AI Safety Has a Mentor Shortage: Junior Researchers Asked to Mentor Too Soon
sethlazar · x · 2026-09-20
AI safety researcher Seth Lazar lays out three structural problems with the current AI safety talent pipeline:
- Mentors aren't scaling: capability-building programs expand fast, but mentorship doesn't — barely-post-fellowship juniors are asked to mentor, and unsupervised junior mentors risk misconduct due to inexperience and power differentials.
- Apprenticeship is too episodic: programs last a couple of months, versus the long-term commitment of traditional apprenticeship; Lazar notes he has hired (or plans to hire) all of his own mentees.
- No progression: people cycle through pipeline programs accumulating papers (with a jab at ICLR churn) without real career advancement.
He notes most of these orgs are funded by Coefficient and calls for systemic design around mentor supply, longer-term mentorship, and progression, asking whether Kairos has the critical mass to drive this unilaterally.
More from AGI Musings
- 'Sutskever instantly saw superintelligence': researcher grapples with failed frontier expectations — teortaxesTex · 2026-09-20
- lateinteraction: longer thinking before hard decisions could invert the AI Jevons paradox — lateinteraction · 2026-09-20
- What does 'take over the internet' actually mean? Dario's botnet warning scrutinized — midwoodgirl10 · 2026-09-20
- Eric Topol on AI in medicine: aging, prevention, and balancing skepticism with hope — EricTopol · 2026-09-20
- Copyright rulings would accelerate licensing shifts, not slow AI training — iamtrask · 2026-09-20
- "If AI automates half of work but we all still work full-time, where did the savings go?" — VraserX · 2026-09-20