Image: made using DALL·E (Custom edited by me)
RSN024
Published 15.09.2025.09.00.MON
Pipes, Power & Probabilities: How Incentives Are Rewiring Tech
AI’s growing pains were on full display this week. OpenAI floated a provocative fix for hallucinations, Google quietly admitted the open web’s ad economy is shrinking, Apple pushed fresh hardware amid muted AI talk, and Wall Street reeled as OpenAI inked a colossal Oracle deal with unanswered power questions. Meanwhile, Nepal’s protests showed how fast young, networked movements can route around censorship with Discord, VPNs and Bitchat. It’s a snapshot of a tech landscape where incentives shape outcomes, distribution trumps purity, and real-world stakes, from ad revenues to civic rights, ride on seemingly abstract engineering choices.
1. Hallucinations Aren’t Bugs, They’re Incentives
Punish confident wrong answers, reward “I don’t know.”
What’s Happening: OpenAI’s new paper argues LLMs hallucinate partly because the community trains and scores them in ways that incentivize guessing. Pretraining optimizes next-token prediction from fluent text without true/false labels, so models pick up robust patterns but fumble sparse facts. Accuracy-only leaderboards act like multiple-choice tests that reward lucky guesses and penalize abstaining. OpenAI proposes uncertainty-aware scoring that penalizes confident errors more than calibrated “I don’t know” replies, and grants partial credit for appropriate uncertainty. They argue mainstream evaluations must adopt these rules or models will keep learning to bluff.
Why it matters to you: Models that admit uncertainty waste less of your time, reduce wild goose chases, and protect your reputation when you rely on AI for work, study, and decisions.
Why it matters to the world: Uncertainty-aware evaluation can shift an entire industry’s optimization target, improving safety across search, health, finance, and governance systems that increasingly sit atop LLMs.
2. Google Accidentally Said the Quiet Part Out Loud
Courtroom candor vs. conference messaging.
What’s Happening: In an ad-tech antitrust filing, Google said the “open web [display advertising] is already in rapid decline,” citing AI’s reshaping of ad markets and capital flows to connected TV and retail media. That stance clashes with recent public messaging that the web is “thriving.” Google later clarified it meant open-web display ads specifically, but the admission mirrors publisher pain: algorithm shifts and AI overviews dampen referral traffic and CPMs. Google argues divestiture would accelerate the decline and harm publishers reliant on open-web ads, while regulators contend structural separation is needed to restore competition. The macro trend is clear: more spend is moving to walled gardens, shoppable media, and AI-summarized results, leaving independent sites squeezed unless they diversify revenue or win direct relationships.
Why it matters to you: Expect more paywalls, newsletters, and memberships as sites chase reader revenue, plus greater reliance on curated apps and aggregators over raw web search.
Why it matters to the world: As ad dollars consolidate into closed platforms, media plurality and independent journalism face fiscal strain, increasing the risk of homogenized information ecosystems.
3. Apple Doubles Down on Refinement with the Awe Dropping Event
Thin phones, steadier wearables, quieter AI.
What’s Happening: Apple’s September lineup brought iPhone 17 with ProMotion on the base model, a super-slim iPhone Air with A19 Pro and Wi-Fi 7, refreshed 17 Pro/Pro Max with pro-video perks, Watch Series 11 and Ultra 3 updates, a Watch SE 3 refresh, and AirPods Pro 3 with improved ANC and live translation. iOS 26 and watchOS 26 extend the Liquid Glass aesthetic and sprinkle AI-assisted features, but the keynote underplayed “Apple Intelligence,” leaning instead on battery, displays, cameras, and accessories. The pitch was practical refinement over flashy frontier demos.
Why it matters to you: If you upgrade, the wins are tangible—smoother screens, sturdier designs, better photos, and more polished health tools without relearning your device.
Why it matters to the world: Apple’s tempered AI posture signals a mainstream pivot: ship reliable, privacy-centric features first; let generative glam follow when it’s robust and culturally accepted.
4. Not Hype, $300B Cloud Deal Meets a Power Grid Reality Check
$300B for compute, now find the power.
What’s Happening: OpenAI inked a five-year, $300B agreement for Oracle Cloud Infrastructure, underscoring a multi-cloud, hyper-scale strategy and Oracle’s muscle in high-performance data centers. The spend telegraphs OpenAI’s appetite to scale inference and training globally but raises two questions: payment and power. The capex-like commitment implies pre-arranged financing or pass-through pricing to enterprise customers. More urgent is electricity: multi-gigawatt footprints require long-term power purchase agreements with solar, storage, gas peakers, or nuclear. Oracle will shoulder much of the physical build-out while OpenAI stays asset-light, tying model roadmaps to grids, permitting, and geopolitics.
Why it matters to you: Bigger backends should mean snappier apps and higher usage caps, but also rising prices for premium tiers as compute and energy costs flow through.
Why it matters to the world: AI’s growth is now an energy story; siting data centers and sourcing clean power will shape emissions trajectories and regional industrial policy.
5. Nepal’s Gen Z Just Gave Us a Masterclass in Routing Around Censorship
When platforms shutter, networks rewire.
What’s Happening: After Nepal blocked major social apps, Gen-Z-led anti-corruption protests surged, coordinated via Discord servers, VPNs, and Bluetooth-mesh tools like Bitchat. Hashtags such as #NepoKids galvanized anger at elite excess, while VPN signups spiked and offline-capable messaging offered censorship-resilient backchannels. Clashes turned deadly; the government restored some platforms, but dissent widened, culminating in political upheaval. The episode spotlights how attempts to throttle speech often backfire, accelerating tool adoption and diaspora amplification.
Why it matters to you: Your communications resilience shouldn’t depend on one app. Knowing VPNs, offline chat, and backup channels keeps you connected when networks wobble or are throttled.
Why it matters to the world: Censorship pushes civic tech to evolve. As states toggle kill-switches, citizens iterate on decentralized coordination, reshaping the balance between control and assembly.
Conclusion
Across these stories runs a single throughline: incentives and infrastructure decide outcomes. Change the evaluations and models stop bluffing. Shift ad spend and the open web strains. Tune devices for reliability and adoption follows. Sign colossal cloud deals and you inherit the energy grid’s constraints. Close platforms and movements route around you. The next era won’t be won by raw capability alone, but by aligning the right incentives with dependable pipes— computational, electrical, and social—so progress compounds without breaking trust.
— Ritik Sharma, Founder, RSN (Ritik Sharma Newsletter)