Spin

Can a dolphin learn to read?

It is a dispiriting prospect.

Chimpanzees, our closest relative, are unable to speak due to fundamental differences in the structure of the throat. Surely a relative limiting factor exists in a brain of the dolphin?

Extraordinary claims demand literate cetaceans. I do not believe I could be convinced were someone to claim they had educated a dolphin, unless they produced it.

We seem to have wagered that LLMs represent a fully tractable pool of human knowledge and judgment, or, at the absolute minimum, a viable pathway to it. We assume that we are capturing knowledge in a meaningful sense, yet fail to grasp that each use is in effect a mirror for the user.

For chatbots, each response is curated, tempered, and unfailingly geared to mimic an interaction. Embedded and pseudo-deterministic applications present a deeper challenge, but the question remains the same.

Can we meaningfully crystallize and apply the human experience to our collective benefit? Can we design a dolphin intelligent enough to read? What happens when we shift to a dolphin-based economy?

There will come a time to ante up. The cost of this will not be what is expected: neither fungible nor mintable, but through application spendable.

Bushwhacked

Studies performed on the consequences of introducing LLMs to teaching have produced disparate results.

In many cases, diminished outcomes across learning have been reported [1]; other approaches yielded uneven, heterogeneous boosts to general learning [2].

It is the distribution of this boost that concerns me. 

Recent investigation 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].

For those 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 tool with specialised vocabulary, structured syntax, and qualitative intuition will define the depth of an interaction for an individual.

Commercial models will not use words a user is judged unlikely to understand. Pedagogical AI will be taken in strange, delirious directions before the dust settles. For better or worse, it is a tool with fundamental potential… and an impossible temptation.

Delinquent

That LLMs continue to find strange new methods of immiseration is unsurprising. We are already seeing passive degradation, particularly in emerging observations on the prevailing habits of users.

The ‘solipsistic reader-writer’, a proposed user classification, generates and consumes fiction within a closed conversational loop, interacting with a machine rather than a human other [4]. Of 573,453 conversations randomly sampled from ChatGPT, more than a third were dedicated to fiction.

This may not sound so terrible: general reading rates are perceived to be perpetually declining. Yet inspecting these tells us fully 50% were dedicated to fanfiction, and 27% to explicit content.

I was born a couple of years before the millennium. Among my peers, I have found many have taken time to examine the strangeness felt at the proliferation of digital life during our childhoods. It is a curiosity; it was a luxury.

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

Now, in the depth of great delusion, I pray it could be the last laugh for the literature major. But most will opt for the path of least resistance.

They will opt for porn, ease of use, the trust of business, and all the unwit consequence.