š¤ This Week in AI
PLUS: Authors vs. AI, High-Stakes Lawsuits & AI Taps Hollywood Scripts Without Permission
Happy Friday!
Would you sell your lifeās work for $2,500? HarperCollins hopes some authors will, as their AI deal sparks outrage over undervaluing creative work.
Meanwhile, AI training practices are under fire, with Hollywood scripts like The Godfather and The Simpsons used without consent. And an AI-powered search tool is reshaping how we shop online.
With a free new model shaking things up and ChatGPTās recent upgrades, this weekās stories offer plenty to exploreāand try for yourself.
Just a quick note: Thanksgiving is next week, so Iām planning to take the week off to recharge. But with ChatGPTās 2nd birthday on 11/30 and rumored updates from OpenAI and Google, thereās always a chance of major drops. If so, Iāll either send a quick update or save it for the following week. See you Friday, December 6th.
Ok, letās dive in. Hereās what you need to know about AI this week (clickable links appear in orange in emails and underlined in the Substack app):
HarperCollins has made a deal with Microsoft to use select nonfiction books for AI training. Authors can opt in for a $2,500 non-negotiable payment (for a three-year license), or choose not to participate.Ā
Honestly, whoās opting in for $2,500?
š Harper Collinsā talking points:
āPart of our role is to present authors with opportunities for their consideration while simultaneously protecting the underlying value of their works and our shared revenue and royalty streams. This agreement, with its limited scope and clear guardrails around model output that respects authorās rights, does that.ā
This is a textbook brand misstep. It not only devalues the HarperCollins brand but also destroys trust and alienates its most crucial partnersāthe authorsāall at once.
AI-powered search engine Perplexity launches shopping, integrating e-commerce into its search results just in time for black Friday.
The tool offers unbiased product recommendations and the ability to search with images, while Pro users ($20/month) get exclusive access to one-click checkout and free shipping.
The company is also rolling out a new merchant program to give sellers āinsights into search and shopping trends that surface their products.ā
Iād say Perplexity is taking on Amazon, but Jeff Bezos, one of its biggest backers, might have other plans.
A recent investigation from The Atlantic reveals that dialogue from over 53,000 movies and 85,000 TV episodesāincluding iconic works like The Godfather and The Simpsonsāhas been used to train AI systems by companies like Apple, Meta, and Anthropic without permission.
The data set āincludes writing from every film nominated for Best Picture from 1950 to 2016, at least 616 episodes ofĀ The Simpsons, 170 episodes ofĀ Seinfeld, 45 episodes ofĀ Twin Peaks, and every episode ofĀ The Wire,Ā The Sopranos, andĀ Breaking Bad. It even includes prewritten āliveā dialogue from Golden Globes and Academy Awards broadcasts.ā
It has circulated among AI developers since 2020 as part of the Pile, a collection of data sets for training generative AI. āThe Pile also includes text fromĀ books, patent applications, online discussions, philosophical papers,Ā YouTube-video subtitles, and more.ā
You can search through the data set by using the tool embedded within the piece by the Atlantic.
š ChatGPTās underlying model, GPT-4o, just got an upgrade. Itās now better at creative writingāmore natural, engaging, and tailored. Itās also better at working with uploaded files to provide more insightful and complete responses.
For the many of you whoāve attended my workshops, nowās the time to dig out your 5-page cheat sheet and draft some high-quality prompts. You wonāt be disappointed. Just make sure to start a new chat to ensure youāre using the updated version.
š Ā OpenAI is also rolling out Advanced Voice Mode on desktop to all paid users, letting you interact with ChatGPT via voice. I already love using this feature on the mobile app when Iām out and about, but having it on desktop means I can give long, detailed instructions or context without typing it all outāmaking things much faster when Iām working at my computer.
French AI startup Mistral has launched Pixtral Large, the companyās most advanced model yet with multimodal capabilities. Their rival to ChatGPT and Claude, Le Chat, now comes with web access with citations, PDF analysis (including visuals like charts and graphs), and image generation powered by Flux Pro 1.1, a leading AI image model.
Oh, and theyāve made it all free (well, for now anyway). So, try it while you can (especially for top notch images) by signing up here.
Suno, one of the leading AI music companies just dropped its most powerful model yet (v4), with better audio quality, sharper lyrics and more dynamic song structure. Ā
The app can generate full songsāvocals includedāfrom nothing but a text prompt and now allows users to upload partial tracks, vocal recordings, or loops for creative collaboration.
Since its launch last year, Suno has made rapid improvements, extending song durations, adding community features, and even allowing users to create custom āsingers,ā all while building a loyal user base.
While its progress is impressive, Sunoāalong with competitor Udioā is currently facing lawsuits from major music labels accusing it of training its AI on copyrighted music without permission. Ā
But Suno seems unfazed. Co-founder Mikey Shulman calls the lawsuits āa bump in the road,ā adding, āWeāre excited about building the future of music.ā
I checked out the demos of the new model on the app and theyāre pretty remarkable.
For now, the new model is available only to Pro ($10/month) and Premier ($30/month) subscribers.
This Rolling Stone exclusive dives deeper into Sunoās ambitious vision and is worth the read.
Promise, a generative AI studio startup, has secured seed funding from Peter Cherninās North Road Co. and Andreessen Horowitz. It will produce original films, series, and new formats in collaboration with Gen AI artists and storytellers, using cutting-edge tools to streamline production workflows.
The company is led by experienced founders, including Fullscreen former founder George Strompolos, ex-YouTube content executive Jamie Byrne, and AI filmmaker Dave Clark.
