Connect with us

Tech News

A deepfakes creator for text so realistic it can’t be made public yet

(TECHNOLOGY) You know about video deepfakes, but the technology exists for doing convincing deepfakes for text. It’s so good that they aren’t ready to release it to the public yet…

Published

on

deepfakes text

Artificial intelligence is being used to complete more and more human tasks. But as of right now, news stories you read online – including all the articles here on American Genius – have been written by real human beings.

Until recently, even the most intelligent computers couldn’t be trained to recreate the complex rules and stylistic subtleties of language. AI-generated text would often wander off topic or mix up the syntax and lack context or analysis.

However, a non-profit called OpenAI says they have developed a text generator that can simulate human writing with remarkable accuracy.

The program is called GPT2. When fed any amount of text, from a few words to a page, it can complete the story, whether it be a news story or a fictional one.

You already know about video deepfakes, but these “deepfakes for text” stay on subject and match the style of the original text. For example, when fed the first line of George Orwell’s 1984, GPT2 created a science-fiction story set in a futuristic China.

This improved text generator is much better at simulating human writing because it has learned from a dataset that is “15 times bigger and broader” than its predecessor, according to OpenAI research director, Dario Amodei.

Usually researchers are eager to share their creations with the world – but in the case, the Elon Musk-backed organization has, at least of the time being, withheld GPT2 from the public out of fear of what criminals and other malicious users might do with it.

Jack Clark, OpenAI’s head of policy, says that the organization needs more time to experiment with GPT2’s capabilities so that they can anticipate malicious uses. “If you can’t anticipate all the abilities of a model, you have to prod it to see what it can do,” he says. “There are many more people than us who are better at thinking what it can do maliciously.”

Some potential malicious uses of GPT2 could include generating fake positive reviews for products (or fake negative reviews of competitors’ products); generating SPAM messages; writing fake news stories that would be indistinguishable from real news stories; and spreading conspiracy theories.

Furthermore, because GPT2 learns from the internet, it wouldn’t be hard to program GPT2 to produce hate speech and other offensive messages.

As a writer, I can’t think of very many good reasons to use an AI story generator that doesn’t put me out of job. So I appreciate that the researchers at OpenAI are taking time to fully think through the implications before making this Pandora’s box of technology available to the general public.

Says Clark, “We are trying to develop more rigorous thinking here. We’re trying to build the road as we travel across it.”

Ellen Vessels, a Staff Writer at The American Genius, is respected for their wide range of work, with a focus on generational marketing and business trends. Ellen is also a performance artist when not writing, and has a passion for sustainability, social justice, and the arts.

Continue Reading
Advertisement
2 Comments

2 Comments

  1. Pingback: A deepfakes creator for text so realistic it can’t be made public yet – tech news

  2. Did You Know?

    March 13, 2019 at 2:44 pm

    They did this because they went from non-profit to for-profit; putting a lesser of their product out there so they can sell the 3x stronger AI to those who wish to purchase it.

Leave a Reply

Your email address will not be published. Required fields are marked *

Tech News

Study finds 1,000 phrases that accidentally activate smart speakers

(TECH GADGETS) Don’t worry about accidentally activating your nosy smart speakers… unless, of course, you utter one of these 1,000 innocuous phrases.

Published

on

smart speakers

It’s safe to say that privacy concerns, especially in today’s digital era, are unquestionably valid. With new video recording technology making it easier to identify people at a glance (whether they like it or not) and concerns that your smart speakers are eavesdropping on you, it may feel like you’re bordering on slightly paranoid around modern technology.

After all, even though there have been cases of smart speakers picking up on intimate conversations, there’s absolutely no risk of them overhearing private things without your consent, right? Even though it’s been documented that these devices — including Cortana, Alexa, Siri, and Google Home — have listened in relationship spats, criminal activity, and even HIPAA-protected data, you’re totally in the clear.

Oh yeah. The thing is, everything that gets broadcast into your smart speaker? There’s a completely random chance that someone back at headquarters may decide to sift through it in order to improve AI learning.

And while most of the time these conversations are totally benign, it doesn’t change the fact that a complete stranger is getting an earful of your private life. In fact, these transmissions? Are actually completely admissible in court, as several murder cases have already demonstrated. Their key evidence was none other than poor Alexa herself.

But wait, wait. These smart speakers can only get your information if you activate them, and that requires you to clearly enunciate their names. Right? Um. Not exactly. Even though you may think that you need to speak crisply into the speaker to activate it, it turns out that these devices are highly sensitive to any suggestion that you might be talking to them. It’s almost like your dog when you even remotely glance at his bag of doggie treats in the corner: one crinkle and Fido comes running, begging for some kibble and ready to serve you.

It’s the same for your smart speakers. As it turns out, there are over a thousand words or phrases that can trigger your device and invite it to start recording your voice. These can range from the perfectly reasonable (Cortana hearing “Montana” and springing to attention) to the downright absurd (Alexa raising her hackles over the words “election” and “unacceptable”). Well, crap. Now what?

It’s no secret that someone is listening in on your conversations. That’s been clearly documented, researched, dissected, and even accepted at this point. However, if you thought that they’d only listen to it if you gave them implicit permission by activating your device (which, to be fair, should not even count as permission in the first place), you were wrong.

So what’s a privacy-loving person to do? Just suck it up and try to choose between the lesser of two evils? On one hand, yes, these smart speakers are super convenient and can make your life easier. On the other?

