It's Levi-O-sa, not Levio-SA!

Limits lie not in language, but in the LLM’s themselves.

The "enchanted" aspects of our discussions of these chatbots are unavoidable.

The language is obvious. Craft your message with the right creativity and precision, and in some unmeasurable way, the master prompt smith is able to pull exactly what they need from the "statistical ether."

In some ways, the discourse around this puts the onus on us.
If only you're able to say the right things, speak the right phrases – you can get what you want. The capabilities of these LLMs are unquestioned. It almost reminds me of pickup culture from the early internet era.  The riches behind Ali Baba's magic cave, waiting to hear "Open Sesame." But it’s worth asking: What exactly are the riches awaiting us? The "blackboxed-ness" of LLMs encourages this mystique, both in their technical architecture and their literal characteristics.

Without the ability to completely articulate how these systems work, our language fills in the gaps. We envision possible capabilities locked away, waiting to be unearthed by the perfect incantation.

By Karan Paranganat,
Augmented with Claude Sonnet 3.5

Dec 2024