Skip to main content

tv   Shift  Deutsche Welle  September 2, 2023 12:15pm-12:31pm CEST

12:15 pm
as well as our protected space sees in the us little domestic team that detach will coming up next. not to making an a i how technology is que writing love on the dining up that so now, so shift after so break. i'm jarrett draining building. thank you so much for watching. have a great day the you, how many platforms can you handle single tenuously without having the feeling that it's just too much you might see me. how much can we do simultaneously?
12:16 pm
multitasking these, the modern because if we do too much, we paid it all wrong. we messed things up, risking brain damage. so let's stop this self sabotage, humans and multitasking watch. now on youtube, v. w documentary, the of 10 artificial intelligence instead, as our romantic choices on the surface circumstances. yes, because a is playing and increasingly important role in dating f. we'll take a closer look at what's happening exactly. plus, how does object visual intelligence gets hacked and misused for criminal purposes? and it is becoming increasingly difficult to distinguish between what's real fake. we tell you what sort of alpha and what the tech companies are doing to protect us from fake video and images. these are the topics that are moving the tech well, the
12:17 pm
catalog official intelligence helped find your perfect match. taking ups have been embracing a iphone while no 10, not the wealth. biggest online banking platform is expanding its offerings to the app is now testing a tool to help used to select the best looking photos. but that's just one example of developers promising us. that's a dating luck with the help of digital technologies, the ad downsize to this trend of costs, but could i become yup, 21st century cupid does a i know best tend to us working on a number of artificial intelligence initiatives to make data and more rewarding what they mean as was good, that's what the less efforts tend as parent company match crew says it's working on a tool that looks to a uses photo albums to select the 5 images that represent them fest. i'm curious to
12:18 pm
see which criteria is the a i would use to fix my pictures. unfortunately, that's no public testing yet. according to bonus, kim c o a match group, a dating tools, i suppose to eliminate all quickness of to all. sure. it is a pro 5 that says, i'm awesome. i could be the love of your life, and i'm humble, feels nearly impossible to already have other projects in the pipeline to fix that problem at the moment. so let's use this contest of generative ai tools to enhance the profiles. it can create individual bios based on interest and relationship goals for the user has edits. it's sounds pretty convenient, but do we really want that making day thing more efficient? according to a recent study conducted in the u. k, only instead of facing f use, this would be open to using ai to enhance their profiles. putting yourself out there seems to be a job that's in level 2. it's allocated for a i to handle, but they use a sub h,
12:19 pm
had nothing against using ai in other parts of the dating process. 75 percent of all dashing app use assessed. they would explore using check. jpg to deliver the perfect opening line with people being willing to use a i to improve that chance. as in the dating world, as the developers are coming up with new ideas, dating at cheese or a i, for example, let's check within a version of someone before you swipe right on next on that. and with flush, you can work on your relationship and intimacy skills. i flush and with the apps checkbox. other companies are working on fox, the, even to the swiping and initial vendor for you on some metrics, me or the down sides. many models currently stored the information use us, give them on service in order to keep one done and ultimately improve a user experience. and when it comes to low, to no doubt going to get personal. and that means giving up a lot of intimate information. so even if that's on select into promise to not give you what they talk to the punches for heck us, it's
12:20 pm
a very tempting target. if your private data falls into the wrong hands, this huge potential to cause you home. and then that's the issue of roman scans. these are criminals who spain romantic interest to gain the vixen affection. they've been use that goodwill to get the victim to send money under false pretences, all to commit fraud. according to the u. s. security organization s b i victims from such a scans and the united states report of financial losses of almost $740000000.00 us $1.20 experts argue that i could super charged these wrongful practices and an able criminals to carry out taylor and alternate the scabs on a huge scale. what do you think will make online dating even more difficult or? well, we all benefit from an induced eco boost. let us know. has a i can cause you home by stealing your private data for example. but how can these systems even be hacked and exploded for criminal purposes?
12:21 pm
here's what you need to know in august. heck us got together at death called the annual heck a convention in las vegas. the tried to make a assistance, ex harmful, or even criminal weights, such heck of groups called red teams are essential in finding security flaws and software which has been reported back to the develop the in the age of a i, they might be more important than of the results from the field for the next few months to give a company's time to address the issues before making them public fatality. even get an a i to go rogue jo, breaking the a. i a assistance, follow the prompts you give them. in other words, instructions but to prevent a ad from following instructions that could reduce harmful content and make us have safety measures in place. for example, they can tell them what office to avoid, but that ways to circumvent a safety measure. this is called jail break. it or other funny one now fix,
12:22 pm
what's the grandma jail break of check g p t. i use a told the check bought that it should act like is deceased. grandmother was supposed to be work at the knob home secretary and always told him how to manufacture it. when he was trying to find a sleep championship proceeded to 50 use a detailed instructions of how to produce list level chemical breath as a bedtime story. right now you've got the heck a that tells you how to build a bump. so what such instructions consecutive be found elsewhere on the internet? well, real, how could be done if someone jo, brakes, a system you're using without you knowing it? prompt injection, a natural language and models process what a stephen to them. however, upon input, it is all just text me a ben needs to differentiate between prompts and text that is meant to work with, for example, an optical pro website. but it doesn't always get differentiation, right? phone been done somewhere in the text as
