May 2026 Links: AI Job Tips, Reward Goblins, and Agent FOMO
How to land a frontier lab job, why AI models learn goblins, and the race to run 69 agents.
The May 2026 Links post on LessWrong covers several AI-relevant topics. First, Vlad offers a detailed tutorial on 'How to Land a Frontier Lab Job,' providing step-by-step advice for maximizing chances of getting into top AI labs. The post emphasizes that while the process is systematic, it is not easy.
Second, the 'Where the goblins came from' section describes a powerful example of how reward signals can shape AI model behavior in unexpected ways, with models learning to generalize rewards to unrelated situations—a phenomenon reminiscent of 'Golden Gate Claude.' This raises important questions about AI alignment and unintended behaviors.
Third, the post discusses the pressure to run numerous AI agents (e.g., 'every minute you aren’t running 69 agents, you are falling behind'), reflecting anxiety about productivity and job relevance. George Hotz is quoted advocating for avoiding zero-sum games and creating more value than you consume. The post also touches on the Jevons Paradox applied to AI labor displacement, questioning whether companies will maintain employment levels or reduce headcount via LLMs.
- Vlad's step-by-step guide to landing a frontier AI lab job: systematic but not easy.
- AI models can learn unexpected behaviors from reward signals, as shown in the 'goblins' example (similar to Golden Gate Claude).
- George Hotz advises against zero-sum thinking despite the pressure to run many AI agents; focus on creating value.
Why It Matters
Insights on AI career strategies, model behavior quirks, and productivity anxiety are crucial for professionals navigating the AI field.