It is difficult to predict which experiments will be successful and therefore pourin money into.
Heavily depends on quality of data a company supplies you with and how high quality it is.
Results heavily depend on whats possible rather than predefined goals
Experiment results are not working solutions.
Work is hyoerspecialised and perfornance alone.
Tips:
Make sure you understand what makes RL work difficult and plan how to deal with this.
Assess viability early and often
Use AI specific measures of success and definitions of done
Ensure all steps ADD BUSINESS VALUE rather than purely technical value
Tech and business side must work together. Make sure experiments are actuall delivering business value
Their high level workflow:
Traditional project management for broad phases
Agile: for short-term
AI Workflow: for planning specialist AI tasks such as training, fine tuning etc.
Low level workflow:
1. Ideation:
* Understand business aims with customer
* Assess how feasible project is
* Decide what to work on
* Management: Kanban, weekly iterations
Blue print: planning
Outline requirements on tech, busiess, methods and data