hi bud buds,
last week, openai launched sora 2, their most advanced video model yet, with far more accurate physics. for example, in earlier versions of sora, a basketball might “teleport” into a hoop. now, shots bounce off the rim, lips sync with dialogue, and sound blends naturally into the scene.
alongside sora 2, openai launched an invite-only sora app for ios. it looks like tiktok (a scrollable video feed), but every clip is ai-generated. the standout feature, “cameos,” lets you use sora 2 to insert yourself or your friends into any video on the app.
a few key takeaways below:
innovative privacy and safety controls
obviously, this platform raises major concerns. what stops people from making harmful or sexual deepfakes of public figures or kids? imo, openai has rolled out an incredibly thoughtful set of safeguards:
opt-in only: if you haven’t opted in, your likeness (or “cameo”) can’t be used, including for public figures
identity verification: cameo creation requires filming a short video and submitting voice samples to prevent impersonation
permissions: you decide who can use your cameo (friends, mutuals, or public) and can change or revoke access anytime
co-ownership: cameo owners automatically co-own any content featuring their likeness, and can delete it at will
teen protection: parents can manage cameo permissions and limit discovery or usage by adults
content filtering: openai claims explicit or extreme content is impossible to generate on the platform
sora is “built for creativity > consumption“… for now…
in their launch materials, openai repeatedly emphasized sora is built for “creativity, not consumption.” for users under 18, infinite scroll is disabled by default, and the feed pauses after a few videos. even adults get nudges if they’ve been scrolling too long. it’s a noble idea, but i’d bet $1,000 they quietly walk this back once they see how much it throttles engagement. imo, most people open social apps to veg out, not make stuff.
intuitive product roadmap
coming next: an android app, an api (with advanced editing capabilities and integrations with video-editing tools), and storyboard controls (for creators to design scenes shot-by-shot)
launch controversy: copyright chaos
unsurprisingly, the sora app went viral after launch for being flooded with videos featuring copyrighted characters doing absurd things (like pikachu stealing from cvs). in response, sam altman quickly announced two changes:
copyright holders will have opt-in control over use of their characters’ “cameos”
openai will share revenue with those rights holders
honestly, anything openai launches now will draw scrutiny. to me, the bigger signal here: sora 2 already has massive engagement, real product-market fit, and a potential new revenue stream for ip owners.
and yet, i still have lingering doubts
a few weeks ago, i wrote that i’m skeptical about mass-market adoption of ai-generated video, since most consumers don’t have real use cases for it. once openai enforces stricter copyright controls, engagement could plummet even further. the overlap between (1) what users want to see and (2) what rightsholders will allow feels incredibly narrow. people don’t want “authorized” pikachu; they want pikachu shoplifting from cvs.
but who knows… it’s exciting times nevertheless, and i would love to be proved wrong. i’m still very eager to try sora myself; someone please invite ya boy.
have a great week, y’all. i love you.
- dj
last week’s biggest product releases
big news! huge!
openai launched chatgpt parental safety controls, including alerts for self-harm conversations, quiet hours, disabling voice mode and memory, removing image generation, and opting out of model training
nothing released playground, an ai tool for creating mini apps from text prompts and deploying them directly to the app store. i think apple will soon streamline its own pipeline from vibe coding → app store too
google launched gemini for home, an llm-powered upgrade to its smart speakers, displays, and doorbells that can handle complex requests like asking “did something eat my plants?”, then checking nest cam footage for bunnies 🐰
perplexity launched comet ai browser, a free ai web browser after months in private beta. i’m excited to try this after not loving arc’s ai browser (dia)
personal faves
amazon released add to today’s delivery, letting users add items to upcoming deliveries with one tap. i always forget to bundle my orders and end up scheduling two (and feeling guilty), so i love the sustainability of this. plus, it’s better for amazon’s margins. win-win.
meta released fan challenges and custom badges, helping creators engage fans through challenges and customizable top fan badges. fans can join by posting reels or photos, with submissions ranked by reactions on a leaderboard
whoop launched blood-testing services to complement its tracking suite for physical activity, sleep, respiratory rate, and blood pressure
amazon released familiar faces, enabling ring cameras to recognize visitors and find lost pets by connecting with other ring cameras in the neighborhood. bring those lost pups home!!!
nice job guys
salesforce released agentforce vibes, an ai-powered dev tool for vibe coding. it lets developers describe tasks in natural language and auto-generate code using their org’s own codebase and policies (great for enterprises)
amazon launched new kindle scribe tablets, with larger screens, improved note-taking, ai search, and ai summaries
amazon released amazon grocery, a private-label brand of grocery items where most products are < $5
doordash is testing dot in phoenix, an autonomous robot for food delivery. dot navigates roads and sidewalks, carries up to 30 pounds, and uses ai for route optimization
amazon released alexa+ for fire tv, allowing for voice-based recommendations, scene search, and live sports trivia
microsoft launched xbox cloud gaming, officially removing its beta tag after five years
paypal released honey integrations with ai chatbots, making its coupon-code browser extension accessible to ai agents
opera launched neon, an ai browser with ifttt-style task automation and “cards” for building repeatable prompts
spotify released exclude tracks from taste profile, letting users remove certain songs (like kids’ music for parents) from influencing recommendations
doordash is testing “going out”, a loyalty program where dashpass members earn in-store rewards at local restaurants
other llm news
anthropic launched claude sonnet 4.5, its flagship coding model for complex agents, improving reasoning and long-task performance
thinking machines lab (mira murati’s ai company) released tinker, an api that simplifies fine-tuning and post-training for llms
google released jules tools, a command-line interface and public api for its jules coding agent, enabling scoped task runs directly from terminals
deepseek released v3.2-exp, an experimental model cutting long-context api costs by ~50%, aimed at developers running massive prompts at scale
huawei introduced sinq, a new open-source quantization method for llms that reduces memory use without hurting quality
idk tbh
paypal launched venmo-paypal native integration, finally letting users send money between venmo and paypal. honestly it feels embarrassing this took over a decade since their 2013 acquisition
snapchat released snapchat paid storage plans for memories, capping free storage at 5gb. interesting to see social apps like snapchat following twitter’s lead in charging for basic features. i wonder if this weakens network effects, and if they’d be better off optimizing ads instead
adobe released premiere for iphones, bringing its flagship editing software to ios. imo, this feels like trying to do too much, and maybe even brand-dilutive. mobile editing is a totally different persona (teen creators on tiktok vs. professional editors for tv shows). honestly, adobe should just acquire a mobile-native editor like capcut and keep them separate
doordash launched a creator program for short-form videos, paying users to post short-form videos of their meals… idk, not the worst idea but feels kind of outside their wheelhouse
hinge is softening its approach to account banning, where if only one part of your profile breaks a rule, they’ll now flag and ask you to fix it instead of banning you entirely. that should’ve just been the case all along?
that’s it for this week; thanks for reading. if you enjoyed this issue, please consider sharing it with a friend so i can get rich and fly to jamaica for a month with my wife and eat curry goat
love, dj