Welcome to this edition of AI Horizon! We’ve gathered six exciting AI developments—from Meta’s plan to automate ad campaigns to AI systems that rewrite their own code. Each topic is boiled down to easy-to-read bullet points so you can catch up fast. Let’s dive in!
Meta’s AI-Powered Ads by 2026 🚀
Meta will let AI handle the entire ad process—creative, copywriting, targeting, budgeting—using only a product image and budget.
Ads will dynamically adapt to each viewer’s context (e.g., showing mountains to someone in Colorado or a cityscape to someone in New York).
Aimed at small businesses without marketing teams, promising professional-grade ads without agency fees.
Since advertising is 97% of Meta’s revenue, this is central to Zuckerberg’s AI strategy.
The Darwin Gödel Machine: AI That Codes Itself 🤖
Developed by Sakana AI and UBC, this agent autonomously rewrites its own code to improve performance.
Uses “mutations” and “selection” like Darwinian evolution—testing code changes on benchmarks and keeping what works.
Achieved up to 150% performance gains on popular coding tests.
Builds an archive of better-performing versions, enabling continuous self-improvement as compute resources grow.
Samsung & Perplexity: New Bixby Buddy? 📱
Samsung is in talks to integrate Perplexity’s AI assistant into upcoming Galaxy devices (possibly launching in 2025).
Perplexity could be preloaded as an app, enhance Bixby’s chat capabilities, and power AI search in Samsung Internet.
This partnership would reduce Samsung’s reliance on Google’s AI services.
Joint R&D discussions include an AI-focused OS layer and a hub to manage multiple AI assistants from different providers.
Apple’s “AI Gap Year” at WWDC 2025 🍏
WWDC 2025 (starting June 9) will feature limited AI announcements—a “gap year” before major reveals in 2026.
Apple plans to open its 3B-parameter models to developers for modest on-device features and rebrand existing features as “AI-powered.”
Rumored projects—an LLM-based Siri overhaul, health-focused Project Mulberry, and a ChatGPT competitor—are delayed to 2026.
Apple’s 150B-parameter model is reportedly nearing ChatGPT quality, but accuracy issues have kept it under wraps for now.
Music Labels vs. AI: Dollars & ♫ Licensing 💿
Universal, Warner, and Sony are negotiating licensing deals with AI music startups Udio and Suno to settle billion-dollar copyright lawsuits.
Labels seek licensing fees plus equity stakes and want fingerprinting tech (similar to YouTube’s Content ID) to track AI usage.
Lawsuits from 2024 demand up to $150,000 per infringed work, potentially totaling billions.
A deal would set a precedent for compensating artists when AI generates new music from their recordings.
AI Outperforms Humans on Emotional IQ Tests 💖
Researchers at the Universities of Geneva and Bern tested six AI models (including ChatGPT-4 and Claude 3.5 Haiku) on emotional intelligence (EI) assessments.
AIs scored an average of 82% compared to 56% for human participants.
ChatGPT-4 even generated new, valid EI test questions on its own.
While AI excels at choosing the most empathetic response, it still can’t truly “feel” emotions—human nuance remains irreplaceable.
Just sharing, I hope you do not mind.
The True Threat of AI is Global Compute Governance, Job Loss, and Economic Displacement https://torrancestephensphd.substack.com/p/the-true-threat-of-ai-is-global-compute