The perfect prompt

In 1936, the foreword to Dale Carnegie's How to Win Friends and Influence People repeated a claim that would echo for the next eighty years: humans use only 10% of their brains. Misattributed to William James, parroted by motivational speakers, and cemented by Hollywood: Limitless (2011), Lucy (2014), the myth refused to die.

It was always wrong. Functional brain imaging has shown for decades that virtually every region of the brain is active across normal daily life. Damage to any small part causes measurable deficits. There is no idle 90%.

Turns out the myth described LLMs better than human brains. Today, we run the world's most capable inference engines at a fraction of their potential not because the models are limited, but because we ask them the wrong questions.

• You can always ask harder.
• You can always add more compute.
• You can always add more context.
• You can always add more window for context.

But a wrong question will never give you the right answer. A right question is not guaranteed to give you the right answer either. But more right questions give more right answers: that's the 90% still left to unlock.

We are a San Francisco-based seed-stage AI startup developing SOTA models for LLM input optimization. Our ML models remove redundant tokens from inputs, allowing for more inference, more context, and lower latencies.

Our goal is to build a universal optimization layer for all LLM inputs, allowing for smarter, longer, cheaper, and faster LLM inference. We are recruiting talented ML researchers and engineers to our San Francisco office. If you would like to join, please reach out below.

Not every token is made the same.