Runway bets video beats language models

TechCrunch

Runway is expanding beyond AI video generation to build world models that learn from observational data rather than text, competing against Google and OpenAI with the belief that video-based AI will unlock scientific breakthroughs in fields from robotics to drug discovery. The $5.3 billion company's unconventional New York roots and focus on early revenue generation may give it an edge, though securing sufficient compute resources remains critical.

arXiv

Researchers propose 'Positive Alignment' as a necessary expansion of AI safety research, shifting from purely preventing harm to actively promoting human flourishing through context-sensitive, pluralistic AI systems with decentralized governance. The framework addresses current failures in engagement, autonomy, and truth-seeking through technical approaches spanning data curation to community customization.

Stanford Digital Economy Lab

Early-career workers in AI-exposed occupations have experienced a 16% employment decline since widespread generative AI adoption, while more experienced workers remain stable, providing the first large-scale evidence of AI's disproportionate impact on entry-level employment.

Prime Intellect

Prime Intellect's autonomous AI agents (GPT-5.5 and Claude Opus 4.7) ran 10k optimization experiments over two weeks, beating human baselines on the nanoGPT speedrun by reducing training steps from 2,990 to 2,930. The agents were effective at hyperparameter search but struggled with novel ideation and needed human records to maintain progress.

TechCrunch

A Menlo Ventures partner estimates 10,000 AI company insiders have achieved $20M+ wealth while widespread layoffs and AI displacement create anxiety among other tech workers, highlighting extreme wealth inequality in the current AI boom.