Deepfakes spiked more than 245% this year

Subscribe | Advertise

Good morning. Elon Musk and Yann LeCun have been fighting on Twitter, OpenAI keeps publishing new blog posts and more media heavyweights have joined OpenAI’s side.

But somehow we made it to Friday.

In today's newsletter: 

• 💹 Gamma: A presentation app for the AI era 

• 🛜 The real problem with Google’s AI Overviews

• 🌎 OpenAI shuts down five global covert influence operations 

• 📈 Research: Deepfakes spiked more than 245% this year

*Gamma: A presentation app for the AI era

Microsoft PowerPoint, believe it or not, is 37-years-old. It was introduced in 1987, before the internet era really began, before social media entered onto the scene and long before ChatGPT was even a blip in the mind of Sam Altman (who was two years old at the time). 

Integration, integration, integration: PowerPoint didn’t start to get fancy new AI features until 2015 with Designer, which generated thematic design options for its users. Designer got a big upgrade in 2019, and more recently, Microsoft announced its AI-powered copilot tool (which is designed to integrate with Microsoft Office). 

• Google is similarly testing AI features with Slides as it strives to bring generative AI features to all of its services, from Search to Google’s entire suite of apps and products. 

A new approach for a new paradigm: Grant Lee, the CEO of Gamma, thinks the time is ripe for a new presentation app, not one that has AI integrations, but one that has AI built directly into the architecture.

• Lee likes to think of Gamma as your “AI design partner.” Someone with no design skills (and no coding ability) can give Gamma’s AI a few brief thoughts and can quickly unlock professional-looking, unique, highly customizable slides. 

• Users can also create mobile-friendly cards & can reinvigorate (or Gammify, if you will) existing documents and slides into more dynamic presentations. 

“It's almost like having a professional designer sit next to you,” Lee said, adding that, unlike the traditional competition, Gamma’s platform also works as an asynchronous content-sharing platform (thank you, The Cloud).

Gamma — which recently raised $12 million in Series A funding — is bent on disrupting and revitalizing the industry of business communication. 

You can join Gamma’s 18 million other users – for free – in dramatically upping your presentation game right now, right here. 

*Indicates sponsored content

💰AI Jobs Board:

• Founding Engineer: Lume · United States · New York, NY · Full-time · (Apply here)

• Head of Product Design: GPTZero · United States · Hybrid; New York, NY · Full-time · (Apply here)

• Intelligence Specialist: MIT Lincoln Lab · United States · Lexington, MA · Full-time · (Apply here)

 📊 Funding & New Arrivals:

• Bessemer Venture Partners is in talks to lead a $250 million investment in Perplexity (The Information). The round would value Perplexity at $3 billion.

• Generative AI platform Maven AGI has raised $28 million in funding.

• Kintsugi AI, an AI platform to automate sales tax compliance, raised $6 million in Series A funding.

🌎 The Broad View:

• News publishers are making a deal with the devil in partnering with OpenAI — from Damon Beres, a senior editor at The Atlantic, the same day The Atlantic announced a deal with OpenAI (The Atlantic).

• How AI is impacting the 2024 elections (Wired).

• L’Oreal is working on a bioprinted skin that it says can “feel” (CNBC).

*Indicates a sponsored link

The real problem with Google’s AI Overviews

We’ve talked a lot about Google’s less-than-stellar rollout of AI Overviews, which most notably suggested that users should add a bit of (non-toxic) glue to their pizza sauce to make sure the cheese sticks. 

But moving past the goofy or downright alarming mistakes that AI Overviews has already made in the short time since its launch, there is a broader problem at play here, a problem that will persist even if AI Overviews could be counted on to be reliably accurate, 100% of the time (and that, my friends, is a very big if). 

The problem, according to AI & linguistics researcher Dr. Emily Bender, has to do with recontextualization, where information is stripped of its context and mushed into the output of an automated system that people might mistakenly trust to be objective. 

• For example: In a recent post, Bender offered a scenario where a user makes a medical search on normal Google, receiving a series of links in return. These links will expose that user to information from known or knowable quantities; a Reddit forum, the Mayo Clinic, etc. 

• Thus, the user knows to take medical advice from the Reddit thread with a grain of salt; they also know (or can find out) that the Mayo Clinic is likely a good source of trustworthy medical information. 

But with AI Overviews, the context of these knowledge sources is gone, both limiting your ability to gain a broader understanding of a topic area and increasing your risk of trusting information that you otherwise wouldn’t. 

