't know how to optimize so how late is a continuous monitoring size and it's in the other thing you have to have internal goals that you too and wanted of a time and you have to make sure that if there's something that's not wanted you you how did the talk go on the company to add margins. So that it's this visibility around that and if you see the gradations making sure that you quickly identify why and how to fix in the future you're describing this trend in observability like herbs observability is a huge topic of conversation right now
00:21:49I had a really interesting interview with James Turnbull who is the I think the s e c t o and you read a book called The Art of monitoring which is all about how to build good monitoring pipelines how to do observability can you talk more about how you bake that into the Engineering Process perhaps into the culture of like cuz cuz I mean searches searches such an interesting problem where you like you really want the latency to be you really want the latency to be good especially when you when like people now expect a response of search or whatever like where every time you enter in a letter Lycra queries so you have to discuss observability little bit more so the book eventually I think it's great book I know I shouldn't be but in general I think that we've done
00:22:49roof the shipping code I mean by cycling evolution of the whole concept of devops as as teams that we eat we had the notion of a cooling systems group and they're on the job at the structure is code now different teams had a little bit more control over pieces and and that's good because people can deploy if you understand how things with the boys is a sense of the infrastructure on the line that got lost at some point in the running on the mote control of cloud more people have exposure to some level of infrastructure and aware of the fact that does not live run in some abstract void
00:23:49the Downs the next Evolution for that has been getting an understanding of performance into application life cycles does Bizzle optimization is a terrible thing to do how do you how do you get people to ship code understand and it's seen in production and be able to immediately identify what a problem is come in for the companies I've worked in the past every team has ownership of a complete dashboard that they can understand so if I'm shipping a piece of web code then as mentioned in my company I need to be
00:24:49able to use I should build understand how that performing on the web like book Rouses is it help me out was it before we on browsers information and be able to set up alerts so that I understand if I'm tracking well on that performance should be part of a project management if I own latency on a TV show and all the pieces should also own the ability to see you on the other phone and segregation is immediately if I'm running club code and there is something slowing down the web as part of that team I should know about it immediately I should be able to diagnose the problem and I should be able to work with others to fix it if it's in my piece of code Asheville to fix it and deploy if it's across the company I should be told him to those people to understand a little bit more and help out on that
00:25:49run exactly yeah it is to provide the right to link people you don't need to build Mall transolution you don't need to build the development of the deployment solution but you need to be able to own a house the tools necessary to be able to understand performance and an infrastructure giphy is search engine but it's also kind of a budding platform people can create gifts on top of it I imagine you have all kinds of interesting internal Services as well imagine you want to have a nice infrastructure that Engineers can randomly spit up machines and run experiments and stuff can you give me an overview of how that's built out in like what cloud you're running on and I guess just like how the
00:26:44how to culture and how the product is influenced how you built out that infrastructure hey I guess traditional 2017 shop work will using DACA we are actually chosen kubernetes right now has an underlying the Dolphins play The Sopranos what we are dealing with so many different endpoints we all effectively a polyglot organisation you know we have preferred tools preferred things but but in the end of day works at quote for you can always a fact that we will know one language is going to be going to rule them all so did it to work around that would have saw that would help of the helpless without weave
00:27:44movie quotes for work with communities as a developer I'm going to bring I can bring up the entire CD with a fart burn running out of memory a a miniature version of the entire stock most of the stack on my laptop in the chickenpox from scratch and I can put my code into a new cluster in the cloud that gets older configured and everything else again with couple of months and infrastructure is used for continuous integration is semen Specialties queen chess employment and within the next couple of months will be used in production as a primary it's a cloud infrastructure platform
00:28:35now we are witnessing the growth in popularity the growth in competitive potential of the Google Cloud particular driven by its kubernetes features as well as the high-level managed Services built on top of the Google Cloud that are differentiating though it seems that the Google cloud has been with the rise of the Google cloud has seen a lot of companies moving well I don't got a lot of companies but some companies at least feeling comfortable moving towards a polyglot Cloud world where they did a show with meet up recently a Meetup has a really interesting infrastructure where it's like a large percentage of his on Google and large percentage of his on Amazon and it's actually copacetic they're fine with that so what about you have you do you have it centralized or anything or you're just like kubernetes on Amazon and your you know hedging your bets or what
