1866: Transatlantic telegraph charged £20 per 20 words. Clerks invented compression shorthand to slash costs.
1951: Claude Shannon theorized an "identical twin" predictor that could reconstruct messages from partial input—foundation of modern compression.
2024: New benchmark shows that same telegraph-era shorthand dialect (still embedded in LLM training data) cuts AI token output by 40-49% across multiple models.
Why it works: LLMs trained on historical text recognize the pattern. Shorter prompts = fewer tokens billed = direct cost savings.
Practical hack: Use telegraph-style compression in prompts for cheaper API calls without sacrificing accuracy. Old-school efficiency meets modern inference economics.
1951: Claude Shannon theorized an "identical twin" predictor that could reconstruct messages from partial input—foundation of modern compression.
2024: New benchmark shows that same telegraph-era shorthand dialect (still embedded in LLM training data) cuts AI token output by 40-49% across multiple models.
Why it works: LLMs trained on historical text recognize the pattern. Shorter prompts = fewer tokens billed = direct cost savings.
Practical hack: Use telegraph-style compression in prompts for cheaper API calls without sacrificing accuracy. Old-school efficiency meets modern inference economics.