Accidental Tech Podcast
Accidental Tech Podcast

568: The Year of Romance

* Pre-show: Time-shifting * Dave Matthews Band Crew Q&A * Follow-up: * AI Agita * Apple’s paper regarding optimizing on-device LLM * Summary by luokai * Apple explores AI training deals with news publishers * New York Times sues OpenAI & Microsoft for copyright infringement * Hard cases make

Featured Speakers

Marco Arment, Casey Liss, John Siracusa Host

Topics Discussed

Episode Summary

Executive Summary: The episode centers on two major debates: the ethics and legality of copying/protecting digital media, and the likely 2024 Apple product cycle. The hosts argue that DRM and streaming limit legitimate users more than pirates, then pivot to AI training copyright fights, Apple’s licensing strategy, and a preview of Vision Pro, OLED iPads, Mac updates, iPhone 16, Apple Watch 10, and emerging LLM features across Apple platforms.

Main Topics: DRM, time-shifting, and digital media ownership: The hosts criticize modern DRM as frustrating legitimate users while doing little to stop piracy. They compare streaming restrictions to past VCR time-shifting rights and discuss screen recording as an imperfect workaround. AI training, copyright, and licensing data: A long discussion weighs whether LLM training on public or copyrighted material should be legal, with emphasis on the New York Times lawsuit, fair use, and Apple/Adobe's strategy of paying for licensed data instead of litigating. Apple Watch import ban and legal maneuvering: They cover the temporary U.S. Apple Watch sales pause caused by the Masimo patent dispute, Apple’s legal appeals, and how quickly the product resumed sales after the pause. 2024 Apple product roadmap and Vision Pro: The hosts preview likely Apple launches: Vision Pro, OLED iPads, M3 Ultra Macs, iPhone 16, Apple Watch 10, AirPods 4/Max, HomePod, Apple TV, and new OS-level AI features. Vision Pro market expectations and developer economics: They debate the device’s early constraints: high price, low supply, prescription/fit issues, gray-market challenges, and whether developers should invest in support despite a tiny initial user base. Apple's move toward on-device LLMs and OS integration: A paper on efficient flash-backed inference leads into speculation that Apple will roll out on-device LLM features for Siri, Photos, autocorrect, and other system services. Yearly themes and work planning: The conversation closes with a discussion of Cortex-style annual themes versus resolutions, with a focus on using themes to guide work and the year's priorities rather than setting brittle goals.

Key Arguments: Modern DRM mostly punishes paying customers; it does not meaningfully prevent piracy, which is why the hosts view it as both socially and technically harmful. Time-shifting logic from the VCR era should apply to paid streaming/video access, but DRM and the DMCA make it harder to exercise that right. LLMs trained on copyrighted material raise a compensation problem: if the model's value depends on others' work, some kind of licensing or market exchange should exist. Apple's approach—paying for licensed training data—is portrayed as a pragmatic hedge against uncertain copyright law. The New York Times case is framed as legally plausible on the basis of verbatim recitation and direct substitution, but the hosts disagree over whether training itself is infringement. Vision Pro will likely be supply-constrained, expensive, and highly dependent on fit, apps, and developer interest, so its first year should be judged as a slow start rather than a success/failure verdict. OLED iPads are positioned as a more immediately compelling 2024 purchase than Vision Pro because the use case is clearer and the upgrade is easier to justify. Apple Watch 10 may be the point at which Apple changes the case and possibly the strap mechanism to free space for battery or design changes. Apple's LLM strategy will likely start with constrained, useful features rather than a full Siri replacement, because current Siri is so poor that even imperfect LLM features could improve the experience. Sideloading in the EU could matter more as an experiment in market behavior than as an immediate consumer revolution, but it may still prove significant for apps, games, or alternative stores.

Data Points: Vision Pro starting price: $3,500 - Repeatedly referenced as the base price of Apple's headset Fish video access window: 48 hours - Paid concert streams are only available for a limited period after the show Fish show video price: about $40 - Cost to watch a single streamed concert Apple licensing talks: at least $50 million - Reported multi-year deals for news archives to train generative AI systems Apple Watch import ban date: December 26 - Date the Biden administration declined to intervene and the ban took effect Court pause timing: December 27, 11:55 a.m. - U.S. Court of Appeals temporarily paused the Apple Watch sales ban Sales resumption time: December 27, 5:21 p.m. - Apple resumed Apple Watch Series 9 and Ultra 2 sales after the pause Fish fan-service references: 30 years - The New Year show was described as weaving in references reaching back decades Dave Matthews Band archival recording: 46 tracks per day - Audio recording setup captured band inputs plus audience mics and other sources Dave Matthews Band archival storage: 138 CDRs per show - Each show's multitrack audio was reportedly copied for vault storage Vision Pro supply speculation: 1-2 million screens per year - Supply-chain rumor about Sony's ability to make the displays Vision Pro first-year volume speculation: maybe a million - Estimate for active devices in the first year if supply constraints hold Apple Watch current product milestone: Series 9 / Ultra 2 - Models temporarily pulled and then re-listed in the U.S.

Pivotal Quotes: "I think we already had several court cases about that, and the VCR was allowed to exist." — John/Speakers: Argument that time-shifting paid content should remain lawful despite DRM restrictions "I continue to think that it is that it should not be legal to train a large language model on content that you don't own because your large language model is worthless without that content." — John/Speakers: Core moral and economic argument about AI training data licensing "It is not a free for all, but it's also not so onerous that no one can ever sing covers anymore and it's also, you know, people do get paid." — Aaron Powell, Jr.: Proposal for a balanced licensing framework for AI training and content use

Implications: Expect more legal fights over AI training, a pragmatic push toward licensing deals, and a busy Apple year defined by Vision Pro curiosity, OLED iPads, and incremental but important platform AI features.

🔓 Sign Up for Unlimited Episode Search

About Accidental Tech Podcast

View all episodes from Accidental Tech Podcast