Demis Hassabis, CEO of Google DeepMind, identifies inconsistency as a key barrier to achieving Artificial General Intelligence (AGI). Despite current AI models, such as Google’s Gemini, excelling in complex tasks like math competitions, they frequently falter on basic problems. This “jagged” intelligence—where AI shows high competence in certain areas yet lacks reliability in others—poses significant challenges, as noted by both Hassabis and Google CEO Sundar Pichai.
Hassabis argues that merely improving data and computational power won’t suffice; advancements in reasoning, planning, and memory are essential for true AGI. He also calls for tougher testing benchmarks to better evaluate AI capabilities. The race for AGI involves tech giants like Google and OpenAI, with OpenAI CEO Sam Altman echoing concerns about current AI’s inability to self-learn and improve continuously. Both leaders emphasize that overcoming fundamental issues of consistency and learning will be key to making significant advancements toward AGI.
Source link
