The Environmental Cost of Training a Big Model

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-ethics-environmental-cost-builders

1. What is the core idea behind "The Environmental Cost of Training a Big Model"?

   1. Training a frontier model uses the electricity of a small city for months. Running inference at scale matches a large country's load. Here is what the numbers actually look like.
   2. Coordinate with peer customers when collective action helps
   3. Human oversight paths that do not require a credit card
   4. You can use AI to help someone else — like writing a kind message for a friend w…

2. Which term best describes a foundational idea in "The Environmental Cost of Training a Big Model"?

   1. inference
   2. training compute
   3. PUE
   4. data center

3. A learner studying The Environmental Cost of Training a Big Model would need to understand which concept?

   1. training compute
   2. PUE
   3. inference
   4. data center

4. Which of these is directly relevant to The Environmental Cost of Training a Big Model?

   1. training compute
   2. inference
   3. data center
   4. PUE

5. Which of the following is a key point about The Environmental Cost of Training a Big Model?

   1. GPT-3 (2020): estimated ~1,287 MWh to train, roughly 500 metric tons of CO2
   2. GPT-4 class (2023): publicly estimated at tens of thousands of MWh
   3. Frontier 2025 training runs: hundreds of GWh, matching small-city annual consumption
   4. Llama 3 70B: Meta disclosed ~1,900 MWh

6. Which of these does NOT belong in a discussion of The Environmental Cost of Training a Big Model?

   1. GPT-3 (2020): estimated ~1,287 MWh to train, roughly 500 metric tons of CO2
   2. Coordinate with peer customers when collective action helps
   3. Frontier 2025 training runs: hundreds of GWh, matching small-city annual consumption
   4. GPT-4 class (2023): publicly estimated at tens of thousands of MWh

7. Which statement is accurate regarding The Environmental Cost of Training a Big Model?

   1. Sparse mixture of experts: only activate parts of a model per query
   2. Custom silicon: TPUs, Trainium, Cerebras, Groq all target more ops per watt
   3. Model efficiency: smaller specialist models often beat giant ones per task
   4. Renewable siting: data centers are chasing hydro, geothermal, and nuclear

8. Which of these does NOT belong in a discussion of The Environmental Cost of Training a Big Model?

   1. Custom silicon: TPUs, Trainium, Cerebras, Groq all target more ops per watt
   2. Model efficiency: smaller specialist models often beat giant ones per task
   3. Sparse mixture of experts: only activate parts of a model per query
   4. Coordinate with peer customers when collective action helps

9. What is the key insight about "Per-query numbers" in the context of The Environmental Cost of Training a Big Model?

   1. A ChatGPT query uses roughly 10x the electricity of a Google search, at rough industry estimates.
   2. Coordinate with peer customers when collective action helps
   3. Human oversight paths that do not require a credit card
   4. You can use AI to help someone else — like writing a kind message for a friend w…

10. What is the key insight about "Reporting quality is poor" in the context of The Environmental Cost of Training a Big Model?

    1. Coordinate with peer customers when collective action helps
    2. Most labs disclose almost nothing about training energy. Public estimates are triangulated from GPU counts and runtime.
    3. Human oversight paths that do not require a credit card
    4. You can use AI to help someone else — like writing a kind message for a friend w…

11. What is the recommended tip about "Key insight" in the context of The Environmental Cost of Training a Big Model?

    1. Coordinate with peer customers when collective action helps
    2. Human oversight paths that do not require a credit card
    3. Training a frontier model uses the electricity of a small city for months.
    4. You can use AI to help someone else — like writing a kind message for a friend w…

12. Which statement accurately describes an aspect of The Environmental Cost of Training a Big Model?

    1. Coordinate with peer customers when collective action helps
    2. Human oversight paths that do not require a credit card
    3. You can use AI to help someone else — like writing a kind message for a friend w…
    4. Every token your favorite AI generates came out of a GPU that was drawing electricity from a grid that was, somewhere upstream, burning some…

13. What does working with The Environmental Cost of Training a Big Model typically involve?

    1. Training a model takes weeks or months and then stops. Running it serves billions of users every day for years.
    2. Coordinate with peer customers when collective action helps
    3. Human oversight paths that do not require a credit card
    4. You can use AI to help someone else — like writing a kind message for a friend w…

14. Which of the following is true about The Environmental Cost of Training a Big Model?

    1. Coordinate with peer customers when collective action helps
    2. Data centers are cooled by enormous amounts of water. A 2023 study estimated GPT-3 training consumed about 700,000 liters of fresh water.
    3. Human oversight paths that do not require a credit card
    4. You can use AI to help someone else — like writing a kind message for a friend w…

15. Which best describes the scope of "The Environmental Cost of Training a Big Model"?

    1. It is unrelated to ethics workflows
    2. It applies only to the opposite beginner tier
    3. It focuses on Training a frontier model uses the electricity of a small city for months. Running inference at scal
    4. It was deprecated in 2024 and no longer relevant