Conversation

Replying to
There’s a potential for a network effect here, and good prompt patterns driving out weak ones. And of course, a clever architecture may allow AIs to discover good prompt patterns to. PromptGAN?
1
20
I have a fair amount of college teaching experience. Enough to know I’m not naturally talented at this. Once I know something, I find it hard to put myself in the shoes (motherboards?) of someone who does not know. But when I’m in a patient mood I can sort of do it okay.
3
13
Some connections from an experienced educator
Quote Tweet
Replying to @vgr
On the off-chance you aren't familiar, I think there are nice connections between this and (a) Perice's notion of "continuity" in learning, (b) Piaget's assimilation/accommodation loop, and (c) a wide variety of perspectives on the power of REPLs (e.g. vimeo.com/223309989)
1
15
Stub to think about debugging and decompiling
Quote Tweet
Replying to @aresnick and @vgr
(I also find it interesting that in many cases, you can think of debugging as decompiling a behavior, i.e. something of a dual/inverse of the "step by step" construction.)
2
9
I’m registering an Official Interest™ in this stebystepism trend so tag me if you spot anything interesting that’s related. It’s also the single most positive development I’ve seen in the zeitgeist in a couple of years. Step-by-step is how you work your way out of shit times.
5
31
Threshold concepts
Quote Tweet
Replying to @vgr
I have a theory in this. I’m very fond of the notion of “threshold concepts”. It’s the idea that there are crucial concepts in any discipline that have a paradigm shifting impact on the mind. They’re take you over a threshold that changes your perspective.
Image
1
19
Replying to
There is something here about growth rates. Stepbystepism is slower initially but accelerates to big momentum if it works. Uncritical trial and error is fast initially but either plateaus or crashes. And just vibing in one-shot ways just finds clever intutions without growing.
1
17
Zooming out to extreme macro, the growth vs. degrowth may be ill-posed. What if step-by-step is the DNA of tortoise like growth that doesn't backslide or crash as much? What if net, the tortoise IS faster than the hare?
4
12
Hmm... now that I think of it, my "sparring partner" approach is a step-by-step alternative to trying to "do" one-shot strategy. Instead of a fully-formed strategy being designed out of 2 days of leadership retreat BS, it emerges gradually, contour by contour
3
7
In any single sparring conversation, you might take 1-2 steps at most, but they add up over time. The strategy "takes shape" at the rate of actual alpha being uncovered in the situation, like a thing gradually revealing its full structure.
1
8
I'm unhappy with the term stepbystepism. I need to know the True Name of this thing. Possibly asynchronous sequentialism, as an antidote to the growing synchronous parallelism of the last few decades.
12
6
Oh shot just figured out the True Name of this thing. It’s not step-by-step, it’s loop-by-loop! Or better still, loopchaining. You don’t move to the next loop until current loop gets to 99.9% and you retest all atomic and subchain loops each time you extend the loopchain.
2
12
This is easily NP complete. To add loop n+1, you have to requalify minimum n loops again and max 2^n loops. If you don’t have nonlinear payoffs and/or “loop fusion” where substrings chunk-up somehow, it’s an intractable way to learn.
1
3
Worse. Each loop is an OODA loop, and the chain only works when the orientations are aligned. Which means potentially n^2 or higher misalignment modes. And this is under non-adversarial supervision.
3
1