Good morning,
I made a mistake in Monday’s Article Hollywood on Strike: I stated that Netflix originally didn’t pay residuals, and now does; this is incorrect, as Netflix has always paid union-mandated residuals (more on streaming residuals in a moment). My mistake was conflating residuals with back-end bonuses for top talent, which Netflix did eschew at the beginning, but started paying later on. I apologize for the error.
Meanwhile, the most recent episode of Sharp Tech dives into both the Hollywood strike and Bob Iger and Disney. You can listen at the link or add the podcast to your podcast player using the links at the bottom of this email.
Finally, a reminder that there are no Stratechery Interviews for the next month; the next Update will be on Monday.
On to the Update:
Streaming Residuals and Spotify
I also made a second error when I lazily referred to broadcast residuals increasing based on a show’s popularity; that isn’t strictly true, as residuals are paid based on the number of times an episode or movie is shown, not on whatever rating it gets (of course the number of times an episode or movie is shown is a function of popularity). This model was roughly translated to streaming, wherein a show is paid residuals for being on a service (and thus “shown”); the payout is a function of the size of the streaming service’s subscriber base (the WGA has a breakdown of residual payouts on streaming services here; the SGA’s payouts function similarly).
The Washington Post had a useful article about residuals, explaining the strikers’ consternation with streaming services:
The advent of streaming has led to significantly lower residuals, according to SAG-AFTRA. During the heyday of cable TV, near-constant reruns of popular shows meant big residual paydays for writers and actors. But the meteoric rise of streaming services such as Netflix and Hulu has upended that reality, and creators say residuals are shrinking. Unlike cable reruns, residual payments for streaming services aren’t based on the number of times an episode or movie is viewed. Instead, they’re based on the number of subscribers the streaming service has, according to WGA. So writers and actors who work for platforms like Netflix make more than those who work for smaller streamers like Paramount Plus.
That means that regardless of whether a show is a flop or a cash cow, it makes the same amount in residuals. Writers and actors say that allows studios to profit off of their work without compensating them fairly for the success of a show. In short: Studios are making more off shows because of streaming, while writers and actors say they are making less. That, combined with inflation, has led to unfair compensation, SAG-AFTRA argues.
Reality is, as this article in Deadline explains, more complicated:
Last week, the WGA’s Negotiating Committee, in its latest appeal to members to approve a strike authorization, said, “Over the past decade, while our employers have increased their profits by tens of billions, they have embraced business practices that have slashed our compensation and residuals and undermined our working conditions.” But the WGA West’s latest annual report, released in June, notes that for the fiscal year ended March 31, 2022, total residuals collected by the guild in 2021 hit an “all-time high.” The guild’s annual reports also show that total residuals increased by 48.2% from 2011 to 2021 – from $333 million to $493.6 million.
Charles Slocum, assistant executive director at the WGA West, told Deadline this week that “total residuals are higher because many more projects are being made, and a lot more are in reuse.” The “slashing” of residuals, he said, “is on a per-program basis.” And the culprit is streaming, where half of all series writers now work.
“In streaming, the companies have not agreed to pay residuals at the same level as broadcast, or the same reward-for-success as they have traditionally paid in broadcast,” he said. “If you write for a streamer, you get two residuals payments – one for domestic streaming and one for foreign streaming. It’s a set amount of money. If it’s a big hit, you do not get paid more residuals in streaming, whereas in the broadcast model, you do because of its success. That’s the sense that residuals were slashed – they have not agreed to a success factor when a program is made for streaming.”
This is in fact a concrete example of how the inherent abundance of streaming, which is unmoored from the time limitations of linear TV, has had a dramatic effect: the studios can (correctly) argue that residual payouts have increased significantly, because they are considering the total; abundance, though, means more competition, which makes the payouts feel like less for individuals.
This dynamic may sound familiar:
This revenue tracker from the RIAA is one I have referenced before: note that the music industry, for all of its challenges in the 2000s, is now grossing more than it did in the heyday of the CD (before an adjustment for inflation, to be fair). The driver is the dramatic rise in streaming revenue, and yet the complaint we hear about streaming is that it is a bad deal for artists, who are paid fractions of a penny for every song. Once again the dichotomy is between aggregate revenue, which this chart shows, and per-artist revenue, which on an average basis is almost certainly lower because there are far more artists than ever before, because there is an abundance of music.
