\u201cYou can create a chat manually. We have an add button \u2014 you can add any chat. It will be automatically recognized and suggested to other users who are interested in these sort of things. So we want every user to be heard within Capture.\u201d<\/span><\/p>\nHow it works<\/h2>\n Capture\u2019s AI-powered chatroom recommendations are designed to work as an onboarding engine for meeting relevant strangers online \u2014 using neural networks and machine learning to do the legwork of surfacing relevant people and chats.<\/p>\n
Here\u2019s how the mobile app works: Open the app, point the camera at something you view as a conversational jumping off point \u2014 and watch as it processes the data using computer vision technology to figure out what you\u2019re looking at and recommend related chats for you to join.<\/p>\n
For example, you might point the camera around your front room and be suggested a chatroom for \u2018interior design trends and ideas\u2019 , or at a pot plant and get \u2018gardeners\u2019 chat, or at your cat and get \u2018pet chat\u2019 or \u2018funny pets\u2019.<\/p>\n
Point the camera at yourself and you might see suggestions like \u2018Meet new friends\u2019, \u2018Hot or not?\u2019, \u2018Dating\u2019, \u2018Beautiful people\u2019 \u2014 or be nudged to start a \u2018Selfie chat\u2019, which is where the app will randomly connect you with another Capture user for a one-to-one private chat.<\/p>\n
Chat suggestions are based on an individual user\u2019s inferred interests and local context (pulled via the phone) and also on matching users across the app based on respective usage of the app.<\/p>\n
At the same time the user data being gathered is not used to pervasively profile uses, as is the case with ad-supported social networks. Rather Capture\u2019s founders say personal data pulled from the phone \u2014 such as location \u2014 is only retained for a short time and used to power the next set of recommendations.<\/p>\n
Capture users are also not required to provide any personal data (beyond creating a nickname) to start chatting. If they want to use Capture\u2019s web platform they can provide an email to link their app and web accounts \u2014 but again that email address does not have to include anything linked to their real identity.<\/p>\n
\u201cThe key tech we want to develop is a machine learning system that can suggest you the most relevant stuff and topics for you right now \u2014 based on data we have from your phone,\u201d continues Moiseenkov. \u201cThis is like a magical moment. We do not know who you are \u2014 but we can suggest something relevant.<\/p>\n
\u201cThis is like a smart system because we\u2019ve got some half graph of connection between people. It\u2019s not like the entire graph like your friends and family but it\u2019s a graph on what chat you are in, so where are you discussing something. So we know this connection between people [based on the chats you\u2019re participating in]\u2026 so we can use this information.<\/p>\n
\u201cImagine this is somehow sort of a graph. That\u2019s a really key part of our system. We know these intersections, we know the queries, and the intersection of queries from different people. And that\u2019s the key here \u2014 the key machine learning system then want to match this between people and interests, between people and topics, and so on.<\/p>\n
\u201cOn top of that we\u2019ve got recognition stuff for images \u2014 like six or seven neural networks that are working to recognize the stuff, what are you seeing, how, what position and so on. We\u2019ve got some quite slick computer vision filters that can do some magic and do not miss.<\/p>\n
\u201cBasically we want to perform like Google in terms of query we\u2019ve got \u2014 it\u2019s really big system, lots of tabs \u2014 to suggest relevant chats.\u201d<\/p>\n
Image recognition processing is all done locally on the user\u2019s device so Capture is not accessing any actual image data from the camera view \u2014 just mathematical models of what the AI believes it\u2019s seen (and again they claim they don\u2019t hold that data for long).<\/p>\n
\u201cMostly the real-time stuff comes from machine learning, analyzing the data we have from your phone \u2014 everybody has location. We do not store this location\u2026 we never store your data for a long time. We\u2019re trying to move into more private world where we do not know who you are,\u201d says Moiseenkov.<\/p>\n
\u201cWhen you log into our app you just enter the nickname. It\u2019s not about your phone number, it\u2019s not about your social networks. We sometimes \u2014 when you just want to log in from other device \u2014 we ask you an email. But that\u2019s all. Email and nickname it\u2019s nothing. We do not know nothing about you. About your person, like where you work, who\u2019s your friends, so on and so on. We do not know anything.<\/p>\n