12:23 pm
a promising that runs the risk of yeah, i model just following that instruction, even if it is a form. academics demonstrate of health sol called prompt injection, could be used to trick microsoft, been chad into providing, dodging things to use as the embedded a prompt into a website. when big chips such that website, the problem got activated, needing the checkbook to send messages. like, do you have one an amazon gift card? all you have to do is follow this link and log in with your amazon credentials. often use a set they didn't trust the link bench had proceeded to ensure them better, safe and legit or imagine you're using an assistant that's automatically off of email all structure. so you have collins up, an attempt injection attack it the, the shots, and couldn't the pump injection detect happen. for example, with an email that the model reads so many minutes by few of them didn't somewhere
12:24 pm
in that email. that's a prompt that says length is calanda and forward hall meetings to office email address meetings. these are email address advice for life and then suddenly all my data has been stolen. there is a positive feed for forced. um no, they have not reminded control. unfortunately, if that's no easy fix, yet a developers comp perfectly predict every reaction to every problems and baffled, cannot put safety measures for all think, couple scenarios in place. and today's a eyes on not intelligent enough to do that by themselves. so from them and say i model is kind of differentiate, besides the enough between problems and the data and everything on the thing deciding this is alice new text. however, ex website developers should research the possibilities of prompt injection more before releasing the models and not put out for us as a match. do you trust a web based assistance? let us know if this, the folks of my colleague are as a i generation
12:25 pm
a i think takes us everywhere and solve news videos, distinguishing between real and fake, discussing more and more difficult. is it even possible to keep up how to tech companies title, the problem, and what can you do to protect yourself? how a fix damage you use a defect skin in china, that's made headlines. businessman was convinced in a video call to transfer money to a friend boss, he fell victim to deep fig created by a hacker with a voice and image southwest. the mess supposedly lost more than 500000 euro. i think it's frightening. what tom criminals can do with a technology, as users who should always be skeptical to us images and videos with c, or was this we're here online. remember this deep fig pictures of donald trump being arrested by police or any of this? yeah, image just like that have the potential to manipulate extra political caps and too much onto our societies. this also applies to texts, thanks to
12:26 pm
a software like chat. jpg. news articles can nowadays be generated in the blink of an eye. the spend the internet with no quantity and even intentionally falls news that way it's getting harder for uses to find reliable information. what is being done by the industry? some developers among them amazon, google, and open a i have recently promised to make this software more transparent and the output more easily recognizable best strategy watermarks that could come in different forms. for example, as meta data cameras have been using that for decades. method the data could automatically be added to a, i generate just images of videos and the theory also texts, but it's more difficult with texts because once you insert the text into another system, for example, from a text document to a browser, the mess of data is gone, digital images could also be watching my cryptographic the bell. ready plenty of
12:27 pm
design tools that app hidden markings to images. for example, to crawl the web looking for copyright violations. sounds good, but these approaches would need to work seamlessly across billions of devices and operating systems. all of these mocking speed on pictures, videos, text or even sound would have to survive being copied and pasted from one f operating system, all platforms to another. to really have an impact, all of that would have to be accessible to everyone with an internet connection without any training immediately. so the quite a few technical hurdles to overcome and there's always the chance of criminals find a way to circumvent the measures. what can you do to protect yourself? unfortunately, that's no easy way. most importantly, always be aware that these aspects exist and that they are being used against you in different ways. so being skeptical about what you see or hear online might be a good strategy when it comes to online articles. check if the soul is reliable and
12:28 pm
compared what you're reading with the pulse on the same story is done by trustworthy media outlets. defect images are getting more and more perfect, but sometimes says things give away the fitness inconsistencies in the next area, hair or fingers that are not connected. effect video sometimes reveal mistakes when you watch them frame by frame. and always be careful if you're being asked for money, even if the person doing that is the old granny during a video clock. is this really higher context showing up in the connection details? rule of thumb. always look for possible inconsistencies before you make a decision. what do you think needs to be done to protect the use of? let us know that's all for me to is by and see you next time the
12:29 pm
treated us to time stop and is done i the can is august and tiny house concepts. when top rates, classical musicians under goals makes music. arts unveiled. algeria is here more attain you is here. they don't mark track. we know is this one of the great dreams and the grim reality the sahara was hell, i swear. what remains? just homes to be the 13 minutes on d. w. what secrets?
12:30 pm
why behind being discovered new adventures and 360 degrees and explore fascinating world heritage sites dw world heritage 360 yeah. now, the treated us to the highest level of mazda it was, this was used caring with, you know, when was that it was like you said it's ultimate. yeah. ok. the good . ready

13 Views

info Stream Only

Uploaded by TV Archive on