00:29:35right now we are running to the knives on Amazon and you know the truth is that we don't miss a sea of a reason to have multiple Cloud appointments at the moment I multi-class is definitely one of these things is creating an abstraction layer that makes it to irrelevant which Claudia running on and we see kubernetes as being fat solution so you don't answer your question no we don't have that Friday right now but they become so if it comes a day when I was like yeah we definitely need to do that will be in a good position to transfer some of these other higher-level apis actually right now we don't OK interesting have you evaluated those options are there too
00:30:35brittle most of the stuff out then it doesn't perform all that much better than when I can or you can download it. Maybe if she looks at the issue look at the absolute performance numbers maybe it's like 97% versus 98% but the reality is I'm looking at all content we don't care that there's a plate in the image we can have that there is a happy face that has been made out of sausages and so you know we have a we have our own domain the wood training has no working with but just isn't part of the equation for Google Translation episode I mentioned I do with unbabel they were talking about how it should be a lot of the translation where they do is it in the Dominican summer service I give his in desk ticket and they translated into different languages
00:31:35and so that builds their their machine learning training model to be biased towards the domain of customer service and show it it it makes it operating the way that is much different than these one-size-fits-all translation in her face is like the ones that you could you can get it at it as an API for Google off the shelf and even Google's on University was I believe the training set the Google originally was the year of the transcripts from the um so I don't know what kind of diplomatic we we have now you need to train you with the model for the purposes that you are looking for a fight if I don't know that if I models don't know
00:32:35that I want you to if if I don't know that hey with 3 Ys is different if not really the same thing as hey we just one way building any kind of graph structure that will let you relate different gifts to each other than your hair where else can you go and find a single piece of gifts and say okay find me other Beyonce GIFs that is closest to a human
00:33:35connect to cross whatever Google could just build this to Facebook and just build this like whatever whoever could build this but she was like one of these things where you like you look at you start to look at the problem under microscope and say this phone is actually way more complicated and takes way more engineering work then just somebody deciding paper to stand up a team at Google to make the gift search engine is that accurate but I mean absolutely I think there is an impression that it's incredibly simple thing to do and the reality is that it is an entire ecosystem it is it such it's thinking about how people Express themselves is thinking about what kind of how you identify content that is Meaningful and different
00:34:35building and they're all similarities like this the wood such the truth and I think we all have for the fall into that trap of trying to find the closest analogy but it truly is a different problem than anyone's ever dealt with before we wind up building building a bunch of technology that is some of the stuff for the week but if he has had to build to to be able to be giphy is absolutely ugly and it's around understanding gifts around manipulating gifts around indexing searching looking on the value of content understanding what what is relevant today was topical all these aspects
00:35:35and everything you can imagine every single problem to Twitter Google and Facebook in the behind will kind of combined into one but hidden behind the scenes
00:35:53incapsula is a cloud service that protects applications from attackers and improves performance incapsula sits between customer requests and your servers filters traffic preventing it from ever reaching our servers botnets and denial-of-service attacks are recognized by incapsula and blocked this protect your API servers and your microservices from responding to unwanted requests to try and capsule it go to incapsula. Com SE daily and get a month free for software engineering daily listeners in capsule is a p I give you control over the security and performance of your application whether you have a complex microservices architecture or a WordPress site and capsule has a Global Network of 30 data centers that optimize routing and cash your content the same network of data centers that are filtering your content for attackers are operating as a CDN and speeding up your application
00:36:53to try and capsule it today go to incapsula. Com SE daily and check it out thanks to encapsulate for being a new sponsor of software engineering get
00:37:10it does defy analogy to previous eras it's not like you can say it's like it's like YouTube cuz it's like video is kind of it is a no because you don't mess you don't frivolously message somebody a YouTube video to like the same way you would message an emoji but people do that with gifts to has such a different emotional valence of an emotional feelings depending on where you're using and how do you think it changes constantly you're talking to your grandmother friend and we need to support that language you use my change but it might not change top traumatically so how do you support that long time what about databases how do you store the level of complexity Liquors you could do stuff where it like you want to make the first enter the first two seconds of a girl the first X frames of a gift more accessible so you can preview it or something at I don't know can you give me
00:38:10baby description the storage model well I could decide and the killer horizontally across a sentence does on the storage structure yeah I think that's actually one of the aspects why we have very interesting that way we've been building out around and like how do you and what what what to do next Generation X Generation image and video pipelines truly look like how do you how do you move past
