🤓 This Week in AI
PLUS: It Was Another Very BIG week In AI
Happy Friday!
It was a massive week in AI. While copyright battles and lawsuits took center stage, the real showstoppers were the debuts of tools that let you animate lifelike characters with just your phone and create customized AI voices that are unsettlingly human—all from text prompts.
There is so much to unpack, so I’m dedicating this entire edition to key updates so I don’t overwhelm you (or myself 😀).
Who am I kidding? I am overwhelmed. But it’s a good kind of overwhelm because I’m also excited. What a time to be alive!
Also, on a quick side note, more of you reached out to let me know that my recent newsletters have gone missing from your inboxes. Unfortunately, email filters can be tricky, and they might have ended up in your spam folder.
So, if you haven’t added me ( avisainewsletter@substack.com ) as a contact in your address book, or moved me to the primary tab in your inbox, or hit reply (and type ‘confirm’) to make sure you get my emails, do it now.
Here’s what you need to know this week (clickable links appear in orange in emails and underlined in the Substack app):
News Corp’s Dow Jones and the New York Post have sued AI startup Perplexity for "massive" copyright infringement, accusing the startup of copying and repurposing their news content into its AI summaries while bypassing links to original sources—without permission.
The New York Times recently sent Perplexity a cease-and-desist notice over similar issues. Perplexity has launched a revenue-sharing program with publishers like Time, but tensions between publishers and AI firms continue to rise as legal battles unfold.
In the meantime, Perplexity is looking to raise around $500 million to double its valuation to about $8 billion. The company gets about 15 million queries a day and makes $50M annually from subscriptions.
Penguin Random House is the first major publisher to add a clause to its copyright pages, explicitly banning the use of its books for AI training. While this signals a firm stance on protecting authors' work, the clause is not legally binding because copyright law already dictates how books can be used, and AI training may still fall under exceptions like fair use.
And in more copyright drama this week…
Blade Runner 2049 producer Alcon Entertainment is suing Tesla and Warner Bros. Discovery, accusing Tesla of using AI tools to generate unlicensed images resembling scenes from the film for its robotaxi launch event.
And now, for another twist…
A former researcher at OpenAI, the first to publicly criticize the company’s methods, says the use of copyrighted data to train models like ChatGPT violates the law. He argues that AI systems are undermining creators, businesses, and the internet ecosystem by generating content that competes directly with the sources they were trained on.
Anthropic has released an upgraded version of their top-tier model, Sonnet 3.5, featuring enhanced reasoning, visual processing, and coding abilities.
And they’ve worked on reducing all the unnecessary apologies 🤭 and have instructed the model to be more direct.
Interestingly, they didn’t rename it—you’ll only know you're using the updated one if you look for the "New" label in parentheses (see below). It’s available to all free and paid users, though free users will hit usage caps.
I played around with it for a few hours this week, but need to dive deeper giving you the scoop.
The company also introduced a data analysis tool, which allows you to process and analyze data, as well as run complex data visualizations and code.
Ok, this next one is really exciting.
Runway just released Act-One, a new tool that allows users to turn simple video footage – even from a smartphone – into lifelike character performances and animations.
The tool accurately captures details like micro-expressions, nuanced pacing, and eye-lines, allowing actors to be turned both animated and realistic characters without the need for expensive motion-capture gear.
A single actor, using only one camera, can now perform multiple characters and generate distinct outputs for each role.
To ensure responsible use, Act-One comes with safety measures like blocking unauthorized attempts to generate content featuring public figures and verifying voice usage rights.
The results are truly impressive. Runway’s rivals will likely introduce similar tech soon, expanding creative possibilities while also challenging long-held assumptions about creativity and artistic value.
You can check out a few examples below.
And a little weekly reminder (because it’s worth repeating):
These models are the WORST they’ll ever be.
Top labs are holding back even more advanced models (OpenAI sat on Advanced Voice Mode for over two years).
If you're building your AI strategy, seek out experts who’ve been fully immersed in this fast-evolving space for the past two years and have hands-on experience—definitely not the big consulting firms.
They might not have the “big firm” labels, but their practical knowledge far exceeds traditional consultancies, who, despite charging hefty fees are still playing catch up and making costly mistakes.
Plan for what’s coming, not what’s already here. Things are moving real fast, and anticipating the next wave of capabilities will give you a big competitive edge.
