Thinking About Thinking
Jun 20, 2024
ML knowledge and capabilities have grown exponentially with unprecedented rate of increase in model abilities, research output, compute resources, and real-world applications. ML related research papers have increased at an annual rate of 35% since 2012. Model parameters, as a first order estimate for model complexities and capabilities, has grown roughly 10 times every 1-2 years. Models like GPT-4 and Claude has achieved near human scores on challenging tasks, such as standardized exams and domain specific benchmarks like GLUE (General Language Understanding Evaluation) and MMLU (Massive Multitask Language Understanding).