The revolt of the reader

Reading holds deep significance for me, transcending mere leisure. The act of delving into lengthy prose has profoundly shaped my worldview over the years. Writing, too, is a cornerstone of my pursuits. It is not only a means to articulate my thoughts but also a rigorous exercise that compels me to refine and crystallize those ideas.

However, my words today arise from frustration: an increasing number of individuals — even those we admire and respect! — are publishing work that unmistakably bears the fingerprints of artificial intelligence. Readers, like myself, are left grappling with a barrage of questions for those promoting such AI-generated content. Do we lack the acumen to discern the LLM's influence, or are we indifferent to its presence?

The answer to the first question is resounding: we can indeed detect the hand of the LLM. To those who consume a wide array of content, the presence of AI is glaringly apparent, akin to the writer's intellectual integrity being compromised. In fact, the abruptness of this revelation is so jarring that I am compelled to believe that those wielding LLMs either lack the dedication to read extensively enough to spot these telltale signs, or they are not even scrutinizing their own work.

As for the second query, readers' concern is unequivocal. The AI-generated prose's peculiarities are so irksome that they trigger an internal alarm, prompting us to disengage from the text mid-sentence, thereby safeguarding our cognitive well-being. This response is not isolated to individuals; Cynthia Dunlop's survey of 668 developers revealed that 78% cease reading immediately upon detecting an LLM's involvement.

Moreover, a staggering 71% of these developers also vow to avoid the author in the future. A striking revelation is that these respondents prioritize authenticity above all else. Notably, 98% expressed a preference for an author's imperfect, manually crafted piece over a flawlessly polished LLM-generated one. This underscores the readers' sensitivity to the authenticity of the writing.

However, dismissing the respondents as a self-selected group may be too hasty. These readers, active on social media, are precisely the individuals most likely to share and endorse content they value — the early adopters and tastemakers of online discourse. The root cause of this reaction lies in the inability to discern between reality and fabrication.

When consuming AI-assisted writing, readers are left uncertain about the veracity of the content. As I posited in RFD 576, utilizing an LLM to create content effectively nullifies the social contract between writer and reader. Readers cannot expect to engage with a laboriously crafted sentence that the writer did not personally create.

The significance of this revolt against AI-generated writing cannot be overstated. When employing an LLM for public-facing material, we abandon our role as creators and fall into the passive role of mere disseminators. The primary objective becomes servicing the reader, rather than cultivating our own ideas. This approach will inevitably alienate readers and weaken our credibility.

Drawing a parallel with the historical battle against spam e-mails, we can foresee a similar trajectory for AI-generated writing. Early in the 2000s, the rampant spam crisis threatened the very fabric of e-mail communication. Yet, as spam filtering advanced and the stigma of spam increased, the practice became untenable. It is conceivable that a similar turning point awaits AI-generated writing.

Identifiable telltale signs of AI authorship, such as those detected by Pangram Labs' Pangram 3 and now improved in Pangram 4, can effectively deter the dissemination of such content. My recent adoption of Pangram 4, due to its superior accuracy, has instilled renewed confidence in combating AI-generated writing. Consequently, I have mandated its use for public writing on Oxide, effectively signaling a commitment to uphold the integrity of our communication channels.