There are an increasing number of observations that LLMs are being used as a shortcut to quickly ingest information which falsely is perceived as knowledge. Furthermore, they are used as truth arbiters which further increases the vulnarilibity to knowledge generation and knowledge sharing. The extensive use of this technology in such a manner leads to convergence in thought, a homogenization, a soft authoritarianism, where diversity of thought is disregarded, cancled.
Relying on such models for education creates a single point of failure. Everyone will be outsorcing critical decisions to stochastic models. Historically, such restrictions were mostly enforced through excessive political force. Today, societies are welcoming such censorship voluntarily via the convenience of easy accessible information through these models.
Over the past decades, technical disciplines shifted from more formal education to a more informal, applied skills. Theory was sidelined as an overhead. However, in the current information-saturated landscape, education functions as an important mental model which helps sift signal from noise.
Without such a model, the practitioner struggles to grasp the output of an LLM, to validate the context. In a high information entropy environment, formal education gives the framework for deep expertise and only viable mechanism for indipendent thought.
When such framework already exists, one can use the LLMs as a map, a tool for navigation within the vast information space. They can quickly organize that information so that initial points for exploration can be marked. When treated as an index for a limited set of information, rather as an oracle, this technology can accelerate discovery of solutions to the problem one faces. Still, a map is not a lossless compression of the territory. Navigating such a lossy information terrain is left to the domain-specific and critical analysis of the person.
Knowledge acquisition is an energy-intensive process. It involves reading, cross-referencing, forgetting, re-reading, synthesizing different pieces of information. This is the mechanism available to us. It is not a bug that has to be overcome with quick information ingestion from an LLM. The latter should serve only as an enhancement to this process, not a substitution to cognitive friction. Bypassing such a friction leads to athropy of the expert’s ability to critically operate in a highly complex contexts, ultimately compromising his/her expertise.