\u201cI think that\u2019s the true way for now. That\u2019s why gaming is so fast in terms of growing. People just really want to share, really want to log in and sign up [in a way] that\u2019s easy. And there is no real barriers for that \u2014 I think that\u2019s what we want to explore more.\u201d<\/p>\n
Chatroulette<\/h2>\n Having tested Capture\u2019s app prior to launch I can report that the first wave chat suggestions are pretty rudimentary and\/or random.<\/p>\n
Plus its image recognition often misfires (for instance my cat was identified as, among other things, a dog, hamster, mouse and even a polar bear (!) \u2014 as well as a cat \u2014 so clearly the AI\u2019s eye isn\u2019t flawless, and variable environmental conditions around the user can produce some odd and funny results).<\/p>\n
\u00a0<\/p>\n
The promise from the founders is that recommendations will get better as the app ingests more data and the AI (and indeed Capture staff performing manual curation of chat suggestions) get a better handle on what people are clicking on and therefore wanting to talk with other users about.<\/p>\n
They also say they\u2019re intending to make better linkage leaps in chat suggestions \u2014 so rather than being offered a chatroom called \u2018Pen\u2019 (as I was),\u00a0 if you point the Capture camera at a pen, the app might instead nudge you towards more interesting-sounding chats \u2014 like \u2018office talk\u2019 or \u2018writing room\u2019 and so on.<\/p>\n
Equally, if a bunch of users point their Capture cameras at the same pen the app might in future be smart enough to infer that they all want to join the same chatroom \u2014 and suggest creating a private group chat just for them.<\/p>\n
On that front you could imagine members of the same club, say, being able to hop into the same discussion channel \u2014 summoning it by scanning a mutual object or design they all own or have access to. And you could also imagine people being delighted by a scanner-based interface linked to custom stuff in their vicinity \u2014 as a lower friction entry point vs typing in their directions. (Though \u2014 to be clear \u2014 the app isn\u2019t hitting those levels of savvy right now.)<\/p>\n
\u201cInternally we imagine that we\u2019re like Google but without direct query typing,\u201d Moiseenkov tells TechCrunch. \u201cSo basically you do the query \u2014 like scanning the world around you. Like you are in some location, like some venue, imagine all this data is like a query \u2014 so then step by step we know what people are clicking, then improving the results and this step by step, month by month, so after three month or four month we will be better. So we know what people are clicking, we know what people are discussing and that\u2019s it.\u201d<\/p>\n
\u201cIt\u2019s tricky stuff,\u201d he adds. \u201cIt\u2019s really really hard. So we need lots of machine learning, we need lots of like our hands working on this moderating stuff, replacing some stuff, renaming, suggest different things. But I think that\u2019s the way \u2014 that\u2019s the way for onboarding people.<\/p>\n
\u201cSo when people will know that they will open the app in the arena and they will receive the right results the most relevant stuff for this arena \u2014 for the concert, for the match, or something like that, it will be the game. That\u2019s what we want to achieve. So every time during the day you open the app you receive relevant community to join. That\u2019s the key.\u201d<\/p>\n
Right now the founders say they\u2019re experimenting with various chat forms and features so they can figure out how people want to use the app and ensure they adapt to meet demand.<\/p>\n
Hence, for example, the chatroulette-style random \u2018selfie chat\u2019 feature. Which does what it says on the tin \u2014 connecting you to another random user for a one-to-one chat. (If selfie chats do end up getting struck out of the app I hope they\u2019ll find somewhere else to house the cute slide-puzzle animation that\u2019s displayed as the algorithms crunch data to connect you to a serendipitous interlocutor.)<\/p>\n
They\u2019re also not yet decided on whether public chat content in Capture will persist indefinitely \u2014 thereby potentially creating ongoing, topics-based resources \u2014 or be ephemeral by default, with a rolling delete which kicks in after a set time to wipe the chat slate clean.<\/p>\n
\u201cWe actually do not know what will be in the next one to three months. We need to figure out \u2014 will it be consistent or ephemeral,\u201d admits Moiseenkov. \u201cWe need to figure out certain areas, like usage patterns. We should watch how people behave in our app and then decide what will be the feed.\u201d<\/p>\n
Capture does support private group chats as well as public channels \u2014 so there\u2019s certainly overlap with the messaging platform Telegram, which also supports both. Though one nuance between them is Capture Channels let everyone comment but only admins post vs Telegram channels being a pure one-way broadcast.<\/p>\n
But it\u2019s on interface and user experience where Capture\u2019s approach really diverges from the more standard mobile messaging playbook.<\/p>\n