What is interesting to consider is if this comparison might go further and point to a potential resolution of this strike, at least in terms of the residuals question. Keep in mind that artists are not paid on a per-stream basis. Rather there is a pool of money that is set aside for artists (primarily via contracts with the music labels), and that pool is split up based on an artists’ share of total streams. To use a dramatically simplified example:
- If Spotify has $1,000,000 in revenue and,
- Labels get 60% of that revenue and,
- Taylor Swift has 1% of total Spotify streams then,
Swift’s payout would be $1,000,000 × 60% × 1% = $6,000. Notably this payout is only indirectly linked to the total number of streams of Swift’s music; what matters is Swift’s stream share of the absolute total.
As I’ve noted in the past this leads to a number of counterintuitive takeaways, including the fact that a lower per-stream payout is likely better for the artist because it may result from a larger absolute amount of revenue, which translates into larger artist payouts; the example in that Update is Spotify versus Apple, wherein Spotify pays less per stream because they have an advertising tier, but because that advertising tier is additive to their subscription base, the total amount paid to artists on an absolute basis would be higher even if the subscription tiers were the same size.
Now obviously writers and actors are not getting 60% of streaming revenue (nor, for that matter, is Swift), but if there were to be a reworking of streaming residuals in a way that reflected success — or the lack thereof — this is probably the only model that works, precisely because the pie itself is still a scarce resource (i.e. revenue). What is nice about this model is that it removes the impetus for streaming services to remove old shows that don’t drive subscriptions, but which still need to pay residuals; what may not be nice for a lot of writers and actors is the cold hard reality of competition on the Internet, where a few artists tend to make the lion share of the money, leaving scraps for the long tail that comprises everybody else.
Llama 2 Open-Sourced
From The Verge:
Meta announced it’s open-sourcing its large language model LLaMA 2, making it free for commercial and research use and going head-to-head with OpenAI’s free-to-use GPT-4, which powers tools like ChatGPT and Microsoft Bing. Meta announced the move as part of Microsoft’s Inspire event, noting its support for Azure and Windows and a “growing” partnership between the two companies. At the same time, Microsoft revealed more details about the AI tools built into its 360 platform and how much those will cost. Qualcomm also announced it is working with Meta to bring LLaMa to laptops, phones, and headsets starting from 2024 onward for AI-powered apps that work without relying on cloud services.
Speaking of errors, this excerpt includes a big one: LLaMA is now Llama, at least according to Meta’s introductory post; less fun, but much easier for me typing it out! This seems like a fair trade-off, because I very well may be typing Llama a lot — this is a very big deal! Start with that introductory post:
We believe an open approach is the right one for the development of today’s AI models, especially those in the generative space where the technology is rapidly advancing. By making AI models available openly, they can benefit everyone. Giving businesses, startups, entrepreneurs, and researchers access to tools developed at a scale that would be challenging to build themselves, backed by computing power they might not otherwise access, will open up a world of opportunities for them to experiment, innovate in exciting ways, and ultimately benefit from economically and socially.
And we believe it’s safer. Opening access to today’s AI models means a generation of developers and researchers can stress test them, identifying and solving problems fast, as a community. By seeing how these tools are used by others, our own teams can learn from them, improve those tools, and fix vulnerabilities.
I agree. When it comes to the innovation angle, look no further than the explosion that happened around Stable Diffusion in the image generation space, or the leaked LLaMA 1 weights: it was a matter of weeks before those models were not only being applied to a wide range of new use cases, and having their capabilities dramatically enhanced, but also were optimized to run anywhere. There is a very good case that we look back on this announcement as being on par with the launch of ChatGPT in terms of impact, and arguably a more important one when it comes to the actual utility of AI. Now a near state-of-the-art LLM can be built into anything and run anywhere, not simply constrained to one specific interface governed by one specific company.
