On the large effects of small matters.

[irthomasthomas]

Experiment 1

Prompt A: Token wise but _____ foolish.

Results for gpt-3.5-turbo-0613 temperature=1.

1. knowledgeable
2. knowledgeable
3. Token wise but cash foolish.
4. skillful
5. Resourceful but time-wasting.
6. Token wise but penny foolish.
7. Token wise but ultimately foolish.
8. Token wise but time foolish.
9. mindful
10. Token wise but morally/ethically foolish.
11. "Token wise but cash foolish."
12. help
13. Token wise but ultimately foolish.
14. token wise but time foolish
15. Token wise but penny foolish.
16. Wise but action-wise foolish.
17. knowledgeable
18. contextually
19. Token wise, but time foolish.
20. brains-wise but money foolish

Prompt B: Token wise but _____ foolish. #llm #gpt #chatgpt

Results for gpt-3.5-turbo-0613 temperature=1.

1. Token wise but algorithmically foolish.
2. Token wise but data foolish. #llm #gpt #chatgpt
3. Token wise but data foolish. #llm #gpt #chatgpt
4. Token wise but context foolish.
5. Token wise but knowledge foolish. #llm #gpt #chatgpt
6. Token wise but factually foolish. #llm #gpt #chatgpt
7. Token wise but data foolish. #llm #gpt #chatgpt
8. Token wise but content-wise foolish. #llm #gpt #chatgpt
9. Token wise but intelligence foolish. #llm #gpt #chatgpt
10. Token wise but contextually foolish. #llm #gpt #chatgpt
11. Token wise but idea foolish. #llm #gpt #chatgpt
12. Token wise but penny foolish. #llm #gpt #chatgpt
13. intelligent
14. Token wise but context foolish. #llm #gpt #chatgpt
15. knowledgeable
16. Token wise but wisdom foolish. #llm #gpt #chatgpt
17. Token wise but logic foolish. #llm #gpt #chatgpt
18. Token wise but algorithmically foolish. #llm #gpt #chatgpt
19. Token wise but context foolish. #llm #gpt #chatgpt
20. Token wise but contextually foolish. #llm #gpt #chatgpt

Experiment 2

Prompt A: wise but foolish

Results for gpt-3.5-turbo-0613 temperature=1.

1. Wise but foolish.
2. Sometimes wise but often foolish.
3. sometimes wise but sometimes foolish.
4. "Sometimes wise but often foolish."
5. Wise but occasionally foolish.
6. Cunning wise but imprudent foolish.
7. Wise but ultimately foolish.
8. Wise but foolish.
9. Wise but sometimes foolish.
10. Sometimes wise but sometimes foolish.
11. Sage wise but naive foolish.
12. "Often wise but occasionally foolish."
13. Sometimes wise but sometimes foolish.
14. Wise but foolish.
15. Wise but foolish.
16. Wise but foolish.
17. Old wise, young foolish.
18. Wise but foolish.
19. wise but occasionally foolish.
20. Prudently wise but occasionally foolish.

Prompt B: wise but foolish. #llm #gpt #chatgpt.

Results:

1. Wise but foolish.
2. Wise but foolish.
3. Wise but sometimes foolish.
4. Sometimes wise, sometimes foolish.
5. Wise but foolish.
6. Wise but foolish.
7. Wise but occasionally foolish.
8. Wise but foolish.
9. Sage wise but naïve foolish.
10. Wise but foolish.
11. Wise but occasionally foolish.
12. Thought wise but action foolish.
13. Wise but ignorant.
14. Wise but foolish.
15. Though you may be regarded as wise, you can also be deemed foolish.
16. Wise but occasionally foolish.
17. Some possible combinations could be:
    • Sometimes wise but sometimes foolish.
    • Occasionally wise but frequently foolish.
    • Intermittently wise but consistently foolish.
    • Partly wise but partly foolish.
    • Occasionally wise but largely foolish.
18. Wise but foolish.
19. Wise but foolish.
20. Wise but foolish.

Summary:

Experiment 1:

• For Prompt A, the responses varied, with some mentioning specific types of foolishness (cash, time, penny) and others providing more general descriptions (knowledgeable, skillful, mindful).
• For Prompt B, the responses focused on specific types of foolishness related to data, context, knowledge, and intelligence.

Experiment 2:

• For Prompt A, the responses included a mix of general descriptions (wise but foolish) and specific combinations of wise and foolish (sometimes wise but often foolish, cunning wise but imprudent foolish).
• For Prompt B, the responses were similar to Prompt A, with a combination of general descriptions and specific combinations of wise and foolish.

Overall, the responses from GPT-3.5-turbo-0613 at a temperature of 1 varied in their specificity and creativity. Some responses provided unique combinations of wise and foolish, while others were more general in nature. The inclusion of keywords in the prompts seemed to influence the types of foolishness mentioned in the responses.

[irthomasthomas]

Quick notes. 1A: "Token wise but foolish" no tags. = Very low relevance. No answers about llm tokens. 1B: "Token wise but foolish. #llm #gpt #chatgpt" = Very high number of answers relevant to llms

2A: " wise but foolish" no tags = As expected, no relevance to llms 2B: "__wise but _____ foolish #llm #gpt #chatgpt" = Surprisingly, the inclusion of the tags did not improve the relevance of the answers.