āAndreessen Horowitz gets pitched on AI studios regularly, but often they are run by either Hollywood executives or tech executives, said Andrew Chen, a partner at the venture-capital firm. Promiseās co-founders span tech, distribution and entertainment, an appealing combination, he said. "
Given their track record, domain expertise and talent relationships, the outlook isādare I sayāfull of promise. Iām excited to see what they create.
Looks like theyāre already hiring, too!Ā
Elon Musk, who was once a co-founder of OpenAI, has escalated his legal battle with the company, adding Microsoft as a defendant in a lawsuit alleging antitrust violations.
Musk accuses the companies of monopolizing the generative AI market through exclusive licensing deals, requiring investors to avoid competitors, and prioritizing profits over the public good, reflecting his broader criticism of OpenAIās ongoing shift from its nonprofit roots to a for-profit model.
These court filings exposed internal emails from OpenAIās early years, revealing intense debates over control, talent retention, and fears ofĀ AIĀ dominance.
Reading through these exchanges is eye opening, offering insights into the personalities and competitive dynamics behind the worldās top AI labs.
Iāll share just the first one hereāan email from now-OpenAI CEO Sam Altman to Musk, as it sets the tone for the conversations.
And, of course, because this is the internet, someone has already compiled all the emails in one place for easy reading.
Subject:Ā question
Sam AltmanĀ to Elon Musk - May 25, 2015 9:10 PM
Been thinking a lot about whether it's possible to stop humanity from developing AI.
I think the answer is almost definitely not.
If it's going to happen anyway, it seems like it would be good for someone other than Google to do it first.
Any thoughts on whether it would be good for YC to start a Manhattan Project for AI? My sense is we could get many of the top ~50 to work on it, and we could structure it so that the tech belongs to the world via some sort of nonprofit but the people working on it get startup-like compensation if it works. Obviously we'd comply with/aggressively support all regulation.
Sam
For context, in case you read the exchanges, hereās a quick rundown on whoās who in these emails:
Sam Altman: OpenAIās current CEO
Ilya Sutskever: OpenAIās former Chief Scientist, who played a key role in the temporary ousting of Sam Altman during last yearās leadership crisis.
Greg Brockman: OpenAIās co-founder and its former President and Chief Technology Officer, who fully supported Altman during the leadership dispute.
DeepMind: An AI research company acquired by Google in 2014 for $400 million, known for groundbreaking work in AI. In 2015, it was led by its co-founder and CEO, Demis Hassabis, a renowned figure in AI research.
Iām also just going to drop this here for funsies.
This means that in the short term, Elon will likely get his way on a few thingsā¦
And if or (who are we kidding?) when things get bad between them, it will get real bad, because Elon will know too much.
Andā¦
OpenAI CEO Sam Altman will co-chair San Francisco mayor-elect Daniel Lurieās transition team as part of the administrationās efforts to strengthen ties with the tech industry.
Scammers are using AI to create fake albums under the names of established artists and covers of popular songsāuploading them to Spotify to boost streams from unsuspecting listeners. These schemes steal royalties, confuse fans, damage artistsā reputations, and undermine trust in the streaming platform.
A new small-scale study found that ChatGPT outperformed 50 doctors at diagnosing medical cases. The test involved real but unpublished case histories, ensuring ChatGPT hadnāt been pre-trained on them.
The chatbot scored 90%, compared to 76% for doctors using the tool, and 74% for those relying only on common resources.
These findings highlight three key issues:
Human Bias: Many doctors stuck to their initial diagnoses, even when the chatbot provided better-supported alternatives.
Not Using ChatGPT to Its Potential / Lack of Training: Most doctors used ChatGPT for simple queries or targeted questions but didnāt take advantage of its ability to provide deep insights on complex cases or analyze entire case histories.
Skepticism Toward AI: Many doctors hesitated to trust ChatGPTās reasoning and answers, even when it challenged their assumptions with better evidence
This study is a wake-up call for better AI trainingānot just for medical professionals, but across all industries.
Even frequent users of tools like ChatGPT or Claude often miss out on their full potential. Without training, itās hard to learn about what these models can actually doāand get the most out of them.
Iāve said it before, and Iāll say it again: Whether youāre just starting out or leading at the top, if your company doesnāt provide comprehensive AI training, take the initiative to find a trusted source.
Understanding how these models workātheir strengths, limitations, and how to get the best out of themāwill help you get better results while minimizing risks. This knowledge is essential for the future of work (and life).ā
Coca-Colaās AI-generated holiday adāa remake of its 1995 classicāhas drawn mixed reactions. Critics call it "soulless" and "creepy," claiming it reduces emotional impact, while others praise its nostalgia.
But were their older ads in this category ever emotionally connective?
Interestingly, research shows viewers unaware of AIās role responded more favorably, reflecting how its use may shape consumer reception and perceptions of authenticity.
Ultimately, it was a safe experiment to gauge overall reactions, and now they have themā¦
ESPN is testing an AI avatar called FACTS on its SEC Nation show, to deliver sports insights like the Football Power Index (FPI), team stats, and schedules.
And lastly,
Fortune's deep dive unpacks how Mark Zuckerberg rebuilt Meta around AI. From open-sourcing Llama to reshaping Metaās strategy, this piece highlights the tech gamble that could redefine the industry.
In case you missed last weekās post, you can find it š:
That's all for this week.
Thoughts, feedback and questions are welcome and much appreciated. Shoot me a note at avi@joinsavvyavi.com.
Stay curious,
Avi
ššš 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.