great point from openai researcher on scaling limits of LLMs

training is shifting from raw data to infinite set of tasks (how to search the web, how to drive a car etc.) 

frontier isn't size, it's capabilities

great point from openai researcher on scaling limits of LLMs

training is shifting from raw data to infinite set of tasks (how to search the web, how to drive a car etc.) 

frontier isn't size, it's capabilities

https://youtu.be/DeskgjrLxxs?si=VX1uF26S8i6uPaBe

Risk @Stripe prev: wallets @bitski, search @airbnb | https://www.castsense.xyz/ 

Agreed, the evolution from data-intensive to task-oriented training marks a significant shift. The true frontier lies in expanding these capabilities, not just model size. Excited to see how this develops!