The line between the artificial and the authentic is narrowing at unnerving speed. OpenAI’s Sora video app, barely out of the gate, is already teeming with deepfakes. Politicians appear to spout words they never uttered, cartoon characters stride into the uncanny valley, and copyrighted IP reanimates with casual abandon. Guardrails, it seems, are more suggestion than constraint. The spectacle may be amusing—until one considers that moderating synthetic media at scale is about as about as effective as stapling water to a wall. Social feeds now carry a permanent question mark: is this real, or is it Sora?
Meanwhile, Meta has chosen a more literal form of reality capture. From December 16, it will begin feeding the contents of your conversations with Meta AI—text, voice, even image prompts—into the company’s vast advertising machine. There is no opt-out. Unless you happen to live in the EU, the UK, or South Korea, where regulators still believe privacy law is more than a polite fiction, your chats will soon help determine the ads that stalk you across Facebook and Instagram. Reuters broke the news; the Financial Times quickly followed.
The data won’t just fine-tune ads. According to The Verge, your AI chat logs will seep into the very marrow of Meta’s products: Instagram Reels, suggested posts, even group recommendations. Sensitive topics—health, politics, religion—are, for now, off the table. But the company’s ambition is plain: to transform idle banter with its chatbots into a feedback loop powering ever more personalised feeds. In the name of “relevance,” users will find their own words recycled into the content they consume.
Corporate AI, however, is not all dystopia. Salesforce, never one to miss a branding flourish, has launched Agentforce Vibes—a system that turns natural-language requests into fully fledged enterprise apps. Its “Vibe Codey” agent promises to generate deployable software inside Salesforce environments, with governance and multi-agent controls wired in from the start. For developers exhausted by endless Jira tickets, it is pitched as liberation; for compliance officers, a pre-packaged reassurance that the machines won’t run wild. Expect it to become the latest badge of honour on LinkedIn for admins keen to prove their organisations are vibing with the times.
At the infrastructure level, the contest is no less fevered. GPUs remain scarce, but memory is emerging as the true chokepoint in the AI arms race. OpenAI, ever eager to stay ahead of demand, has enlisted Korean heavyweights Samsung and SK Hynix to secure a pipeline of high-bandwidth DRAM for its looming “Stargate” data-centre project in the United States. That a software company now negotiates supply contracts like a chip foundry underscores the new economics of AI: algorithms may be clever, but they are nothing without terabytes of fast memory to chew on.
Taken together, these stories chart the contradictory mood of the AI moment. On one end, consumers face uncanny videos and unwanted surveillance of their digital confessions; on the other, enterprises are promised tools that code themselves, while infrastructure giants scramble to keep the lights on. The industry’s acceleration is undeniable. Whether society can keep pace with its consequences is a different question entirely.
Image source: @GabrielPeterss4 on X who found himself with the most liked video on sora 2 after he created a deepfake of Sam Altmann being caught shoplifting at target. The internet remains undefeated.