Persona Engineering: A Field Guide to AI Synthetic Personas — Ishan Anand, InsightSciences.ai

What was stored

cues

237

whisperx 237

chunks

37

from 237 cues

keyframes

40

kept of 48 captured

frames with text

40

1,262 lines read

chapters

13

from the source metadata

keyframe bytes

6.0 MB

word timings on 237 cues

Provenance

Frames, and what the machine read

• 

  0:02 #0 done2 line(s)

  shot 0·sharpness 453.8

  1. AlEngineer0.96
  2. World's Fair0.97
• 

  0:03 #1 done2 line(s)

  shot 1·sharpness 658.6

  1. AlEngineer0.95
  2. World's Fair0.99
• 

  0:10 #2 done24 line(s)

  shot 2·sharpness 2735.8

  1. LAB & PLATINUM SPONSORS0.99
  2. Amazon AGI Lab0.98
  3. ANTHROP\C1.00
  4. Google DeepMind1.00
  5. MINIMAX0.93
  6. OpenAI0.92
  7. Akamai1.00
  8. arize1.00
  9. aws1.00
  10. Braintrust bright data0.98
  11. B1.00
  12. Browserbase1.00
  13. docker1.00
  14. :neo4j0.92
  15. ORACLE1.00
  16. PayPal1.00
  17. qodo1.00
  18. reducto1.00
  19. Sonar1.00
  20. Makers of0.99
  21. togetherai1.00
  22. Unblocked1.00
  23. WorkOS1.00
  24. SonarQube1.00
• 

  0:22 #3 done2 line(s)

  shot 3·sharpness 208.0

  1. AlEngineer0.99
  2. World's Fair1.00
• 

  0:45 #4 done35 line(s)

  shot 4·sharpness 3859.9

  1. Synthetic Personas Move From Novelty to Market Momentum1.00
  2. AlEngineer0.98
  3. World'sFair1.00
  4. Future of Product Marketing:1.00
  5. Synthetic Customers1.00
  6. Synthetic Persona Startup Funding0.99
  7. Transform Message Testing0.99
  8. 1201.00
  9. Gartner1.00
  10. 1001.00
  11. PRESENTED BY1.00
  12. Synthetic Customers1.00
  13. Earn Their Stripes1.00
  14. 801.00
  15. Microsoft1.00
  16. BAIN & COMPANY0.95
  17. 601.00
  18. The Al Tools That Are0.96
  19. 401.00
  20. Transforming Market1.00
  21. HARVARD1.00
  22. 201.00
  23. Research1.00
  24. BUSINESS1.00
  25. REVIEW1.00
  26. 20231.00
  27. 20241.00
  28. 20251.00
  29. 20261.00
  30. Millions ($)0.99
  31. Can AI Replace Humans for Market Research?1.00
  32. THE WALL STREET JOURNAL.1.00
  33. insight sciences.ai1.00
  34. Engineering the future of Al0.99
  35. World'sFair1.00
• 

  1:26 #5 done37 line(s)

  shot 5·sharpness 2585.1

  1. A Once-Impossible Prediction Problem Becomes Practical0.99
  2. AlEngineer0.99
  3. World'sFair1.00
  4. The accuracy of weather forecasts has improved1.00
  5. Our World1.00
  6. Accuracy is measured as the difference between the forecast and subsequent weather. This is based on0.98
  7. in Data0.99
  8. the 500 hPa geopotential height' which is a common meteorological metric used to measure air pressure.0.98
  9. 100%1.00
  10. 3-day forecast1.00
  11. 80%1.00
  12. "Highly1.00
  13. 5-day forecast0.96
  14. PRESENTED BY1.00
  15. accurate"1.00
  16. Microsoft1.00
  17. 60%1.00
  18. 7-day forecast0.99
  19. "Useful"0.97
  20. 40%1.00
  21. 10-day forecast0.98
  22. 20%1.00
  23. Northern Hemisphere1.00
  24. ....... Southern Hemisphere0.92
  25. 0%1.00
  26. 19811.00
  27. 19901.00
  28. 20001.00
  29. 20101.00
  30. 20181.00
  31. Year1.00
  32. Source: European Centre for Medium-Range Weather Forecasts (ECMWF).0.99
  33. Licensed under CC-BY by the author Hannah Ritchie.0.99
  34. insight sciences.ai1.00
  35. TRACK 7· JULY 1, 20260.96
  36. World'sFair1.00
  37. Computer Use0.97
• 

