I recently made a career move and changed my focus from NLP to biology. I made this decision after months of thinking and deliberation. Therefore, I decided to document my decision making process here so that I can revisit and optimize it in the future.
Build personal moats by developing high in demand skills that have high barriers to entry. Don't be easily replaceable by humans. Example skills that are hard to get: Machine learning, distributed software, vertically integrated software etc.
Build moats by combining a set of these compounding specialized skills. Keep in mind that Humans and machines have opposite strengths and weaknesses. Humans are better than machines in open ended, broad domains. Generalizing across multiple compounding domains can bring unique skill set combinations with high returns, especially in a world where the importance of skill in specialized domains disappear due to the paradox of skill.
Work on hard & vertical & long term problems with long term people
Work on high risk vertical progress problems. Talent is equally distributed, but opportunity is not. The number of people who work on hard problems is very low compared to the number of people who work on horizontal progress - copying things that work and bringing those to the entire world. Working on hard problems require long term thinking with long term oriented people and result in more enjoyable work with compounding returns for all participants.
Most people can't take risks either due to a high cost of living, risk averseness, family responsibilities or cultural biases. However, * The world is driven by tail events and high returns come from risky bets which require getting used to repeated failures. * To make higher than average returns, embrace being uniquely wrong.
Don't follow the crowd, instead, find things that are interesting to you, you can be uniquely good at and that are also valuable for society. It's ok if those things are undervalued by society, it's better if anything because * The only valuable things are the worthless things.