Dwarkesh Patel > Quotes

“Let’s go back to the original question of how you measure when human-level AI has arrived,”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“It depends on how you want to define reasoning.”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“General intelligence is the ability to approach any problem, any skill, and very quickly master it using very little data.”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“Figure 4 shows results from one core benchmark for such general ability: BIG-Bench, a collection of 200 tasks selected to be hard for language models to accomplish.”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“Two ways to fit the same data.”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“tasks, which include identifying anachronisms, reading ASCII art, and determining the author of a passage,”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“We’re looking at that in different ways, partly through simulation—for example, using very realistic game environments to generate realistic data—but also through self-play. That’s where systems interact with each other or converse with each other.”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“Honestly, we have to admit it’s weird. It doesn’t match up. This is one reason I’m a bit skeptical of biological analogies. I thought in those terms five or six years ago.”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025

“We want the models to complete real tasks, like writing code, doing our grad school homework, or, eventually, making money for us while we kick back.”
― Dwarkesh Patel, The Scaling Era: An Oral History of AI, 2019–2025