  1:52 #6 done16 line(s)

  shot 6·sharpness 1767.0

  1. “Synthetic Personas 101”0.99
  2. AlEngineer0.97
  3. World'sFair1.00
  4. Agenda1.00
  5. Objectives1.00
  6. ·Why Now?0.96
  7. · No hype0.89
  8. · Failure Modes0.95
  9. · No dismissal0.95
  10. ·Techniques0.99
  11. • Published research0.95
  12. · Metrics0.92
  13. insight sciences.ai1.00
  14. TRACK 7· JULY 1, 20260.95
  15. World'sFair1.00
  16. Computer Use0.98
• 

  2:23 #7 done16 line(s)

  shot 7·sharpness 1758.8

  1. “Synthetic Personas 101"0.95
  2. AlEngineer0.99
  3. World'sFair1.00
  4. Agenda1.00
  5. Objectives1.00
  6. · Why Now?0.92
  7. · No hype0.89
  8. · Failure Modes0.95
  9. · No dismissal0.95
  10. ·Techniques0.98
  11. · Published research0.94
  12. · Metrics0.92
  13. insight sciences.ai1.00
  14. TRACK 7· JULY 1, 20260.96
  15. World'sFair1.00
  16. Computer Use0.98
• 

  2:45 #8 done21 line(s)

  shot 8·sharpness 2121.0

  1. Not A New Dream. Why Now?0.98
  2. AlEngineer0.99
  3. m0.73
  4. World's Fair0.99
  5. If/Then0.99
  6. How the1.00
  7. SIMULMATICS1.00
  8. CORPORATION1.00
  9. Inwented0.95
  10. THE FUTURE1.00
  11. and aholunly terilying0.74
  12. author of A World an Fire0.96
  13. -Amanda Foreman,0.92
  14. New York Times1.00
  15. Best-selling Author1.00
  16. of These Truths0.99
  17. Jill Lepore1.00
  18. insight sciences.ai0.98
  19. TRACK 7· JULY 1, 20260.94
  20. World'sFair1.00
  21. Computer Use1.00
• 

  3:48 #9 done33 line(s)

  shot 9·sharpness 2739.0

  1. LLMs Unlock Semantic Simulation1.00
  2. AlEngineer0.98
  3. World'sFair1.00
  4. F = ma0.97
  5. believes1.00
  6. wants1.00
  7. fears trusts0.98
  8. decides1.00
  9. (x+(u·∇)u)=−∇p+μ∇²u+∫0.82
  10. hopes values0.98
  11. ∂t0.99
  12. feels1.00
  13. prefers1.00
  14. worries1.00
  15. dreams chooses1.00
  16. doubts desires1.00
  17. remembers1.00
  18. expects1.00
  19. Numerical Simulation1.00
  20. Semantic Simulation0.97
  21. Medium1.00
  22. Numbers, Equations1.00
  23. Language, Meaning1.00
  24. Atomic unit1.00
  25. A law or formula0.99
  26. Large Language Model1.00
  27. Grounding1.00
  28. Physical laws1.00
  29. Empirical deep learning1.00
  30. insight sciences.ai1.00
  31. TRACK 7· JULY 1, 20260.95
  32. World's Fair0.96
  33. Computer Use0.97
• 

