(Welcome to the Entertainment Strategy Guy, a newsletter on the entertainment industry and business strategy. I write a weekly Streaming Ratings Report and a bi-weekly strategy column, along with occasional deep dives into other topics, like today’s article. Please subscribe.)
It’s time for another edition of my “Four Horsemen of the Media-pocalypse” series, which I’m now aiming to get out quarterly. (You can find past editions here and here. And note: many of the links aren’t from the last quarter, since I try to group them by topic.)
I like writing this series because I want to check on theaters regularly (they’re sort of just surviving) and the cable bundle (yikes!) without getting caught up in the week-by-week twists and turns, which can drive you crazy. (Though if you want to follow those twists and turns, check out my “blogroll” to find out who you should be reading on the box office every week…) And I especially want to check in on piracy regularly, since I think that (barely discussed) issue really, really matters to the future of Hollywood.
Oh, and I have more thoughts on AI than I know what to do with.
So let’s dive right in.
Theaters – Who’s to Blame for Theaters’ Supply Problems?
Last go around, I shared The-Numbers recap of the year so far (and you should be subscribed to their monthly newsletter), but this go around, I wanted to share David Gross’ FranchiseRe’s update:
I particularly like this article because it highlights a major issue facing theaters today, which I’ve written about before: theaters have a supply problem. Q2 and Q3 have been doing okay; Q1 is sinking the US box office.
Who’s to blame for that supply problem? First culprit: the streamers. Just look at the weeks starting 22-Sep to 5-Oct. The streamers sent six major films, starring legitimate movie stars, to streaming, including…
- Prime Video’s Play Dirty, a heist film starring Mark Wahlberg, LaKeith Stanfield, and Rosa Salazar, and directed by Shane Black.
- Apple TV+’s The Lost Bus, a thriller starring Matthew McConaughey and America Ferrera.
- Netflix’s Steve, a prestige drama starring Cillian Murphy.
- Netflix’s Ruth & Boaz, a Tyler Perry Studios film on Netflix.
- Hulu’s The Man in my Basement, a 20th Century thriller starring Corey Hawkins and Willem Dafoe.
- Apple TV+’s All of You, a prestige sci-fi drama.
That’s a lot of movie stars in a lot of movies that aren’t going to theaters! (Five of those six films were produced by the tech companies’ streamers.) Yeah, some of those films are too small to make a dent in the box office, but many of them aren’t. Since you never know where the next surprise hit film may come from (see Sinners), losing these films to streaming represents real losses for the box office.
But corporate consolidation is just as big a threat to theaters. Last month, Hulu released two S-to-S films: Swiped and The Man in My Basement, both 20th Century films. Reminder: once Disney bought Fox, the two studios’ combined theatrical output declined. The big potential news is that Paramount might buy Warner Bros. and, yeah, they can promise to release thirty films a year, but that’s more than unlikely.
Both tech companies avoiding theaters and consolidation are killing movie theaters’ supply of films, but I only really see one of those issues (consolidation) garnering any larger outrage from the entertainment media.
Others are sensing a market opportunity here. As Matt Belloni noted on The Town podcast, with so many weekends “free” of studio films, other companies see open space on the theatrical calendar. Sure enough, Ted Hope shared a list of new film distributors that have popped up in the last year.
Cord-Cutting – Are the Cable Companies Finally Fighting Back?
Another quarter, another set of cable company declines. Unfortunately for those in the “Could the cable bundle hit a floor?” camp (which has included me), Q2 posted another decline of 1.6 million cable subscribers, according to MoffettNathanson. Sean McNulty of The Ankler’s “The Wakeup” newsletter estimated 1 million lost subs after he tracked all of the cable company’s public announcements, on par with past years. S&P estimated a 7% decline in 2024.
Importantly, the losses were “flat”, meaning they didn’t grow or shrink, a bad sign for the “we hit the floor” scenario. For cable to hit its bottom, it needs the losses to, well, stop, but the declines don’t appear to be slowing down…with one big exception.
Charter is now selling skinnier bundles and ad-tiers to the non-tech-company streamers…and it appears to be working, cutting their quarterly losses to just 80K, down from 400K the year before. (I’d love to subscribe to this bundle, but Charter isn’t available in Southern California right now.) Spectrum and DirecTV are attempting similar, but not as aggressive, tactics.
Frankly, strategically, cable companies should have adopted these tactics years ago.
AI – Two Big Pieces of News
There’ve been two big pieces of AI/LLM and Hollywood news since we last checked in on this potentially groundbreaking, world-changing technology:
- ChatGPT 5 came out. I’ve seen little to no improvement using it. If anything, ChatGPT seems worse now. I really can’t square the circle between my personal experience (ChatGPT 5 has the functionality of an assistant you’d fire within two weeks, and it’s crashing constantly) and the breathless, but often detail-free hype.
- OpenAI released Sora 2, an AI video program with even greater capabilities. And they launched their own app. And Meta launched Vibes, another AI video app.
Sora 2 represents a genuine improvement in video output from AI/Large Language Models, but there’s still two huge issues:
- How much do these video apps actually cost OpenAI? As even Sam Altman said, users are using it too much!
- When will Sora 2 be able to remember the prompts, scenes and characters? This is the biggest roadblock to AI making movies in Hollywood. Right now, Sora 2 can’t remember the content of its output, but can use reference images, though even these often shift over time. This limitation will keep these videos from replacing feature films. (For now.) (This article had some good notes on its limitations.)
We’ll see how much this field progresses in the years (months?) to come. Right now, I can’t get my text LLM to remember super basic prompts within a small convo, and it’s constantly crashing, so color me skeptical.
