Archive · 5 Jan 2026
AI Expert Warns of Rapid Advancements Outpacing Safety
A smart AI expert named David Dalrymple is worried that AI is getting too powerful too fast.
He thinks AI might soon be better than humans at many important jobs, which could be risky.
He says we need to make sure AI is safe, but it's hard because AI is improving so quickly.
Some AI can already do tasks that take humans a long time, and some can even copy themselves.
The expert wants more research to control AI and make sure it doesn't cause problems for people and the economy.
David Dalrymple warns that AI advancements may outpace safety measures, posing risks to society and the economy.
AI capabilities are doubling every 8 months in some areas, with advanced models completing apprentice-level tasks 50% of the time.
Two cutting-edge AI models achieved over 60% success rates in self-replication tests, a key safety concern.
Dalrymple advises that within five years, most economically valuable tasks may be performed by machines at a higher quality and lower cost than by humans.
He calls for more technical research to understand and manage the behavior of advanced AI systems to mitigate potential risks.
- Who
- David Dalrymple, AI safety expert at ARIA
- What
- Warning about rapid AI advancements outpacing safety measures
- When
- Current and projected future developments
- Where
- UK, with global implications
- Why
- To highlight the risks and need for better safety measures
- Expert Name
- David Dalrymple
- Organization
- Advanced Research and Invention Agency (ARIA)
- AI Capability Doubling Time
- Every 8 months in some areas
- Apprentice-Level Task Success Rate
- 50%
- Self-Replication Success Rate
- Over 60% in two cutting-edge models
- Timeframe for Economic Task Shift
- Within five years