“When a wise man points at the moon the naïve examines the finger.”

— Confucius (?)

I was invited to a meeting on integration with Groq, the application specific integrated circuit (ASIC) for LLM inference that’s recently “bought” by Nvidia to counter Google’s TPU. We barely fine-tune LLMs with general purpose GPU and I see zero need for the ASIC so Why Groq? The only explanation is it is yet another shiny vehicle which my coworkers cannot say no.

That got me thinking of the three different levels in which we operate:

  • Vehicle. The particulars, the hyped, the trees.
  • View. The underlining, the overall, the forest.
  • Vision. The goal, the value, the climate.

And the seemingly endless domains the vehicle-view-vision stratification applies to:

  • Work
  • Life
  • Leadership
  • Education
  • …

View

An uncrooked view of anything can only emerge from the above, the under and the around simultaneously:

  • Over-view. The thing in relation with other things. It’s place in the grand scheme of things. The breadth of knowing.
  • Under-the-hood-view, the depth of knowing, which may come from:
    • Formal Education.
    • Practical Experience.
  • Local-view. The local context around the thing. The situational knowing.

Interview

As an application of the 3V Stratification theory, whenever I was asked to interview someone, I gauge the candidate along the three dimensions below:

  • Vehicle. Does the candidate know the tools of the trade?
  • View. Does the candidate have a view of her own to the field?
    • Breadth. Does the candidate know the grand scheme of things in the field?
    • Depth. Does the candidate possess under-the-hood knowledge of the field?
  • Vision. Does the candidate has a vision which is contextualized to the local reality?
The 3V Stratification