Piracy Part 1 – Enforcers Finally Get Some Wins Against Big Piracy Sites
When I write about piracy, I now have two different topics I have to discuss:
- Regular ol’ copyright infringement.
- LLMs pirating material to train on, training on copyrighted material, and outputting material that uses copyrighted characters/works.
Let’s start with old fashioned piracy. where there’s been the most action I’ve ever seen. ACE (Alliance for Creativity and Entertainment) shut down Rare Breed IPTV, Outer Limits IPTV, the world’s largest illegal sports streaming site, Streameast, which even Walt Hickey seemed to admit to using. The DoJ got a conviction against the founders of Jetflicks, the largest piracy case in US history. After these wins, in a tactic I appreciate, ACE and the MPA “scoop up” piracy sites’ domains to prevent them from coming back.
And on the political front, Senators are trying to pass the “Block BEARD” act to “allow copyright holders to ask a federal court to block piracy websites” and this legislation has bipartisan support.
This is all good news—which is nice before we get to the next section—but I’d argue that piracy on YouTube is still the real issue, which Google isn’t liable for due to section 230 protections. As an Adalytics report revealed this summer, thousands of copyrighted titles remain on the platform.
Okay, that’s the good news. On to the bad news.
Piracy Part 2 – The AI Companies Train on a Lot of Pirated Materials…And Can Keep Doing It
LLMs have been trained on copyrighted material, but with no recognition or compensation to the creators whose work was used to train LLMs. And in the last year, we’ve learned the scale of this training.
Alex Reisner at the Atlantic has written some great pieces on this issue, breaking news last November that multiple AI companies uploaded dialogue from Hollywood films (“53,000 other movies and 85,000 other TV episodes”) to train their AIs. Reisner had a follow-up story on how many pirated books Meta’s Llama gobbled up. As a result, according to a new report, Meta’s Llama can recall 42% of the first Harry Potter book, meaning it remembers and can repeat the text. The British Film Institute released a report citing that hundreds of thousands of works have been used to train AI. (Sadly, though, that same report recommends that creatives should license more content to AI companies.) 12.6 million pieces of fan fiction were used to train AI. Also, this summer, in a new lawsuit, a porn company accuses Meta of pirating and distributing porn on BitTorrent, for years, to get copyrighted material to train its AI program.
Legally, it doesn’t seem like LLMs will be stopped from training (or having been trained) on copyrighted data, but pirating the material the LLMs were trained on is a legitimate legal issue.
The authors suing Anthropic for copyright infringement settled their case, after a judge ruled that LLMs training on copyrighted material is fair use. But they were liable for pirating seven million books. (I’m no lawyer, but I’m surprised the writers didn’t just take them to court. Then again, class action lawsuits can often mostly benefit the lawyers bringing the cases.) And the authors suing Meta got their case thrown out, but the judge left the door open to future lawsuits.
Piracy Part 3 – The Studios Try to Stop AI Copyright Infringement
When OpenAI released Sora 2, users immediately started generating videos of copyrighted characters. And OpenAI said that every piece of copyright material had to be “opted out” of its program.
They’ve since walked that back a bit, offering more “granular control” to rights holders, probably because it’s just a matter of time before they get sued. Last June, Disney and Universal sued Midjourney for obvious copyright violations; now Warner Bros. joined that lawsuit. The three studios also sued Chinese AI company, Minimax, for training their work on stolen material. Universal added a warning to its films that they cannot be used for AI training.
But I would make a key distinction here. Many book publishers and studios have been hesitant to pursue legal action against AI companies for training on copyrighted material, because they hope to someday use the fruits of the AI companies’ labor, i.e., replace workers with AI. In the case of Disney and Universal’s lawsuit against Midjourney, they don’t like that their copyrighted characters are making someone else money.
Here Are My Five Recommendations To Handle This AI Piracy
Though I feel like most of what I suggest below has little chance of coming to pass—almost no politicians anywhere in the world have tried to meaningfully regulate LLM/AI—I thought I’d share my suggestions anyway. The point is the current “anything goes” era of AI just isn’t fair to content creators, nor good for society long term:
- First, there’s no reasonable, justifiable argument against total transparency for AI training. Perhaps if there was some “secret sauce” combination of training material, I could see it, but that’s not how LLMs work; their developers are trying to find and train with as much data as possible. (You don’t upload 50,000 screenplays with a discerning eye. Same goes for millions of pieces of fan fiction.) Thus, we should all know what those training materials are. The US should, tomorrow, require total AI training transparency laws, especially to ensure that none of the material was pirated.
- Second, Congress should pass a law protecting the images and likenesses of all people. This would especially apply to actors, but all Americans would be protected by this from deep fake porn or misinformation.
- Third, LLMs/AI companies shouldn’t have Section 230 protections. This seems obvious, but if the LLM is doing the creating, why wouldn’t the company be liable for what its product creates? It’s not users making these things, but the companies’ own technology.
- Fourth, LLMs/AI should compensate the creators of the materials they’ve trained on. This especially applies to journalists (which is easy to do looking at the sources that get cited), but I’d like to see this applied to other realms as well.
Some of these suggestions strike LLM proponents as counter-productive. They worry that regulating LLMs on copyrighted material will cause America to “fall behind” in the LLM/AI arms race, comparing it to America falling behind in the Cold War with nuclear weapons.
So first off, pause, and think about that analogy.
But second, what is the downside? First, let’s make a distinction between words/written material (I wouldn’t and don’t want my writing training LLMs, but I understand how that could potentially lead to super intelligent AI) and video…what really is the downside of America “falling behind” on AI video creation? Fewer and fewer Americans will become addicted to low-quality, low-effort, AI-generated social video slop. (And fewer Americans lose jobs.)
I’m not sure I see the national security threat on the video front.