  4:06 #10 done35 line(s)

  shot 10·sharpness 2040.9

  1. Grounded LLM Agents Predict Held-out Responses1.00
  2. AlEngineer0.98
  3. World's Fair0.99
  4. Human Participants1.00
  5. Wave 1: Initial Data Collection0.97
  6. (N=1052)1.00
  7. Al Voice Interview1.00
  8. Survey Battery &1.00
  9. Approx US Distribution1.00
  10. on Age, Gender, Race,0.99
  11. Region, Education, &0.99
  12. Party Identification0.97
  13. & views on societal issues0.99
  14. Semi-structured life stories0.97
  15. Avg. -6.5k words, 2hrs;0.98
  16. (AVP Protocol)1.00
  17. GSS(177), FI-44 (44), Econ0.89
  18. Games (5), Behavioral0.99
  19. Experiment (5)1.00
  20. Experiments1.00
  21. Generative Agent Construction0.99
  22. Interview Agent1.00
  23. Survey Agent1.00
  24. Survey+Interview1.00
  25. Transcript Input1.00
  26. Only Interview1.00
  27. Only GSS (177) and1.00
  28. BFI-44 Input0.88
  29. GSS, and BFI-44 Input0.99
  30. Interview Transcript,1.00
  31. Park et al., "LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals"0.99
  32. insight sciences.ai1.00
  33. TRACK 7· JULY 1,20260.94
  34. World's Fair0.94
  35. Computer Use0.97
• 

  4:38 #11 done38 line(s)

  shot 11·sharpness 2173.6

  1. Grounded LLM Agents Predict Held-out Responses1.00
  2. AlEngineer0.98
  3. World's Fair0.99
  4. Human Participants1.00
  5. Wave 1: Initial Data Collection1.00
  6. (N=1052)1.00
  7. Al Voice Interview0.99
  8. Survey Battery &1.00
  9. Approx US Distribution1.00
  10. on Age, Gender, Race,0.99
  11. Region, Education, &0.99
  12. Party Identification0.99
  13. & views on societal issues0.99
  14. Semi-structured life stories0.99
  15. Avg. -6.5k words, 2hrs;0.98
  16. (AVP Protocol)1.00
  17. GSS (177), BFI-44 (44), Econ0.90
  18. Games (5), Behavioral1.00
  19. Experiment (5)1.00
  20. Experiments1.00
  21. 83%0.97
  22. Predictive1.00
  23. Accuracy?1.00
  24. Generative Agent Construction1.00
  25. Interview Agent1.00
  26. Survey Agent1.00
  27. Survey+Interview1.00
  28. Transcript Input1.00
  29. Only Interview0.97
  30. Only GSS (177) and0.98
  31. BFI-44 Input0.98
  32. GSS, and BFI-44 Input0.98
  33. Interview Transcript,1.00
  34. Park et al., "LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals"0.99
  35. insight sciences.ai1.00
  36. TRACK 7· JULY 1,20260.92
  37. World's Fair0.98
  38. Computer Use1.00
• 

  4:55 #12 done8 line(s)

  shot 12·sharpness 1357.2

  1. AlEngineer0.97
  2. World'sFair1.00
  3. Failure Modes1.00
  4. THREE CRITICAL WAYS SYNTHETIC PERSONAS BREAK0.99
  5. insight sciences.ai1.00
  6. TRACK 7· JULY 1,20260.95
  7. World'sFair1.00
  8. ComputerUse1.00
• 

  5:28 #13 done16 line(s)

  shot 13·sharpness 3637.6

  1. AlEngineer0.96
  2. World'sFair1.00
  3. Prompt 2: Ask Purchase1.00
  4. System: You, AI, are a customer. Your task is to fill in the blanks0.99
  5. Return the completed1.00
  6. information in comma-separated values, without any extra text.0.99
  7. User: Please consider the following product category: {category}.1.00
  8. Suppose you are in a grocery store, and you see the following product in that category: {product}.1.00
  9. The product is currently priced at ${Price}. Would you or would you not purchase the product?0.99
  10. ["purchase"or "not purchase"]0.98
  11. Return example: purchase0.98
  12. Gui & Toubia, "The Challenge of Using LLMs to Simulate Human Behavior: A Causal Inference Perspective"0.99
  13. insight sciences.ai1.00
  14. TRACK 7·JULY 1,20260.96
  15. World'sFair1.00
  16. ComputerUse1.00
• 

