☕️ Elon Musk & The Controversial Use of Political AI Deepfakes

And we're back! After a short hiatus, AI Tangle is back to delivering what's new in AI - starting with Elon Musk's controversial deepfake video post about Vice President Kamala Harris. Other key takeaways of the week include:

• Apple pushes back Apple Intelligence's release date to a few weeks after iOS 18

• Meta debuts a tool to allow US creators on Instagram to create AI versions of themselves

• Biden-Harris Administration endorses open source models like Meta's Llama in a recent report by the US Commerce Department

Join us at AI Tangle as we untangle this week's happenings in AI!

Elon Musk under fire as he is accused of spreading "manipulated lies" by Harris' campaign

Elon Musk recently sparked controversy when he reposted a doctored video of Vice President Kamala Harris on his X/Twitter platform. The video clip, viewed 131 million times as of writing, features a fake Harris voiceover making inflammatory statements, following which Harris's presidential campaign accused Musk of spreading "manipulated lies." Many see it as especially concerning when considering Elon's endorsement of Donald Trump, Kamala Harris' direct opponent in the presidential race.

​

Let's elaborate a bit

Reposted on the 26th of July by Elon Musk, the Harris deepfake voiceover says lines such as "I was selected because I am the ultimate diversity hire" and that anyone who criticizes her is "both sexist and racist." The original video's author, posted by the @MrReaganUSA account, properly labels the clip as a parody, while Elon's does not. Amy Klobuchar, a Democratic senator, argues that this, in turn, violates the platform's synthetic and manipulated media policy. Hany Farid, professor at the University of California, Berkeley, added that the Harris video showed the power of generative AI and deepfakes and that we "will be seeing more and more of this type of nonsense."

Apple delays Apple Intelligence for iOS devices by "a few weeks" following stability concerns

Apple Intelligence, the company's first deep dive into AI-based enhancements announced during its Worldwide Developer Conference (WWDC) in June, has been delayed to launch in October. Apple Intelligence, and the suite of AI-powered features and enhancements, was set to release along with iOS 18 and iPadOS 18 in September, but Bloomberg's Mark Gurman noted that users will have to wait "a few more weeks." This means that Apple Intelligence should be available starting from iOS 18.1, which software developers are expected to get their hands on a beta release of in just a few days.

​

Meta debuts a tool to allow Instagram users in the US to create AI clones of themselves

Meta recently introduced AI Studio, allowing creators on Instagram in the US to make customized AI chatbots of themselves that interact with followers, answer common questions, and respond to comments on their behalf. It can be argued that AI Studio is Meta's response to compete with similar chatbot startups, such as character.ai and Replika, by enabling the creation of new AI characters. Due to the tricky nature of the concept, Meta has assured that such AI profiles will be clearly labeled everywhere, along with providing a guideline handbook for creators that goes more in-depth about the process.

​

The latest report from the US Commerce Department endorses open source models like Meta's Llama

The US Commerce Department issued a report endorsing "open-weight" generative AI models like Meta's Llama 3.1, emphasizing their availability to small companies, researchers, and nonprofits. However, the report recommends that the government find additional ways of monitoring these models for potential risks. While open models promote innovation and competition, the NTIA calls for more active risk assessment and restrictions on model availability when needed, which is in line with the current Biden-Harris Administration to strike a balance between AI potential and AI safety.

​

NIST released a tool called Dioptra to help assess and test AI models' security

The National Institute of Standards and Technology (NIST) recently re-released Dioptra, an open-source web-based tool from 2022 designed to determine the impact of malicious attacks on AI systems that affect their performance, such as training data "poisoning." Dioptra helps companies and users assess AI risks by benchmarking and researching models, exposing them to simulated threats. While Dioptra works as intended for locally downloaded models, it currently doesn't support models like OpenAI's GPT-4o that are gated behind an API.

​

Canva acquires AI image generation startup Leonardo.ai, setting its sights on Adobe's market share

The Australian-based online design platform Canva is acquiring AI image generation startup Leonardo.ai, making it the company's second acquisition this year as it tries to compete with Adobe. The deal, which took a month to finalize, allows Canva to integrate Leonardo.ai's video and image generation models into its own products. Valued at $26 billion in 2024, Canva aims to lure away Adobe's enterprise customers and accelerate revenue growth with its ever-growing suite of AI tools as Leonardo.ai joins Canva's list of acquisitions, which includes the likes of Affinity, Kaleido.ai, Pexels, and Pixabay.

​

AMD's "Strix Point" series' Ryzen AI 9 HX 370 undergoes its first tests and benchmarks with impressive takeaways

The AMD Ryzen AI 9 HX 370 processor, part of AMD's latest Ryzen AI 300 series nicknamed "Strix Point," was recently tested in the Asus Zenbook S 16 (UM5606) laptop. Allegedly built from the ground up for AI workloads, AMD certainly wasn't pulling any punches with its statements. Early testing shows the HX 370 processor topping AI workload benchmarks, like Geekbench, and casually excelling elsewhere, such as graphics and productivity-related tests.

​Mem.ai - Mem organizes the information you save to it, making the process of remembering important meetings, ideas, and articles easy with instant access to your personal knowledge. Level up your productivity and creativity with a 7-day free trial at get.mem.ai.*

​Humanize - Humanize is a text transformer tool to make AI-generated content sound more human-like by combining sentiment analysis and contextual intelligence.

​Ariglad - Enlist Ariglad's AI-powered knowledge base to automatically update and create articles by analyzing support tickets to automate your company's help center, ensuring your information is always relevant and up-to-date.

​Blaze - Blaze uses AI and automation to scan millions of online signals to help companies target the right potential customers as soon as they show interest and generate leads - make marketing easy with Blaze.

* indicates a sponsored tool

Google's Latest AI Commercial Misses The Mark (3-min read)

Google's latest commercial for Gemini AI raises questions about authenticity as it shows a father using AI to help his daughter write a fan letter to Sydney McLaughlin-Levrone. Understandably, critics argue that it completely undermines the personal touch of such letters and that, out of everything, Google shouldn't be trying to push AI-written fan letters.

A Sliver Away From an Olympian Gold (17-min watch)

The combined efforts of Google DeepMind's latest breakthrough models, AlphaProof and the upgraded AlphaGeometry 2, made waves as AI snagged itself a (fictional) silver medal at the International Mathematical Olympiad (IMO) - but how did they do it?