Bet

I sometimes worry that our hatred of generative AI is born out of a silent recognition that our minds are not as cherished and lovely as we once thought. 

If it is indeed so linked to our future, as certain titans have blithely designed, the effect epistemological thought can yield on their development is potentially unlimited.

A theoretically refined and practicable theory of knowledge could be applied to the densest form of the subject of its study we’ve yet created.

Call

There are hurdles.

Can we crystallize the human experience? If we cannot, what knowledge do we lose?

In what way do we know knowledge? Historically, the oral or printed delivery of human perspective in one form or another. All forms of mass-transmissible knowledge were human-made. 

Perhaps we can make an exception for the whales, who are still in need of one. Could a dolphin learn to read? It is a dispiriting prospect. 

We have wagered that LLMs represent a tractable distillation of human knowledge or, at the absolute minimum, a viable pathway to it. At this point, that is the course we will follow. 

There will come a time to ante up. With apologies to Mephistopheles, the cost of this will not be what is expected. It is not fungible or mintable, but it is spendable.

Bushwhacked

Studies performed on the consequences of introducing LLMs to pedagogy produce disparate results. Diminished outcomes across learning metrics have been reported in many cases [1]; other approaches have yielded extraordinarily heterogeneous general boosts to learning [2].

It is the distribution of this boost that concerns me. 

There are obvious points, frankly, over the effects of unequal access to apparently effective learning enhancement methods. More insidious implications lie in the vast differences in effective literacy rates between users. 

Lest I be accused of elitism: recent investigations of effective literacy rates in the US found that over a quarter of adults struggled with basic integration of information from short textual sources, while over 50% exhibited significant difficulty engaging with simple medium-length texts [3].

Unable or unwilling to accommodate the growing scale of our collective digital enterprise, the stratifying effect will be dramatic.

The ability to effectively query an interactive educational tool with specialised vocabulary, structured syntax, and qualitative intuition will define the depth of the mirror.

Sometimes, in delusion, I pray it could be the last laugh for the literature major.

Commercial models will not use words it judges a user not to be likely to understand. We are already seeing the beginning of this passive degradation in the language model solipsism many, myself included, have observed in our peers, friends, and families.  

Canticle

That LLMs continue to find strange new methods of immiseration is unsurprising. The question of intrinsic value will be raised in ways we are unable to fully grasp at this stage of our leveraged accelerando. They will compound the effect of digital illiteracy, a process that began before I was born.

I have the perspective of a 1997 birthday. Among my peers I have found, nearly universally, a desire to examine the strangeness we felt at the proliferation of digital life. 

A child can get used to anything. Or so I used to believe.