Well, if you’re a fan of your privacy, then perhaps these devices aren’t meant for you. At this point, you’ve got little recourse. These companies will continue to use your data, and there’s nothing stopping them from spying on you. That is, unless you prevent them from doing it in the first place.

If you want to keep your private conversations private, either unplug your smart speaker when you’re not using it, or don’t get one in the first place. Otherwise, you’ll continue to give your implied consent that you’re totes cool with them butting in on your personal life, and they’ll continue to be equally totes cool with using it without your permission.

Continue Reading

Tech News

HEY needs to fix its issues to be the Gmail killer it claims to be

(TECH NEWS) You would hope that HEY, the paid email service, would launch without issues but it has a few. Let’s hope some of that money goes to fixing them.

Published

on

Hey email

Last week, we covered HEY–a new email service that seemingly has a lot to offer–and while we largely praised the service despite it being a paid client awash in a sea of free email options, not everyone is fully on board with HEY’s inimitable charm–at least, not yet.

Adam Silver, an interaction designer focused on user experience, had some criticisms of HEY–many of which he identified as “pretty surprising oversights.” Though Silver does mention that his overall opinion of the service is good, these oversights are the focus of his review.

“HEY isn’t very accessible,” says Silver in his notes. His assessment, while self-admittedly not a holistic view, includes issues related to JavaScript (specifically when it isn’t enabled, which is something more and more companies are requiring) and lack of reasonable keyboard shortcuts for anyone using a screen reader. As Silver points out, these are fairly simple–and, thus, surprising–problems that probably should have been caught from the onset.

“All of these things are really easy to fix,” amends Silver.

Another issue Silver highlights is the inbox (imbox?) sorting. As we mentioned previously, there are three locations for email: the imbox, the feed, and the paper trail, each of which serves a different purpose. The problem with this system is that organizing emails by only three overarching categories affords little flexibility; furthermore, Silver notes that the menu for accessing each folder leaves a lot to be desired from a design standpoint.

The feed is also the subject of Silver’s criticism in that it doesn’t function enough like a traditional inbox to the point that it is actually difficult to use. Especially given the feed’s purpose–to store newsletters and such in a free-scrolling manner–this is a hold-up for sure; coupled with the feed’s lack of notifications, you can see how this problem cripples the user experience without active attention to the ancillary feed inbox.

Lastly, Silver mentions that the name “imbox” is, well, stupid. “This is not a typo but it’s not good,” he says. “You need a really good reason not to keep things simple.”

This is actually a point that we initially glossed over in our overview, but it’s another instance of a company doing a little too much to stand out–and, in doing so, potentially disrupting the user experience. “Keeping it simple” by calling the delivery place for your email the “inbox” won’t sink your brand, and the name “imbox” is sure to, at best, annoy.

It’s important to reaffirm that HEY’s driving principle–accessible email that prioritizes your privacy and charges you a relatively nominal fee for doing so–is good, and that’s the tough part of any app’s development; should they choose to follow Silver’s lowkey advice and make a few tweaks, they’ll have a winning product.

Continue Reading

Tech News

Live captioning via AI is now available for Zoom, if a little limited

(TECH NEWS) In order to be more inclusive, and improve the share of information with your team, live captioning is a great option for your next Zoom call.

Published

on

Zoom live captioning

The ubiquitous all-father Zoom continues to prompt innovation–and in a time during which most companies are still using some form of remote communication, who can blame them? It’s only fitting that someone would come along and try to flesh out Zoom’s accessibility features at some point, which is exactly what Zoom Live Captioning sets out to accomplish.

Zoom Live Captioning is a Zoom add-on service that promises, for a flat fee, to caption up to 80 hours per month of users’ meetings via an easy-to-implement plugin. The allure is clear: a virtual communication environment that is more time-efficient, more accessible, and more flexible for a variety of usage contexts.

Unfortunately, what’s less clear is how Zoom Live Captioning proposes to achieve this goal.

The live-captioning service boasts, among other things, “limited lag” and “the most accurate [speech-to-text AI] in the world”–a service that, despite its sensational description, is still only available in English. Furthermore, anyone who has experienced auto-captioning on YouTube videos–courtesy of one of the largest technology initiatives in the world–knows that, even with crystal-clear audio, caption accuracy is questionable at best.

Try applying that level of moving-target captioning to your last Zoom call, and you’ll see what the overarching problem here is.

Even if your Zoom call has virtually no latency, everyone speaks clearly and enunciates perfectly, your entire team speaks conversational English at a proficient degree across the board, and no one ever interrupts or experiences microphone feedback, it seems reasonable to expect that captions would still be finicky. Especially if you’re deaf or hard of hearing–a selling point Zoom Live Captioning drives home–this is a problematic flaw in a good idea.

Now, it’s completely fair to postulate that any subtitles are better than no subtitles at all. If that’s the decision you’d like to make for your team, Zoom Live Captioning starts at $20 per person per month; larger teams are encouraged to contact the company to discuss more reasonable rates if they want to incorporate live captioning across an enterprise.

Nothing would be better for speech-to-text innovation than being wrong about Zoom Live Captioning’s potential for inaccuracy, but for now, it’s safe to be a little skeptical.

Continue Reading
Advertisement

Our Great Partners

The
American Genius
news neatly in your inbox

Subscribe to our mailing list for news sent straight to your email inbox.

Emerging Stories

Get The American Genius
neatly in your inbox

Subscribe to get business and tech updates, breaking stories, and more!