TickerTain — Make Analysis Fun Again

←

▶ 3:53 · Frontier AI Models · Deployment becomes training in continual learning regime accelerating returns to scale for leading AI labs

now playing · Frontier AI Models

AI Safety & Alignmentriskscore 8/10dwarkesh patel

Continual learning breaks current AI safety regulatory framework assuming frozen model deployment

Current AI safety regulation assumes a distinct training-then-deployment phase, but continual learning merges these phases, making point-in-time safety evaluations obsolete and requiring on…

AI Regulation & Policyriskscore 7/10dwarkesh patel

Train-then-deploy regulatory frameworks become obsolete under continual learning

Current safety regulations assume a clear boundary between training and deployment. With models improving daily from real-world usage, that boundary disappears, making point-in-time evaluat…

AI Safety & Alignmentriskscore 6/10dwarkesh patel

Alignment research must shift from frozen weights to constant weight updates under continual learning

Current alignment focuses on ensuring frozen weights behave well during deployment. With continual learning, weights update constantly, requiring new research on preventing jailbreaks, dece…

Frontier AI Modelstailwindscore 6/10dwarkesh patel

Continual learning drives diversification of AI minds, avoiding current mode collapse

Today's base models are similar because they train on the same data. When models learn from diverse real-world deployments across different companies and instances, model outputs will diver…

Frontier AI Modelstailwindscore 9/10dwarkesh patel

Deployment becomes training in continual learning regime accelerating returns to scale for leading AI labs

When deployment data directly improves model weights daily, the lab with the best model and most usage enters a self-reinforcing loop where usage generates better models which attract more…

AI Economics & Business Modelstailwindscore 9/10dwarkesh patel

Continual learning creates switching costs and moats for leading AI labs

When models improve from user interactions across sessions, switching AI providers becomes equivalent to firing an experienced employee, enabling labs to charge high margins and create dura…

AI Economics & Business Modelstailwindscore 8/10dwarkesh patel

Continual learning creates switching costs that give leading AI labs durable moats and pricing power

When models improve from each user interaction, switching AI providers becomes like firing an experienced employee. This lock-in lets labs charge high margins, similar to cloud providers, a…

AI Infrastructuretailwindscore 8/10dwarkesh patel

Inference economics of continual learning favor large organizations due to massive batching requirements

Serving personalized model weights efficiently requires batch sizes of thousands of concurrent sequences, giving large enterprises with many employees and agents orders-of-magnitude compute…

AI Infrastructuretailwindscore 7/10dwarkesh patel

Inference batching economics at scale favor large organizations serving personalized model weights

Serving personalized weights efficiently requires batch sizes of thousands of concurrent sequences. Large enterprises with many employees and agents can amortize compute efficiently, while…

←

▶ 3:53 · Frontier AI Models · Deployment becomes training in continual learning regime accelerating returns to scale for leading AI labs

now playing · Frontier AI Models

AI Safety & Alignmentriskscore 8/10dwarkesh patel

Continual learning breaks current AI safety regulatory framework assuming frozen model deployment

Current AI safety regulation assumes a distinct training-then-deployment phase, but continual learning merges these phases, making point-in-time safety evaluations obsolete and requiring on…

AI Regulation & Policyriskscore 7/10dwarkesh patel

Train-then-deploy regulatory frameworks become obsolete under continual learning

Current safety regulations assume a clear boundary between training and deployment. With models improving daily from real-world usage, that boundary disappears, making point-in-time evaluat…

AI Safety & Alignmentriskscore 6/10dwarkesh patel

Alignment research must shift from frozen weights to constant weight updates under continual learning

Current alignment focuses on ensuring frozen weights behave well during deployment. With continual learning, weights update constantly, requiring new research on preventing jailbreaks, dece…

Frontier AI Modelstailwindscore 6/10dwarkesh patel

Continual learning drives diversification of AI minds, avoiding current mode collapse

Today's base models are similar because they train on the same data. When models learn from diverse real-world deployments across different companies and instances, model outputs will diver…

Frontier AI Modelstailwindscore 9/10dwarkesh patel

Deployment becomes training in continual learning regime accelerating returns to scale for leading AI labs

When deployment data directly improves model weights daily, the lab with the best model and most usage enters a self-reinforcing loop where usage generates better models which attract more…

AI Economics & Business Modelstailwindscore 9/10dwarkesh patel

Continual learning creates switching costs and moats for leading AI labs

When models improve from user interactions across sessions, switching AI providers becomes equivalent to firing an experienced employee, enabling labs to charge high margins and create dura…

AI Economics & Business Modelstailwindscore 8/10dwarkesh patel

Continual learning creates switching costs that give leading AI labs durable moats and pricing power

When models improve from each user interaction, switching AI providers becomes like firing an experienced employee. This lock-in lets labs charge high margins, similar to cloud providers, a…

AI Infrastructuretailwindscore 8/10dwarkesh patel

Inference economics of continual learning favor large organizations due to massive batching requirements

Serving personalized model weights efficiently requires batch sizes of thousands of concurrent sequences, giving large enterprises with many employees and agents orders-of-magnitude compute…

AI Infrastructuretailwindscore 7/10dwarkesh patel

Inference batching economics at scale favor large organizations serving personalized model weights

Serving personalized weights efficiently requires batch sizes of thousands of concurrent sequences. Large enterprises with many employees and agents can amortize compute efficiently, while…