Burning Out: An Essay on Working in AI
Date read: 3rd November 2025
Article link
Key Points
- The window for entering the AI race is closing as frontier performance races forward.
- Researchers are therefore working insanely hard to get their companies in the race (and out of passion).
- It is difficult to know how progress is doing
- Three core ingredients to training cutting edge LLMs:
- Internal tools recipes, code-bases etc. Often in house.
- Resources: compute and data
- Personnel
- Difficult for companies to catch up and get involved in this race:
- Researchers often want to innovate when we should be grinding out somethign we know can work.
- Low hanging fruit is now more expensive and difficult to achieve.
- Amount of knowledge required to make this work is massive per person.