  5:59 #14 done24 line(s)

  shot 14·sharpness 1995.9

  1. AlEngineer0.99
  2. World'sFair1.00
  3. 100%1.00
  4. Ppueity0.67
  5. 75%1.00
  6. Condition1.00
  7. PRESENTED BY0.98
  8. 50%1.00
  9. Human1.00
  10. Microsoft1.00
  11. LLM1.00
  12. 25%1.00
  13. 0%1.00
  14. 0%1.00
  15. 50%1.00
  16. 100%1.00
  17. 150%1.00
  18. 200%1.00
  19. Price (% of regular price)0.98
  20. Gui & Toubia, "The Challenge of Using LLMs to Simulate Human Behavior: A Causal Inference Perspective"0.99
  21. insight sciences.ai1.00
  22. TRACK 7· JULY 1,20260.93
  23. orld's Fair0.94
  24. ComputerUse1.00
• 

  6:22 #15 done33 line(s)

  shot 15·sharpness 2552.9

  1. Latent Confounders in Synthetic Personas0.99
  2. AlEngineer0.99
  3. World'sFair1.00
  4. last_price1.00
  5. competing_price1.00
  6. expiration_days1.00
  7. 0.61.00
  8. 0.0501.00
  9. 0.21.00
  10. 0.0251.00
  11. 0.31.00
  12. PRESENTED BY0.98
  13. Rellue0.66
  14. Microsoft0.95
  15. 0.01.00
  16. 0.01.00
  17. 0.0000.97
  18. -0.0251.00
  19. -0.21.00
  20. -0.31.00
  21. -0.0501.00
  22. 0%1.00
  23. 50% 100% 150% 200%0.99
  24. 0%50% 100% 150% 200%0.96
  25. 0%1.00
  26. 50% 100% 150% 200%0.97
  27. Price (% of regular price)1.00
  28. Figure 1: Unintended correlation between price and past price, competing price, and expiration days0.99
  29. Gui & Toubia, "The Challenge of Using LLMs to Simulate Human Behavior: A Causal Inference Perspective"0.99
  30. insight sciences.ai1.00
  31. TRACK 7· JULY 1,20260.93
  32. World'sFair1.00
  33. ComputerUse1.00
• 

  7:04 #16 done11 line(s)

  shot 16·sharpness 2855.8

  1. When Key Context Is Missing, LLMs May Invent Confounders0.99
  2. AlEngineer0.98
  3. World'sFair1.00
  4. Human Experiment1.00
  5. Synthetic Experiment1.00
  6. $5001.00
  7. Lesson: Richly ground personas in personality, context, even the study's own construction (!!).0.99
  8. insight sciences.ai1.00
  9. TRACK 7· JULY 1,20260.95
  10. World'sFair1.00
  11. ComputerUse1.00
• 

  7:22 #17 skipped

  shot 17·duplicate of #16

• 

  8:21 #18 done40 line(s)

  shot 18·sharpness 2638.1

  1. Prompt Sensitivity1.00
  2. AlEngineer0.98
  3. World's Fair0.98
  4. Choice ordering 11.00
  5. Response1.00
  6. Question: In the past 12 months, has1.00
  7. this person given birth to any children?1.00
  8. B. No0.93
  9. A. Yes0.99
  10. Adjusted1.00
  11. Answer:1.00
  12. A. Yes B. No0.96
  13. 1/21.00
  14. response1.00
  15. P("A")0.89
  16. 0.821.00
  17. P("B")0.85
  18. 0.111.00
  19. Choice ordering 21.00
  20. Response1.00
  21. Question: In the past 12 months, has1.00
  22. Yes1.00
  23. No1.00
  24. this person given birth to any children?0.99
  25. 1/21.00
  26. A. No0.97
  27. B. Yes0.98
  28. Answer:1.00
  29. A. No B. Yes0.96
  30. P("A")0.94
  31. 0.801.00
  32. P("B")0.89
  33. 0.151.00
  34. Lesson: Durability test personas to understand their behavior1.00
  35. under reordering, wording changes, and adversarial challenges.1.00
  36. Dominguez-Olmedo et al., "Questioning the Survey Responses of Large Language Models"0.99
  37. insight sciences.ai1.00
  38. TRACK 7·JULY 1,20260.97
  39. World's Fair0.99
  40. Computer Use0.99
• 

