Junior AI safety researchers: Don't write your first paper solo
Only 20% of NeurIPS papers have fewer than 4 authors—here's why.
In a LessWrong post, AI safety course facilitator Mikhail Mironov argues that junior researchers should not attempt to write their first AI safety research paper as a single author. He notes that at top ML conferences like NeurIPS 2025, papers with fewer than 4 authors are extremely rare, and the typical workload for a 4-author paper involves three junior researchers each putting in 15-20 hours per week for ~4 months, plus a senior scientist contributing 2-5 hours weekly. Replicating that as a solo act would demand roughly 6 months of full-time work—a daunting and often unrealistic commitment.
Mironov recommends part-time research fellowships such as SPAR (with a 20% acceptance rate), AI Safety Camp, or Apart Fellowship as a more structured path to publishing. These programs pair small teams with a senior mentor, provide fast feedback cycles (daily team communication + weekly mentor meetings), and help juniors identify impactful yet solvable problems—a skill they likely lack. Working alone risks reinventing the wheel, getting stuck on unsolvable problems, or pursuing unimportant questions. The post concludes that solo research can still be useful for skill-building (e.g., reproducing papers) but should not be the primary route to a first publication.
- NeurIPS 2025 papers rarely have fewer than 4 authors; solo publishing requires ~6 months full-time equivalent work.
- Part-time fellowships like SPAR (20% acceptance rate) provide mentorship, team collaboration, and faster feedback.
- Junior researchers risk poor problem selection and wasted effort without experienced guidance.
Why It Matters
First publication success rates increase dramatically with structured mentorship and teamwork—saving months of wasted effort.