The wave of DeepSeek takes reveal more about our own hopes and concerns than they do about DeepSeek.

One of my college professors, Donna Haraway, used the term “power objects” to define topics that we don’t fully understand, yet recognize to be important. As a result, we end up projecting onto these “power objects” our own concerns.

DNA was a power object. As were PET Scans, Big Data, the Internet, nuclear power, and so much more. Anything new we believe to be important while not understanding why it’s important is a “power object.” (At least until we understand it sufficiently to remove the mystique.)

DeepSeek is a textbook “power object,” with every hot take spotlighting a person’s desires and fears. Here’s some I’ve noticed:

  • To AI Boosters, DeepSeek proves LLM progress won’t slow down.
  • To AI Skeptics, DeepSeek proves AI companies have no moat.
  • To Open Source Advocates, DeepSeek proves open development is superior.
  • To AI Doomers, DeepSeek proves we aren’t being careful enough.
  • To Security Researchers, DeepSeek proves the risks of backdoors in models.
  • To Privacy Advocates, DeepSeek proves LLM services are vectors for collecting sensitive data.
  • To China Hawks, DeepSeek proves we need more US sanctions and investment in US champions.
  • To China Doves, DeepSeek proves sanctions don’t work and only inspire new innovation.
  • To NVIDIA Bears, DeepSeek proves we won’t need new data centers full of NVIDIA clusters.
  • To NVIDIA Bulls, DeepSeek proves we’ll need more GPUs for synthesizing training data and running inference.

…and so on!

Whenever these big waves of takes hit – and a wave of DeepSeek’s magnitude is rather rare – be aware of “power objects” and learn from the takes.