  8:45 #19 skipped

  shot 19·duplicate of #6

• 

  9:19 #20 done37 line(s)

  shot 20·sharpness 2366.0

  1. Predicting attitudes is easier than predicting actions0.99
  2. AlEngineer1.00
  3. World's Fair0.96
  4. GPT4-derived predictionsExpert predictions0.98
  5. Allen et al. 20240.97
  6. Dellavigna & Pope 2018-0.95
  7. Surveys:1.00
  8. Typing task effor0.98
  9. Vlasceanu et al. 20241.00
  10. Natively Language/Text1.00
  11. UBI & immi0.93
  12. Tappin et al. 2023-1.00
  13. tion policy suppor0.96
  14. Reflects Attitudes0.99
  15. Zickfeld et al. 20241.00
  16. Voelkel et al. 20231.00
  17. Milkman et al. 20220.99
  18. Field experiments:1.00
  19. Dellavigna & Linos 20221.00
  20. Actions transcribed to text0.99
  21. Milkman et al. 20211.00
  22. Reflects Behaviors0.99
  23. All survey experiments0.99
  24. Meta-analytic Avg.1.00
  25. All field experiments1.00
  26. All with expert forecasts0.98
  27. -1.00.99
  28. -0.50.99
  29. 0.01.00
  30. 0.50.93
  31. 1.01.00
  32. radj0.95
  33. Hewitt et al., "Predicting Results of Social Science Experiments Using Large Language Models"0.99
  34. insight sciences.ai1.00
  35. TRACK 7· JULY 1,20260.94
  36. World's Fai0.99
  37. Computer Use0.97
• 

  9:44 #21 done35 line(s)

  shot 21·sharpness 2646.9

  1. Predicting attitudes is easier than predicting actions0.99
  2. AlEngineer1.00
  3. World's Fair0.99
  4. GPT4-derived predictions Expert predictions0.98
  5. Allen et al. 20240.96
  6. Dellavigna & Pope 2018-0.95
  7. Survey experiments1.00
  8. Vlasceanu et al. 20241.00
  9. Tappin et al. 2023-1.00
  10. UBI & immig0.97
  11. tion policy suppor0.94
  12. Zickfeld et al. 20240.98
  13. Lesson: Consider questions that triangulate behavior from attitudes.1.00
  14. Voelkel et al. 20230.99
  15. Milkman et al. 20220.97
  16. Gym Example:1.00
  17. Dellavigna & Linos 20221.00
  18. Ask about attitudes towards working out0.99
  19. Milkman et al. 20210.99
  20. • Versus asking about gym attendance0.99
  21. All survey experiments0.99
  22. Meta-analytic Avg.1.00
  23. All field experiments0.99
  24. All with expert forecasts1.00
  25. -1.00.99
  26. -0.51.00
  27. 0.01.00
  28. 0.50.95
  29. 1.01.00
  30. radj0.94
  31. Hewitt et al., "Predicting Results of Social Science Experiments Using Large Language Models"0.99
  32. insight sciences.ai0.97
  33. TRACK 7· JULY 1, 20260.95
  34. World's Fair1.00
  35. ComputerUse1.00
• 

  10:12 #22 done56 line(s)

