5. Deep Research and contextualize

First, have both ChatGPT and Gemini generate a Deep Research report for you. To do this, make sure you select “deep research” (on ChatGPT, click the “+” in the text entry bar, on Gemini this is under “tools”).

Tell it what resource you want to learn about, and that you want to a prediction of how much of this research will be consumed worldwide by AI by 2030, as well as a comparison of how much of this resource is used by some other industry, like the Library of Congress or streaming video or cell phone service or golf courses or steel manufacturing or the military or something. (You might have one in mind, or you might have a brief conversation with it about how to think about other industries.)

It will probably ask you some follow-up questions, or propose a “research plan” that you should look at, and possibly edit, before it starts the research. Make sure that its plan includes finding a lower estimate and a higher estimate, as well as the main estimate. Depending on the resource you are asking about, you may want a global estimate, or an estimate for the United States, or an estimate that breaks out different regions separately. There may be other questions it asks you, and you should try to think intelligently about whether these are relevant or things you can tell it to ignore.

Once you tell it to start that, it may take 5 minutes or it may take 30 minutes. You might want to have both ChatGPT and Gemini working in different windows.

Once you have both reports, get Claude to help you interpret them. Upload both reports into Claude, and ask it to help you understand and contextualize them. See if they roughly agree with each other, or if there are different estimates or trends they are using. If you think anything seems weird or surprising, see if Claude can explain, or if it shares your surprise. Does it think the estimates given might be overestimates or underestimates, or is there a lot of uncertainty?

The numbers involved will probably be in the millions, billions, or higher. Ask Claude to help you understand these numbers, by comparing them to things like the total amount of that resource that is currently used, both in total, and by things you understand. Is this as much water as people use for drinking, or for golf courses, or for agriculture? Is this as much electricity as is used by a single factory, or for YouTube, or for streetlights, or for aluminum refining? Is this as much storage space as a large video game needs, or all of Wikipedia, or the whole Library of Congress?

Once you’re done, write a few sentences summarizing what you think are the most interesting results. In the file you submit, put your summary at the top. Then share the Claude conversation (either with a link, or by copy and paste). Also include the two Deep Research reports (again, either as links, or by copy and paste).

1. Amount of training data used for LLMs
2. Amount of electricity used for training LLMs
3. Amount of electricity used for running LLMs
4. Amount of computing power used for training LLMs
5. Amount of computing power spent running LLMs
6. Amount of money spent on training LLMs
7. Number of GPU chips (or other specialized chips) for LLMs
8. Amount of storage needed to hold the weights and biases of the largest trained LLM
9. Amount of water used for data centers running LLMs
10. Total amount of computing power available in the world
11. Total amount of cloud storage in the world
12. Total amount of renewable electricity generated worldwide
13. Total number of GPU chips (or other specialized chips) manufactured worldwide

Assignments 1 2 3 4 5 6 7 8 9 · Final project

Module 5 Back to AI Literacy