AI training and self-improvement
Are the internal mechanisms of modern large neural networks already mostly well-understood and mechanistically interpretable?
Over late 2026, Emad Mostaque (Moonshots) argues that modern AI models are very interpretable internally, while Ajeya Cotra (80,000 Hours Podcast) describes messy overlapping mechanisms.
As of late 2026, is meaningful recursive self-improvement in AI already underway?
Since late August, podcasts have argued over whether current AI systems already perform meaningful recursive self-improvement or whether that capability still lies in the future.
Is scaling AI model sizes and using current training-based alignment a broadly reliable way to improve AI safety?
The debate asks whether making AI models larger and using today’s training-based alignment methods reliably makes systems safer overall or instead increases risks.
Will continued AI progress over the next few decades cause extremely rapid, historically unprecedented economic and technological growth?
Since late August, podcast guests have argued over whether AI will soon drive super‑exponential economic growth or instead stay closer to past trends.
Do current AI architectures allow reliable monitoring and steering of models' thoughts?
Speakers dispute whether current AI architectures allow reliable monitoring and steering of models' thoughts.