Podcast, Interview
Alex Lawsen on avoiding 10 mistakes people make when pursuing a high impact career
- Spending two years writing intermittently without publishing prevents determining if one is excelling, potentially blocking a pivot to fields like podcasting where the individual might succeed.
- Failing to receive clear signals to try a different path after an ambiguous start can harm an individual's career trajectory and make the path difficult to navigate.
- Individuals who are a poor fit for technical AI safety but rule themselves out may incorrectly conclude they "suck" rather than identifying viable alternative roles.
- In technical AI roles, the primary bottleneck is not a lack of people good at math and programming, but a shortage of those who are exceptionally great at specific tasks, making the value of a better fit significantly higher than a mediocre performance on a top task.
- If an individual tries three distinct things without generating excitement or a sense of "crushing it," they have gathered sufficient information to change their career path.
- Hiding work to avoid negative feedback prevents receiving the necessary signal to switch to a better path, often stemming from a fear of the pain associated with failure.
- Applying to only five jobs and receiving rejections at the first stage suggests a lack of ambitious application or a failure to apply to a sufficient number of roles.
- Applying to 20 roles with a decent chance of success and receiving no feedback indicates a completed attempt to "fail hard," signaling it may be reasonable to stop pursuing that specific role.
- Setting success criteria for an application strategy, such as seeing different stages of hiring across applications, helps determine if the strategy is functioning correctly.
- Applying to four hyper-competitive roles and reaching the final round of two is consistent with ambitious application rather than failure.
- Individuals currently in fulfilling jobs are less likely to need advice on "trying hard to fail" or switching careers compared to those in unsatisfying roles.
- Early-career individuals should follow normal heuristics like optimizing for learning and grades, whereas those who are great programmers but dislike technical safety research should not force themselves into it for more than a couple of years.
- Running a student group instead of focusing on personal technical learning may hinder preparation for a technical research career.
- Those who are great managers or mentors may have a big multiplier effect by setting themselves up now to steer future talent, especially as fields become less neglected.
- Switching jobs every 9 to 12 months prevents the practice necessary to become "really, really good" at a specific set of skills and incurs significant costs in emotional resilience and time.
- Most people should adopt a policy of staying in a chosen role for two years before re-evaluating, unless convinced otherwise by strong evidence.
- The information-gathering phase for early-career individuals should be short (e.g., a week), whereas commitment duration should be longer later in a career.
- For those early in their career, spending a month deciding on a 12-month role is reasonable if options are limited and the choice is difficult.
- Spending a year writing intermittently without publishing yields an ambiguous outcome where it is impossible to discover if one is truly excelling.
- A PhD program might take seven years, which is rational within a 10-year AI timeline, leaving only three years for actual work.
- Individuals with 10-year timelines for AI should be prepared to spend seven or eight years setting themselves up rather than doing immediate impact work.
- For most of the next 20 years, the most useful alignment work will occur close to AGI arrival, shifting incentives toward setting up later rather than immediately.
- A 7-year research project is useless if AI arrives in less than 7 years, but should still be undertaken if it represents the individual's only hope.
- An expected value calculation supports long-term career capital building even with a 15% chance of AI arriving in three years and a 35% chance in the following seven, provided the individual can significantly increase their effectiveness in later worlds.
- Individuals in fields with catastrophic risk face unusually high uncertainty, making the standard for being "reasonably sure" a much higher bar than in average decisions.
- Intuitions that contradict the consensus of "super smart people" in a field should not be automatically dismissed but should be examined for tension.
- A career path an individual would choose if guaranteed to be great for the world is likely a good candidate to consider for actual impact.
- Hating community building is a clear reason it might be a mistake to pursue it as a proxy for impact.
- If an individual struggles for time and does not see long-term skills development in community building, they should set up for a different long-term path.
- Updates from 20 people in a community regarding career advice should not be treated as independent data points but as a single correlated update.
- Listening to a podcast and agreeing with a confident conclusion without evaluation may lead to the wrong conclusion if the advice does not apply to the specific situation.
- If the speaker is wrong, the listener may be the person who actually needs to hear the opposite advice.
- The AI governance team at Open Philanthropy represents the current working environment, implying a future trajectory of focusing on AI governance.
- The speaker anticipates listeners will find the episode resonant and learn how to avoid mistakes they have already made.
- The speaker expects personal fit to be a significantly recurring theme during the conversation.
- Setting a "success criterion" for an application strategy helps an individual determine if their strategy is working correctly.