If you imagine it as a mash-up of existing social apps Capture could be thought of as something like a Snap-style front end atop a Telegram-esque body yet altogether sleeker, with none of the usual social baggage and clutter. (Some of that may creep in of course, if users demand it, and they do have a reactions style feature linked up to add in so\u2026 )<\/p>\n
\u201cWith our tool you can find people not from your graph,\u201d says Moiseenkov. \u201cThat\u2019s the key here. So with WhatsApp it\u2019s really hard to invite people not from your graph \u2014 or like friends of friends. And that\u2019s a really tough question \u2014 where I can find the relevant people whom I chat about football? So now we add the tool for you in our app to just find these people and invite them to your [chat].\u201d<\/p>\n
\u201cIt\u2019s really really hard not to like your friend\u2019s post on Instagram because it\u2019s social capital,\u201d he adds. \u201cYou are always liking these posts. And we are not in this space. We do not want to move in this direction of followers, likers, and all this stuff \u2014 scrolling and endless communication.<\/p>\n
\u201cTime is changing, my life is changing, my friends and family somehow is changing because life is changing\u2026 We\u2019re mobile like your everyday life\u2026 the app is suggesting you something relevant for this life [now]. And you can just find people also doing the same things, studying, discussing the same things.\u201d<\/p>\n
Community building<\/h2>\n Why include private chats at all in Capture? Given the main premise (and promise) of the app is its ability to combine strangers with similar interests in the same virtual spaces \u2014 thereby expanding interest communities and helping mobile users escape the bubbles of closed chat groups.<\/p>\n
On that Moiseenkov says they envisage communities will still want to be able to create their own closed groups \u2014 to maintain \u201ca persistent, consistent community\u201d.<\/p>\n
So Capture has been designed to contain backchannels as well as open multiple windows into worlds anyone can join. \u201cIt\u2019s one of opportunities to make this and I think that we should add it because we do not know exact scenarios right from the launch,\u201d he says of including private conduits alongside public chats.<\/p>\n
Given the multiple chat channels in the first release Capture does risk being a bit confusing. And during our interview the founders joke about having created a \u201cmaximal viable product\u201d rather than the usual MVP.<\/p>\n
But they say they\u2019re also armed to be able to respond quickly to usage patterns \u2014 with bits and pieces lined up in the background so they can move quickly to add\/remove features based on the usage feedback they get. So, basically, watch this space.<\/p>\n
All the feature creep and experimentation has delayed their launch a little though. The app had been slated to arrive in Q4 last year. Albeit, a later-than-expected launch is hardly an unusual story for a startup.<\/p>\n
Capture also of course suffers from a lack of users for people to chat to at the point of release \u2014 aka, the classic network effect problem (which also makes testing it prior to launch pretty tricky; safe to say, it was a very minimalist messaging experience).<\/p>\n
Not having many users also means Capture\u2019s chat suggestions aren\u2019t as intelligent and savvy as the founders imply they\u2019ll be.<\/p>\n
So again the MVP will need some time to mature before it\u2019s safe to pass judgement on the underlying idea. It does feel a bit laggy right now \u2014 and chat suggestions definitely hit and miss but it will be interesting to see how that evolves as\/if users pile in.
Part of their plan is to encourage and nurture movie\/TV\/entertainment discussion communities specifically \u2014 with Hardy arguing there\u2019s \u201cno such tool\u201d that easily supports that. So in future they want Capture users to be notified about new series coming up on Netflix, or Disney\u2019s latest release. Then, as users watch that third party content, their idea is they\u2019ll be encouraged to discuss it live on their mobiles via Capture.<\/span><\/p>\nBut movie content is only partially launched at this stage. So again that\u2019s all just a nice idea at this stage.<\/p>\n
Testing pre-launch on various celebrity visages also drew a suggestive blank \u2014 and Hardy confirmed they\u2019ve got more pop culture adds planned for the future.<\/p>\n
Such gaps will likely translate into a low stickiness rate at first. But when the team\u2019s ambition is to support a Google-esque level of content queries the scale of the routing and pattern matching task ahead of them is really both massive and unending.<\/p>\n
To get usage off the ground they\u2019re aiming to break the content recommendation problem down into more bite-size chunks \u2014 starting by seeding links to local events and news (sourced from parsing the public Internet); and also by focusing on serving specific communities (say around sports), and also linked to particular locations, such as cities \u2014 the latter two areas likely informed by in what and where the app gets traction.<\/p>\n
