In short

A roundup surveying live tensions in AI adoption: OpenAI's proposal to curb hallucinations by penalising confident wrong answers over honest uncertainty; China's 'AI+' strategy prioritising rapid application-layer deployment (targeting ~90% intelligent-terminal penetration by 2030) over foundational research; and Common Sense Media data showing a third of teens treat AI as a companion. It also asks whether AI is just another hype cycle, concluding mass adoption sets it apart.

The second Tech Futures Project roundup ties together three stories that look unrelated but aren’t: why models lie, how China is deploying AI, and how teenagers are bonding with it.

On hallucinations, the newsletter highlights OpenAI’s argument that the problem is partly self-inflicted: we benchmark models in ways that reward confident guessing over honest “I don’t know,” so we get confident guessers. On China, it explains the “AI+” strategy as a bet on deployment, pushing AI across the whole economy rather than chasing the frontier, with eye-watering 2030 adoption targets. And on AI companions, it sits with the unsettling Common Sense Media finding that a third of teens already treat AI as a friend.

Running underneath is a regulatory point worth noting: as capability shifts from training to inference-time reasoning, rules pegged to compute thresholds start to lose their grip.

Key takeaways

  • Hallucinations may be reduced by changing evaluations to reward calibrated uncertainty rather than confident guessing.
  • China's 'AI+' policy favours fast, economy-wide deployment over frontier research, aiming for ~90% intelligent-terminal penetration by 2030.
  • A third of teens engage AI as a companion and a quarter share personal information, raising developmental and privacy concerns.
  • Mass adoption (800m weekly ChatGPT users) distinguishes AI from prior hype cycles.
  • The reasoning/inference-scaling paradigm weakens compute-threshold-based AI regulation.

Read the full piece

This is a summary. Read the complete essay, with all the sources and argument, on Substack.

Read on Substack →

Frequently asked questions

How can AI hallucinations be reduced?
OpenAI argues hallucinations partly stem from evaluations that reward confident guessing; changing benchmarks to penalise confident errors and reward acknowledged uncertainty would discourage models from fabricating answers.
What is China's 'AI+' strategy?
A national policy emphasising rapid, economy-wide deployment of AI at the application layer, aiming for roughly 90% penetration of intelligent terminals by 2030, rather than competing primarily on foundational model research.
How are teenagers using AI companions?
Common Sense Media research suggests about a third of teens treat AI as a companion and a quarter share personal information with it, raising privacy and developmental concerns.

People & ideas in this piece

OpenAICommon Sense MediaGrace ShaoArvind NarayananAI+ strategyModel hallucinationsReasoning paradigmCompute-based thresholds

Topics: AI & Geopolitics , How AI Actually Works