Sunday Curations – Survive your Phd, Small Groups & more
Andrej Karpathy's Phd Survival Guide and other tidbits
As always, late on Sunday, here our favourite finds of the last week. If you want an escape from FTX and Twitter drama, this is for you.
Surviving your Phd / University
Two really good resources from Andrej Karpathy (ex-Director AI at Tesla) on the student and research life:
Really enjoyed James’ essay on Small Groups – check out his blog for more.
Lying somewhere between a club and a loosely defined set of friends, the SMALL GROUP is a repeated theme in the lives of the successful. Benjamin Franklin had the Junto Club, Tolkien and C.S. Lewis had The Inklings, Jobs and Wozniak had Homebrew. The Bloomsbury Group was integral to the success of Virginia Woolf, Clive Bell, and John Maynard Keynes, while MIT’s Model Railroad Club spawned much of modern hacker culture.
The SMALL GROUP must strike a delicate balance between indirection and real progress. Pure business drive is not desirable. The goal here is not to invest more in the skills you use at work. Instead, it is to be truly exploratory for no immediate purpose. It is to waste time (yet to savour it), to wander off in the wrong direction (and to find an exciting new path). Indirection and exploration should not come at the cost of doing and building. Doing and building should not come at the cost of having fun.
The Structure of Ill-Structured Solutions: Boundary Objects and Heterogeneous Distributed Problem Solving
A paper by Susan Star:
Understanding community goals means analyzing the problem of due process in open systems. Due process means incorporating different viewpoints for decision-making in a fair and flexible manner. It is the analog of the frame problem in artificial intelligence. From analyses of organizational problem solving in scientific communities, the paper derives the concept of boundary objects and suggests that this concept would be an appropriate data structure for distributed intelligence. Boundary objects are those objects that are plastic enough to be adaptable across multiple viewpoints, yet maintain continuity of identity. Four types of boundary objects are identified: depositories, ideal types, terrains with coincident boundaries, and forms.
Brings to mind Christopher Alexander's concept of centers and edges for a timeless way of building. With a pattern language, boundary objects can be emergent dynamic commons. Applied to artificial intelligence, this opens the possibility of a distributed AI based on a social metaphor rather then a psychological one.
Happy reading! And jump into the Discord to discuss.