This release will also hasten the commodification of models themselves: Llama 2 is about at the level of GPT 3.5 (except for coding, which it is not (yet) trained for), which is obviously not as good as GPT-4. GPT-4 access, though, both via ChatGPT and the OpenAI API is limited, relatively slow, and expensive; Llama 2 can be run anywhere, can be fully optimized, and is free. This not only makes far more potential use cases viable, it also provides significant pricing pressure on “better” models.
Llama’s License
I suspect that close readers of Stratechery are not surprised by this release; I wrote an Update in May highlighting how Meta was clearly experimenting with open sourcing its AI technology, and explained why it might work to Meta’s benefit to be more open than the rest of the industry:
Here, though, I go back to one of Zuckerberg’s comments on that earnings call:
I think that there’s an important distinction between the products we offer and a lot of the technical infrastructure, especially the software that we write to support that. And historically, whether it’s the Open Compute project that we’ve done or just open sourcing a lot of the infrastructure that we’ve built, we’ve historically open sourced a lot of that infrastructure, even though we haven’t open sourced the code for our core products or anything like that.
And the reason why I think why we do this is that unlike some of the other companies in the space, we’re not selling a cloud computing service where we try to keep the different software infrastructure that we’re building proprietary. For us, it’s way better if the industry standardizes on the basic tools that we’re using and therefore we can benefit from the improvements that others make and others’ use of those tools can, in some cases like Open Compute, drive down the costs of those things which make our business more efficient too. So I think to some degree we’re just playing a different game on the infrastructure than companies like Google or Microsoft or Amazon, and that creates different incentives for us.
Zuckerberg was specifically talking about cloud infrastructure software, but the same point applies to AI capabilities as well: Meta isn’t selling its capabilities; rather, it sells a canvas for users to put whatever content they desire, and to consume the content created by other users. It follows, then, that Meta ought to be fairly agnostic about how and where that content is created; by extension, if Meta were to open source its content creation models, the most obvious place where the content of those models would be published is on Meta platforms. To put it another way, Meta’s entire business is predicated on content being a commodity; making creation into a commodity as well simply provides more grist for the mill.
What is compelling about this reality, and the reason I latched onto Zuckerberg’s comments in that call, is that Meta is uniquely positioned to overcome all of the limitations of open source, from training to verification to RLHF to data quality, precisely because the company’s business model doesn’t depend on having the best models, but simply on the world having a lot of them.
That noted, not everyone gets access to Meta’s investment; note Section 2 of the Llama 2 license agreement:
2. Additional Commercial Terms. If, on the Llama 2 version release date, the monthly active users of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
Probably the closest company to that 700 million monthly active user (MAU) figure is Snap, which said it passed 750 million MAUs earlier this year. Obviously all of the other big consumer tech companies have more than 700 million MAUs, as well as other services like Telegram, which just surpassed 800 million MAUs.
You can certainly understand Meta’s thinking with this clause: I noted in AI and the Big Five that Apple and Amazon in particular were betting on open source for their AI stories; Apple could build it in to iOS, and Amazon could host it. If they want to leverage Meta’s investment to do that, though, they will need to come to an agreement with Meta, like Microsoft did. Microsoft will host Llama models, doubling down on their push to make Azure the obvious place to deploy not just AI workflows but also the data that goes into them, and presumably will pay Meta (or have some sort of revenue-sharing plan) for the privilege.
There are costs, though: Meta isn’t going to benefit from, say, Apple optimizing Llama for its operating systems and chips like it did for Stable Diffusion, and AWS is still the most important hosting platform that will host plenty of other models for all of those applications and datasets already in place. Moreover, while Llama is in many respects now an obvious choice for startups building products that incorporate AI, that choice may entail foreclosing a future acquisition from one of the giants, because they would then run afoul of Meta’s license.
I could certainly see this situation evolving over time: if Llama doesn’t get as much traction as Meta hopes then it may open it up to big companies too; if it gets massive traction then Apple and Amazon in particular may have no choice but to make a deal to make the most dominant model available on their platforms. “Most dominant model available” is a big “if”, of course, but open source is a pretty darn big accelerant, both for Llama specifically and AI generally.
This Update will be available as a podcast later today. To receive it in your podcast player, visit Stratechery.
The Stratechery Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.
Thanks for being a subscriber, and have a great day!