Bitter Lesson
An idea about AI development put forward by Rich Sutton, holding that approaches relying on large-scale computation and general learning methods will ultimately prevail over elaborate designs that rely on human domain knowledge.
Timeline (last 90 days)
The developers' choice to adopt frontier LLMs let them "stand on the right side of the Bitter Lesson."
The Bitter Lesson proposed by Richard Sutton points out that in the long run, AI systems that leverage large-scale computation and general learning methods such as search and learning will ultimately prevail over those systems that rely on human domain knowledge and hand-designed features.
Kohli believes that the Bitter Lesson has clear boundaries of applicability in biology, because biological data is far less abundant and cheap than internet text.
The correct interpretation of the Bitter Lesson is that 'only algorithms that can scale matter,' and model accuracy equals efficiency multiplied by resources
All Facts (4)
The developers' choice to adopt frontier LLMs let them "stand on the right side of the Bitter Lesson."
50%UnverifiedThe Bitter Lesson proposed by Richard Sutton points out that in the long run, AI systems that leverage large-scale computation and general learning methods such as search and learning will ultimately prevail over those systems that rely on human domain knowledge and hand-designed features.
50%UnverifiedKohli believes that the Bitter Lesson has clear boundaries of applicability in biology, because biological data is far less abundant and cheap than internet text.
50%UnverifiedThe correct interpretation of the Bitter Lesson is that 'only algorithms that can scale matter,' and model accuracy equals efficiency multiplied by resources
50%