80,000 Hours Podcast

#219 – Toby Ord on graphs AI companies would prefer you didn't (fully) understand

367 snips
Jun 24, 2025
Toby Ord, an Oxford philosopher and bestselling author of 'The Precipice,' dives into the shifting landscape of AI development. He highlights how AI companies are moving from simply increasing model size to implementing more thoughtful reasoning processes. This transformation raises crucial questions about accessibility and the ethical dilemmas we face as AI becomes more powerful. Ord also discusses the economic implications of these changes, emphasizing the urgent need for adaptive governance to tackle the complexities of evolving AI technologies.
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
INSIGHT

Shift to Inference Scaling

  • AI companies have shifted from scaling training compute to scaling inference compute due to data scarcity and high costs.
  • This inference scaling makes AI more powerful but significantly increases the cost per usage of advanced models.
INSIGHT

Gradual, Costly AI Access

  • Inference scaling means AI progress will be gradual and costly at deployment, delaying widespread access.
  • Early superhuman AI access will be limited to those who can afford massive usage costs, impacting accessibility and pacing.
INSIGHT

Weights Less Critical with Inference Costs

  • When inference cost dominates, possession of model weights matters less than access to compute to run those models.
  • This reduces the risk from weight leaks but also diminishes the impact of open sourcing AI model weights.
Get the Snipd Podcast app to discover more snips from this episode
Get the app