Alexander Huso, a 31-year-old software tester from Salt Lake City, currently working in a hotel, harbors a deep passion for coding. Despite his coding proficiency, the absence of a formal degree makes it challenging to prove his skills in the industry, leading him to frequently leverage AI tools for coding assistance.
Driven by the goal of reducing output token costs for his Claude Pro subscription, Huso embarked on an unconventional experiment: he prompted Claude to communicate in "caveman speak." His hypothesis was that by forcing the AI to use simpler vocabulary and grammar, the generated output would have fewer characters, thereby saving tokens.
The experiment indeed yielded some token savings. However, this economy came at a significant cost to quality. Huso reported that the AI's output became severely degraded and largely unusable. "I wouldn't trust it to write any serious code," Huso stated, highlighting how this extreme token-saving strategy severely compromised the AI's practical utility.
Subsequently, Huso shared his experiment and its outcomes on Reddit. To his surprise, this peculiar endeavor rapidly gained widespread attention and went viral. This experience not only underscored the potential pitfalls of overzealous cost-saving in AI prompt engineering but also offered him a profound lesson in the mechanics of online content virality.