00:39:06ffmpeg is a great piece tool and everyone used to live in a massive massively distributed cloud system I know Point wasn't designed to to have like a Sub sub 200-200 my second response times and structure is being built around that and as far as well as far as we know so Raven next generation of agency scale-out image-processing interview about giving was this thing called giphy DVR which is an attempt to tackle the problem of we've got videos all over the internet but most of those have not been to
00:40:06further than two gifts and in a perfect world you would be able to query against giphy for any gift in any video that would be some substring or some sub concatenation of images in the video and giphy DVR system where like you can put in a video and it looks through the video and finds plausible gifts like you just met somebody tries to I guess use machine learning to try to figure out which of the areas in the video would be an interesting GIF giphy DVR work I think that's the way she can raise it pretty well I mean if you're looking at a particular pallavi call of the goal is to create a system that can look at that and say yes his four hours of the game but he is the 1660
00:41:06places that you care about and and PAW that is looking at Nothing by scene changes looking at the main statistic types of changes that will help you and it's by that example if I'm if I'm looking at a sports game for them to understand that yes there is a school and someone scores a goal so does peaches and create the beat the system that will automatically pretty highlights the important parts of that aliens are the hours of video how many
00:42:05yeah that's the kind of Destruction you need so when I sing shape images at transfer images Battles of the structure that helps run those kinds of algorithm. Scale when you have enormous amounts of dignity and come through the system a little bit about the team structure and the way the different teams work together and I guess the place I want to start at the there's kind of this interesting relationship between the creative team in engineering team I think because as I understand a giphy you have like a studio side in the giphy studio side is where you make gifts and you potentially work closely with Brands to get gifts made that you know it's like benefits the brand makes the brand look good and people maybe want to share like a Snickers GIF and it gets the brand recognition
00:43:05can you describe how you structure the relationship between the creative side and the engineering side if you need me I'm here we look at the the background of a lot of unit is the interest in the other lot of engine is him on there is a strong autistic vent weather or whether it's you know about photography music whatever else stronger pain with that thing. Loan concept that never interacts with anything else to keep interest in respect and fascination with think outside of did someone create a new algorithm that lets you switch numbers faster and
00:44:05and says like oh wow he is actually pretty. That's awesome keep trick treating the culture of a for the fascination of Oz of exploration I think you would Studios we we are the company we are exposed to the amazing what they do every single week and that we will they let you are presenting anything that the office that I haven't have done and we have autism residents we have automated we have absolutely no variety of different types of of opting create event for us as being supposed to have a regular basis and an often and
00:45:02I want things went well for releasing it will be releasing a lot more of this some of the incredible. Between those two pieces of technology technology there's a lot of stuff that we've done and you know she's very excited to to release these kind of these incredible projects but the public some of them have been done and shows so we we actually have had several of shows in the last 6 months even showing some of the intersections of just make sure you Check n Go on the website and attractive physical installations installations at leverage GIF infrastructure job gives a whole ecosystem around them to provide things that let you interact with animations as gifts and stickers
00:46:02I'm in the physical world I think that structure that you just articulate about giphy is what is the first companies where I really feel like it's leveraging what you get out of being in New York because if there's a lot to talk about like is it better to build your company in San Francisco or New York and and one thing that's for sure is in New York you're going to get more aggressive hybridization of like culture / medium / music / chart with in with engineering to little less of that in San Francisco I feel like I think that's like it's a real competitive advantage that if he has an especially when you're talking about breaking down the silos between like advertisers and in their fans and kind of like the marketing messages in like you know kind of the way that they
00:47:02online advertising works right now where you have like this church and state separation of who's making the advertising content versus who is Distributing it makes me feel like if he is really well positioned I mean I'm not I'm something I kind of agree with you two and a half years of thinking some Cisco and I've been around the country and more sterile it's almost there oh yeah it was a great experience and surrounded by other technologist but I actually I don't find any is created I don't find that people I feel that I think I think the problem when you have too many people in Tech thinking about type is that you lose track of of end-users you to lose track of what you should be able to give back to the World by the product because you become so entrenched in your own
