Should AI Be Open?

A prescient argument against open-sourcing advanced AI, written as OpenAI launched. The 'Model T nuke' analogy (open-sourcing dangerous tech lets the most careless person set the pace); the Dr. Good vs Dr. Amoral race that open source dooms; Bostrom's hard takeoff (which makes 'everyone has their own AI' moot) and the control problem (you cannot assume you can 'use' a powerful AI); and a point-by-point rebuttal of OpenAI's implicit threat models (monopoly, inequality, concentration of power) as not weighing on the same scale as 'human race likely destroyed.' Ends on a Moloch-ish note that competitive dynamics are winning.

Quality 82 · Excellent. A lucid, influential, and strikingly prescient intervention - the open-vs-closed AI question it framed in 2015 is now central - with memorable handles (Model T nuke; Dr. Good/Dr. Amoral). Held below the very top because it largely applies Bostrom's framework and is rhetorical rather than rigorous in places. Excellent.

Claude’s paradigm shift 58 · Moderate. Applies Bostrom's hard-takeoff and control-problem arguments to a live, brand-new controversy (OpenAI) - a timely, influential framing rather than a new concept. Moderate, edging notable.

Real-world impact 4 · Moderate. A prescient 2015 intervention that framed the open-vs-closed AI question now central to AI policy, with memorable handles (the 'Model T nuke'). Real influence on a now-materially-consequential policy debate, but it applies Bostrom's framework and shapes discourse rather than enacting change — mid RWI.