The glue pizza recipe came from a Reddit post … if a normal Google search led you to that link, you would have known instantly that it was a joke. 

OpenAI shuts down five global covert influence operations 

As OpenAI itself acknowledged, though ChatGPT might be intended for enterprises (or fun games where you try to get the chatbot to return 10 sentences ending in the word orange) the system is ripe for potential abuse. 

The company said in a Thursday blog post that, over the past three months, it has identified and “disrupted” five global covert influence operations that were using ChatGPT to spread content across social media. 

The details: 

• Two of these influence operations originated from Russian organizations that were active on Facebook, Twitter, Telegram and various websites. 

• One originated from the Chinese organization Spamoflauge, which posted across Twitter, Medium and Blogspot. 

• One originated from the Iranian organization the International Union of Virtual Media, which posted content to a website linked to a threat actor. 

• And the last originated from an Israeli company called Stoic, which posted content on Twitter, Instagram, Facebook and associated websites. 

OpenAI said that the bulk of the content focused on geopolitical events, adding that the organizations do not “appear to have benefited from meaningfully increased audience engagement or reach as a result of our services.”

My thoughts: I spoke with someone once who said something along the lines of ‘tech always creates a new problem, then ships solutions to that problem.’ I feel like I’ve been in this feedback loop of asking: ‘Okay, but is it worth it?’ I think it’s great that OpenAI shared this report, disrupted the activity and is using AI to detect when its AI has been used by threat actors. 

I just can’t stop imagining if they had never created this whack-a-mole problem to begin with.

Together with Zero To Mastery

Unlock the Secret to 2024's Hottest Skill!

Figuring out how to use AI tools and chatbots can 10x your career opportunities.

But it can be confusing and daunting to truly unlock their potential (trust us, we’ve been there).

That's why Zero To Mastery created their Prompt Engineering Bootcamp to teach you how to craft powerful prompts that will set you apart. The course will also teach you some of the fundamentals around how LLMs like ChatGPT work & how to create your own AI tools that will take your career to the next level. Plus, it's constantly being updated with the latest tactics (it's about to be updated with 3+ hours of additional lessons on Advanced Prompting Techniques that only experts know about).

This is your chance to stay ahead of the AI curve.

Join thousands of other students and master Prompt Engineering today.

Special Offer: For the next 48 hours, enjoy 15% off any plan with code DEEPVIEW, exclusively for The Deep View readers!

Deepfakes spiked more than 245% this year

A key element to the ethics of AI — and a key component of this ever-unfolding AI story — has revolved around deepfakes, that AI-powered capability to create an image or video of someone that is both super convincing and completely fake. 

This aspect of the tech has already been weaponized in ways that run the gamut from horrifying to disturbing, i.e. nonconsensual deepfake pornographic abuse, election interference and theft. 

The research: Verification platform Sumsub found that instances of worldwide deepfakes surged some 245% year-over-year in the first quarter of 2024; that number was 303% for the U.S. specifically. 

• Why it matters: This year is an enormous year for global elections, and the impact AI will have (and is already having) on these elections will likely be profoundly negative.

• Indeed, Sumsub found a “noticeable growth” of deepfake instances in countries that will hold elections this year. 

“Media platforms … need to make sure they are not inadvertently contributing to the spread of misinformation," Sumsub’s head of AI/ML Pavel Goldman-Kalaydin said in a statement. 

My thoughts: While content provenance technology does exist, such processes are imperfect and unlikely to be usable at scale this year. All I can do is urge you not to rely on social media as a source of actionable information. And always verify any piece of information from multiple trusted sources (following journalists you respect on Twitter is helpful here).

We can no longer trust anything we see online (as our next section regularly proves).

• Codara: Review your code and diagnose errors using AI.

• Brewed: Bring your web ideas to life and create websites and web components using AI.

• Huree: An AI-powered analytics dashboard.

Have cool resources or tools to share? Submit a tool or reach us by replying to this email (or DM us on Twitter).

*Indicates a sponsored link

SPONSOR THIS NEWSLETTER

The Deep View is currently one of the world’s fastest-growing newsletters, adding thousands of AI enthusiasts a week to our incredible family of over 200,000! Our readers work at top companies like Apple, Meta, OpenAI, Google, Microsoft and many more.

If you want to share your company or product with fellow AI enthusiasts before we’re fully booked, reserve an ad slot here.

One last thing👇

That's a wrap for now! We hope you enjoyed today’s newsletter :)

We appreciate your continued support! We'll catch you in the next edition 👋

-Ian Krietzberg, Editor-in-Chief, The Deep View