If you work on making superintelligence go well, is there something you’re trying to do that keeps getting stuck?
Maybe it’s writing, running, or starting something. Or maybe it’s focusing, deciding, or coordinating.
When output gets stuck like this, I’ve seen two kinds of bottlenecks:
External bottlenecks. Things like health or sleep issues, financial or housing instability, and active crisis. Anything outside of your mind that gets in the way. (These are not my comparative advantage.)
Internal bottlenecks. Even when external circumstances are great, there are times when acting is easy and there are times when acting gets stuck. Isn’t that odd? Maybe your best times have been certain days or weeks, while working on past projects, collaborating with specific people, past roles, other areas of your life, or when you were younger. What’s different about these times? What stops the rest of your output from feeling like this?
After removing internal bottlenecks:
Daniel Tan, coauthor of Emergent Misalignment, was trying to do more research and enjoy life, but kept losing weeks and months to burnout. Willpower and deadlines would help with work, but only temporarily. In the summer of 2025, I showed him how to remove his internal bottlenecks. In the 12 months so far since, he reports that long periods of burnout no longer happen and that he’s regained ~5h/day from distractions. He’s used this time to pursue personal projects and host events for his friends instead. Details, video
An AI researcher was trying to speak his mind in relationships but kept holding back. Seven years of meditation, 150 hours of IFS therapy, and multiple retreats gave him useful skills but he still held back. After one afternoon of unlearning, he began expressing romantic attraction easily. After three more, he reported naturally disagreeing with colleagues, being direct with his mom, and texting friends “I miss you”, even though we never discussed/Goodharted on these. Video interview. Within months he asked out a girl he liked and entered a happy relationship, which he’s now been in for over a year. This then freed his attention to join a lab, where he stood up for himself when someone tried to take credit for his work.
A researcher at a certain frontier lab wanted to share more of his takes in Slack, but held off when this involved disagreeing with colleagues he looked up to. After two evenings of unlearning in late 2024, he started sharing his takes, sometimes quite stubbornly until the other person got what he was saying. He’s confirmed this is still the case 18+ months later; he almost forgets what it was like to ever hold back. In this time he was also promoted to run his own team. (Reference iff same lab.)
1. Is what you’re trying to do limited by internal bottlenecks?
If you’d like to know whether your output is limited by internal bottlenecks, I offer internal bottleneck evals. You’ll learn how to recognize the effect of internal bottlenecks firsthand. Takes 1–2 meetings; SF / remote. [Update: This is nearing invite-only.]
Contact: Twitter DM or chris@chrislakin.com
“I definitely wouldn’t’ve done anything like this without being directed at it by you or spending money… It seems to have already materially improved my life.”
—Anthropic MTS, September 2026
2. Unlearning
To remove internal bottlenecks limiting output, I offer unlearning. Unlearning has grown out of my research into locally optimal strategies and tracking results years later.
Because payment is used to track results, payment is forbidden unless/until you feel genuinely satisfied with post-hoc outcomes.1
It is available after applicable evals.
“I got exactly the results I intended to.”
—OpenAI MTS (reference available on request)
Twitter @chrislakin | About Chris Lakin | References available upon request.
Similar to Paul Christiano’s Certificates of Impact.