| https://twitter.com/k_cieslak | |
| GitHub | https://github.com/Krzysztof-Cieslak |
| GitHub Next | https://githubnext.com/team/krzysztof-cieslak/ |
| https://twitter.com/k_cieslak | |
| GitHub | https://github.com/Krzysztof-Cieslak |
| GitHub Next | https://githubnext.com/team/krzysztof-cieslak/ |
People are misunderstanding and misusing LLMs.
Their most considerable value is not what they "know" (i.e. the ability to [sometimes] answer questions) but rather their ability to reason about a given text.
With 32K tokens prompt size in GPT-4, the vast majority of work should be spent on filling this prompt with relevant context and designing the prompts to force AI to answer questions based on the prompt rather than its internal "knowledge".
*CEOs earn 1000% more over the last 40 years*
Economists: this is fine.
*employee wages grow for the first time in 40 years*
Economists: THE LABOR MARKET IS TOO HOT AND IS DRIVING INFLATION.