The claims of DeepSeek are highly credible, and should strike fear into the heart of anyone playing this bubble. AI Winter follows AI Summer. A serial “kill-chain” of risks can now be described, of which the failure of any one will lead to DeepFreeze of DeepWinter. The serial possibilities:

• Insolvency of investments in generation infrastructure.
• The  U.S. is no longer the top player.
• Projected benefits fail to materialize.
• Labor unrest, spreading as disenfranchisement proceeds apace. Luddites are inspired to make the last stand of humanity.
• A distinct category of unrest, caused by the use of AI by the unscrupulous to manipulate the information space.

The failure of any one item in the above list punctures the current projections. The multiplicative uncertainty of multiple factors may be sufficient.

We assume that frontier engines such as ChatGPT4 approximate human capabilities. In fact, they are clearly superior, in many categories of formal reasoning, to the average human. Rumor has it they may shortly exceed superior human intellects.

We seek to answer, in a napkin calc, whether the claims of DeepSeek are plausible.

• A center value estimate for the amount of electricity used to train ChatGPT4 is 5.7 * 10¹⁰ watt-hours.
• There are 8760 hours in a year.
• The brain develops anatomically till maturity at age 25, a space of 219,000 hours.
• The brain consumes about 25 watts of power = 25 watt-hours per hour.
• Over 25 years, the brain consumes roughly 5.47 * 10⁶ watt-hours of energy.
• The ratio of total training energy consumption, ChatGPT4 ÷ human = 5.7 * 10¹⁰ ÷5.47 * 10⁶, which is approximately

10,000 : 1

The energy reduction of DeepSeek is made plausible by what is possible.

• This ratio is approximately 2¹³.
• The time remaining until the Technological Singularity, posited by some for 2038, is 13 years. This implies an annual halving of power consumption per AI.

Anyone with skin in the game who has given this ratio consideration has sleepless nights worrying about being blind-sided, which is exactly what happened. Group-think takes care of the rest. Ironically, human nature suggests if the ratio of energy consumption were more like 100:1, it would get our attention. But for the quants who “know everything”, the number does not compute. Incomensurable numbers do not exist. Hence, it does not exist.

So why not copy the brain? Do we need to? What is the prospect for further disturbing surprises? Due to time constraints , this will be continued shortly.