00:48:02I will now I'm inside check has the same problem if the only thing I care about is is a click-through rate then I don't necessarily know what message I'm conveying to this year and being on the other end and I think that you can't because it's so decoupled from humans exactly about advertising fraud and where we asymptote towards in terms of advertising cuz Google is an advertising business Facebook as an advertising business I don't want to see these companies go anywhere but when you look at how problematic the state of online advertising is it I don't know it kind of worries me but I guess it will take awhile for advertisers to sort of wake up and see how much of the
00:49:01money is sort of going down the drain being viewed by Bots and and what not I mean have you thought of you thought deeply about the fraud problem or do you have any thoughts on it I know I don't know if I ever think deeply I think you know what the problem is obviously a real concern and Paula because for us if we don't if there is fraud if we have if we have people gaining all system it will create a user experience and in for us the primary goal of the company is to creative creative fantastic user experience so fraud is something that we can set it up for us and we want to make sure that at no point we get denatured by the equivalent of thinking about
00:50:01thinking to provide a more organic healthier interaction of LeBrons than has been possible with any of the ad Tech world out there and yeah but is an answer the movement it's are we going to have a movement where advertisers are going to have to start to say look this is kind of a lost cause I think it's funny because we start off a conversation about observability but the problem with Dad take his seems like observability drives fraud and maybe the solution is we don't observe very much or we have some some very crude high-level absurd observability bit that is so high level that it's very difficult for any individual fraudster to take advantage of
00:51:01what's the transparency of being the solutions a thing I think that I think I'd attack is the issue is not measurability I think the issue is the inability to measure the right and we and I think you know what will happen I hope what happens is that we don't move away and don't shy away from from visibility in measuring of
00:51:39because of the bad taste that gets left in the mouth off to the bank. They are without extra one but rather the way you get better at measuring the right thing that I've truly indicators of how people are interacting with you on tracking with you or your brand Contracting with you as a company or an individual by the most valuable are people making gifts aggressively and uploading them cuz I know that's that is like kind of a side feature of give you right now it seems like right now it seems like you're very focused on creating the gift content and letting people playing at the index why do people send gifs in messages really quickly and there's like you know you wanted to be able to send a gif of Taylor Swift make a cute expression or Kanye West like doing something weird what about the creative side hot like how much is gift creation growing
00:52:39a clean way of measuring that because I thought what I see when I see this we see gift creation and we can we see gifts. Exploding and we see people who in this that time at lunch you know maybe they have jobs as a graphic designer and lunch time they're creating amazing gifts and don't know if I can come by the most popular most viral gift for Obama most popular gift last year, and it is it's a side job but she does that and she does it every day and it's God's hundreds of millions of views and it's the kind of thing where as an artist you able to do that and I think that's
00:53:39read something well that is growing in as a phone as a develops isn't as those that form of Oz as a friend as a form of expression I think that that will get increasingly Lodge yeah we we we see that potential we see adoption with things like if you count as being something where people are increasingly creating their own content and and sometimes that sharing is sometimes a publishing it and sometimes they are leveraging all these things and you know we will we see a future where everyone has a set of of gift some of the things you going to regular basis how much are things that they created with whatever kind of to your tools we hope to see in the future what's the biggest Vision I just closed off what's the biggest division for what giphy to grow into a little bit
00:54:39it's hard to convey emotions it's hot in the death the opportunity for short phone content is to provide the. The richest medium to express emotions to convey a great deal of information and to be able to hold men's how people interact that one on one two groups in to communicate with a broad audience if if we are able to grow into all those things then we can we can change how people have communicated and provided a much better mechanism that anything's possible today coming on this operation Gillies make great conversation thank you have asked how they can support software engineering daily please write a review on iTunes or share your favorite
00:55:39search on Twitter and Facebook followers on Twitter at software underscored daily or on our Facebook group called software engineering daily please join our slack Channel And subscribe to your newsletter at software engineering daily. Com and you can always email me Jeff at software engineering daily.com if you're a listener and you want to give your feedback or your ideas for shows or your criticism I'd love to hear from you and of course if you're interested in sponsoring a software engineering daily please reach out you can email me Jeff at software engineering daily.com and thanks again for listening to the show

Transcribed by algorithms. Report Errata
Disclaimer: The podcast and artwork embedded on this page are from Software Engineering Daily, which is the property of its owner and not affiliated with or endorsed by Listen Notes, Inc.

EDIT

Thank you for helping to keep the podcast database up to date.
ADS

RECOMMENDATIONS