Ai deception: a plugin that mimics human errors – and why it matters
The digital world is now awash in a new layer of deception: a Chrome plugin, dubbed ‘Sinceerly,’ intentionally injects typos and grammatical errors into ai-generated text, mimicking the imperfections of human writing.
A calculated mess: introducing ‘sinceerly’
Developed using vibe coding – essentially, ai generating the code itself – Sinceerly leverages a model trained on common human writing mistakes. Think forgotten capitals, misplaced commas, and even deliberate substitutions of ‘sincerely’ with its misspelled cousin, ‘sinceerly.’ It’s a calculated attempt to blur the lines between machine and human authorship, a strategy with potentially serious implications.
The project originated with investor Ben Horwitz, who observed a curious trend: the more senior a company executive, the more riddled their emails became with errors. He utilized Claude Code to rapidly prototype the plugin – a testament to the speed of contemporary ai development. The result? An unsettlingly accurate simulation of a CEO’s digital slip-ups.
During testing, the plugin yielded shockingly realistic results. Four out of five CEOs contacted responded to emails crafted with Sinceerly’s ‘CEO’ mode – designed to amplify the most egregious errors – with terse, barely-there replies. Two contained typos, and one demanded a call from Ben himself, rather than the intended contact.
This isn't a harmless prank. The tool offers selectable levels of imperfection – ‘Slight,’ ‘Normal,’ and ‘CEO’ – ranging from subtle stylistic quirks to deliberately chaotic text. The underlying message is clear: ai-generated content, even when meticulously crafted, can be convincingly disguised as human work.
OpenAI is currently under investigation by Florida’s Attorney General for multiple charges, including accusations that ChatGPT, if treated as a person, would face a murder indictment. The implications are profound – suggesting that deploying ai tools without critical assessment carries inherent risks.
While the plugin’s primary use is likely to generate amusement, the underlying technology highlights a critical vulnerability. Current AI detection tools struggle to differentiate between genuinely flawed human writing and deliberately introduced errors. Using Sinceerly to deceive classmates or bypass editorial oversight is a fool’s errand—detection methods are improving rapidly. However, it significantly increases the chances of fooling less discerning readers, particularly those who may not utilize AI detection software.
Sinceerly is available for free download via its website, offering a limited daily word count. A paid subscription unlocks greater usage. But let this be a cautionary tale: the ease with which AI can mimic imperfection underscores the urgency of developing robust safeguards against manipulation in the digital realm. The race to detect and defend against AI-driven deception is only just beginning.