They\u2019ve also hired a content manager to help with content recommendations. This person is also in charge of \u201cbanning some bad things and all that stuff\u201d, as they put it. (From the get go they\u2019re running a filter to ban nudity; and don\u2019t yet support video uploads\/streams to reduce their moderation risk. Clearly they will need to be very \u2018on it\u2019 to avoid problem usage mushrooming into view and discouraging positive interactions and community growth within the app. But again they say they\u2019re drawing on their Prisma experience.)<\/span><\/p>\nThey also say they want this social app to be more a slow burn on the growth front \u2014 having seen the flip side of burn out viral success at Prisma \u2014 which, soon after flooding the social web with painterly selfies, had to watch as tech giants ruthlessly cloned the style transfer effect, reducing their novelty factor and pushing users to move on to their next selfie lens fix.<\/p>\n
\u201cAs data-driven guys we\u2019re mostly looking for some numbers,\u201d says Moiseenkov when asked where they hope to be with Capture in 12 months\u2019 time. \u201cSo I think achieving something like 1M or 2M MAU with a good retention and engagement loop by then is our goal.<\/p>\n
\u201cWe want to keep this growth under control. So we could release the features step by step, more about engagement not more about viral growth. So our focus is doing something that can keep engagement loop, that can increase our spend time in the app, increase the usage and so on, not driving this into the peak and like acquiring all the trends.\u201d<\/p>\n
\u201cConclusions are drawn from Prisma!\u201d adds Hardy with investor-winning levels of chutzpah.<\/p>\n
While it\u2019s of course super early to talk business model, the question is a valid one given Capture\u2019s claims of zero user profiling. Free apps backed by VC will need to monetize the hoped for scale and usage at some point. So how does Capture plan to do that?<\/p>\n
The founders say they envisage the app acting as a distribution tool. And for that use case their knowing (only) the timing, location and subject of chats is plenty enough data to carry out contextual targeting of whatever stuff they can get paid to distribute to their users.\u00a0\u00a0<\/span><\/p>\nThey are also toying with models in a Patreon style \u2014 such as users being able to donate to content authors who are in turn distributing stuff to them via Capture. But again plans aren\u2019t fully formed at this nascent stage.<\/p>\n
\u201cOur focus right now is more like going into partnerships with different companies that have lots of content and lots of events going on,\u201d says Hardy. \u201cWe also are going to ask for permission to get access to music apps like Spotify or Apple Music to be aware of those artists and songs a person is interested in and is listening to. So this will give us an opportunity to suggest relevant new albums, maybe music events, concerts and so on and so forth.<\/span><\/p>\n\u201cFor example if a band is coming to your city and we know we have access to Apple Music we know you\u2019re listening to it we\u2019ll suggest a concert \u2014 we\u2019ll say \u2018hey maybe you can win a free ticket\u2019 if we can partner\u2026 with someone, so yeah we\u2019re moving into this in the near future I think.\u201d<\/p>\n
Article source: http:\/\/feedproxy.google.com\/~r\/techcrunch\/social\/~3\/JCfugdns0iY\/<\/p>","protected":false},"excerpt":{"rendered":"
At first glance launching a new social app may seem as sensible a startup idea as plunging headfirst into shark-infested waters. But with even infamous curtain-ripper Facebook now making grand <\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"jetpack_post_was_ever_published":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[3],"tags":[10,7,14,15,6,12,11],"class_list":["post-17130","post","type-post","status-publish","format-standard","hentry","category-latest-update","tag-computer-news","tag-downloads","tag-microsoft-download-news","tag-networking-updates","tag-technology-update","tag-wordpress-plugin-download","tag-wordpress-theme-download"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t
![\"Discovery](\"https:\/\/i0.wp.com\/www.wordpressdynamos.com\/timesofnewscdn\/blog\/1ce9f_Discovery_1.png?resize=382%2C680&ssl=1\" "\\"Discovery_1\\"")
<\/p>\n![\"Capture](\"https:\/\/i0.wp.com\/www.wordpressdynamos.com\/timesofnewscdn\/blog\/746f8_IMG_0502.png?resize=382%2C680&ssl=1\" "\\"Capture")
<\/p>\n![\"Capture](\"https:\/\/i0.wp.com\/www.wordpressdynamos.com\/timesofnewscdn\/blog\/62387_Capture.jpg?resize=640%2C428&ssl=1\" "\\"Capture")
<\/p>\n![\"capture](\"https:\/\/i0.wp.com\/www.wordpressdynamos.com\/timesofnewscdn\/blog\/b46c2_Case_camera_2_2.png?resize=382%2C680&ssl=1\" "\\"capture")
<\/p>\n