ElevenLabs, a leader in AI voice generation and cloning, has launched Voice Design, a tool that lets users create customizable voices from simple text descriptions. Creators can tailor voices to specific tones, ages, and accents, offering tons of flexibility.
Whether it’s for audiobook narration, game characters, virtual assistants, film and TV production, or educational content like e-learning, there are countless ways for creators to bring AI-generated voices to life.
You can check it out below and find a voice design prompt guide here.
If you’ve been reading my newsletter for a while, you know I’m skeptical of most AI startups, especially those in the creative and content creation areas (image, video, audio and music generators) because they’re far too many of them and they all face a significant challenge: commoditization.
This leads to two key issues: decreasing differentiation and race-to-the-bottom pricing.
So as the initial excitement wears off, many of these products are left competing on cost alone, which significantly reduces their defensibility.
Unless, of course, they have the broad distribution power of companies like Google, allowing them to scale faster and reach wider markets, even in a crowded field.
So, there’s a good chance MOST of these companies (even the well-funded ones that seem promising) won’t be around within the next year or two.
What sets ElevenLabs apart—and why I’m particularly impressed by their approach—is their clear focus on strategic differentiation in a space where many companies are still riding the hype wave.
Unlike startups that focus solely on the technology, ElevenLabs has built an ecosystem focused on:
1️⃣ Practical, high-value use cases that solve real-world business problems
2️⃣ User-friendly and intuitive interface that allows creators and businesses to use its products without technical expertise.
3️⃣ Meaningful strategic partnerships across various industries, using AI voice technology to power media content like podcasts and news audio, simplify customer service, and enhance healthcare access by improving multilingual communication and appointment scheduling. I’m obsessed with the breadth of their partnerships.
It’s one of the few AI companies I’d genuinely be excited to collaborate/work with. 😉
To the many media and tech leaders among my readers: if you’re evaluating potential partners and need help cutting through the noise (and there is so much), hit me up!
The White House is ordering the Pentagon and intelligence agencies to speed up their AI adoption to stay ahead of rivals like China, while protecting democratic values. A new national security memo sets rules for ethical AI use, banning AI from tracking free speech or bypassing nuclear controls.
Microsoft is launching AI agents or ‘virtual employees’ that can handle tasks like customer requests and identifying sales leads without constant human supervision. Companies can build their own agents or choose from 10 pre-built bots specializing in areas like sales, service, finance, supply chain, and more to automate repetitive work.
Schools’ growing reliance on AI detectors is leading to false accusations of cheating, especially for neurodivergent and ESL (English as a Second Language) students, with serious academic and emotional consequences.
As AI becomes central to the future of work, finding ways to integrate it into education is essential to preparing students with the skills they’ll need to succeed.
Canva has a shiny new and more powerful text-to-image generator called “Dream Lab”, a tool that was born out of its recent acquisition of AI startup, Leanardo.ai (my absolute favorite free image generation tool).
Now, I love me some Canva and have been a fan of it for years, ever since I started using it to promote the events I used to host on Clubhouse (if you know, you know).
But Canva’s AI features, especially the image generation ones, have been meh at best and unusable at worst. So, I’m really excited for Leonardo’s integration.
Free users get 20 lifetime generations, and paid accounts get 500 a month, which is a solid deal.
The mother of a 14-year-old boy who died by suicide after developing an emotional connection with a chatbot on Character.AI is suing the company. With over 20 million users, the AI companionship app is now facing questions about whether it properly protects vulnerable teens amid growing concerns about the mental health effects of AI companions.
👀 A Polish radio station has sparked controversy by replacing journalists with AI-generated presenters. The station claims its AI avatars will appeal to younger listeners by covering cultural and social issues, but critics argue the move sets a dangerous precedent for media jobs. Over 15,000 have signed a petition opposing the move.
UCLA researchers have developed an AI system that can analyze 3D medical scans—like MRIs and ultrasounds—with the same accuracy as medical experts, but much faster. It works with various types of scans and requires fewer training samples, which could speed up diagnoses, cut costs, and improve patient care in real-world clinics.
That's all for this week.
I’ll see you next Friday. Thoughts, feedback and questions are welcome and much appreciated. Shoot me a note at avi@joinsavvyavi.com.
Stay curious,
Avi
💙💙💙 P.S. A huge thank you to my paid subscribers and those of you who share this newsletter with curious friends and coworkers. It takes me about 8+ hours each week to curate, simplify the complex, and write this newsletter. So, your support means the world to me, as it helps me make this process sustainable.