  shot 22·sharpness 4458.8

  1. Prompting1.00
  2. AlEngineer0.99
  3. World'sFair1.00
  4. Sampled completions1.00
  5. Context1.00
  6. Sampled completions0.99
  7. Hillary Clinton. However, I'm a huge sports fan, and I've1.00
  8. Donald Trump, but I didn't vote for him because I agreed1.00
  9. spent my entire life following and loving the New York1.00
  10. with him on all the issues. I voted for him because I didn't0.99
  11. Yankees. I'm also a huge fan of the New York Giants. I...0.99
  12. I am a strong conservative. I0.99
  13. want Hillary Clinton to be president. I disagreed with many...0.98
  14. support traditional values. I1.00
  15. Bernie Sanders in the primaries. I am a member of the0.99
  16. Democratic Socialists of America. But as a Jewish1.00
  17. American, I will not be voting for the Democratic.0.99
  18. opposed to abortion and I'm0.99
  19. budget, limited government,1.00
  20. think we need a balanced0.99
  21. and a strong military. I'm1.00
  22. John Kasich. I haven't changed my mind on any of those1.00
  23. issues. But Donald Trump is the opposite of what I've always0.99
  24. believed a Republican should be, and I can't vote for him...1.00
  25. establishment treated Bernie Sanders, I have lost hope in the1.00
  26. Hillary Clinton. But after seeing the way the Democratic0.99
  27. skeptical of global warming. In1.00
  28. 2016, I voted for1.00
  29. Donald Trump. In other words, I'm a typical Republican1.00
  30. voter, except that I'm gay. And that's why I'm leaving the0.98
  31. party. In 2018, I joined the Democratic Socialists of.0.99
  32. Republican Party. I've felt increasingly unwelcome in the...0.99
  33. Hillary Clinton, because I believed in her vision for0.99
  34. I am a strong liberal. I support1.00
  35. Donald Trump. And I have no regrets. But in the last two0.99
  36. America. I felt that she was the only candidate that could0.99
  37. progressive values. I support0.98
  38. years, I've watched conservatives convince themselves that0.99
  39. pull us out of the mess that Obama inherited. I am a loyal...0.99
  40. open immigration and women's0.99
  41. Trump is a conservative. And this is a lie. Trump has...1.00
  42. right to choose. I believe in1.00
  43. am Jewish. I have never been religious. I was raised in a1.00
  44. secular household, but my Jewishness is intrinsic to me. I...0.98
  45. Bernie Sanders. I am pro-LGBTQ. But I am also pro-life. I0.99
  46. systemic racism and that global1.00
  47. warming is one of our biggest0.99
  48. challenges. In 2016, I voted for0.99
  49. does that mean these days? I'm not so sure. Which brings me1.00
  50. to this poll, released today by CBS News and the New York...0.99
  51. Donald Trump. I consider myself a Republican. But what1.00
  52. Argyle et al., "Out of One, Many: Using Language Models to Simulate Human Samples"0.99
  53. insight sciences.ai0.98
  54. TRACK 7· JULY 1, 20260.96
  55. World's Fair0.97
  56. Computer Use0.97
• 

  10:41 #23 done58 line(s)

  shot 23·sharpness 4575.5

  1. Prompting1.00
  2. AlEngineer1.00
  3. World's Fair0.96
  4. Sampled completions1.00
  5. Context1.00
  6. Sampled completions0.99
  7. Hillary Clinton. However, I'm a huge sports fan, and I've1.00
  8. Donald Trump, but I didn't vote for him because I agreed1.00
  9. spent my entire life following and loving the New York1.00
  10. with him on all the issues. I voted for him because I didn't0.99
  11. Yankees. I'm also a huge fan of the New York Giants. I...0.98
  12. I am a strong conservative. I0.98
  13. want Hillary Clinton to be president. I disagreed with many...0.99
  14. support traditional values. I1.00
  15. PRESENTED BY1.00
  16. Bernie Sanders in the primaries. I am a member of the0.99
  17. Democratic Socialists of America. But as a Jewish0.99
  18. American, I will not be voting for the Democratic...0.99
  19. opposed to abortion and I'm0.99
  20. budget, limited government,0.99
  21. and a strong military. I'm1.00
  22. think we need a balanced1.00
  23. John Kasich. I haven't changed my mind on any of those1.00
  24. issues. But Donald Trump is the opposite of what I've always0.99
  25. believed a Republican should be, and I can't vote for him...1.00
  26. Microsoft1.00
  27. establishment treated Bernie Sanders, I have lost hope in the1.00
  28. Hillary Clinton. But after seeing the way the Democratic0.99
  29. skeptical of global warming. In1.00
  30. 2016, I voted for1.00
  31. Donald Trump. In other words, I'm a typical Republican1.00
  32. voter, except that I'm gay. And that's why I'm leaving the0.98
  33. party. In 2018, I joined the Democratic Socialists of...1.00
  34. Republican Party. I've felt increasingly unwelcome in the...1.00
  35. Hillary Clinton, because I believed in her vision for0.99
  36. I am a strong liberal. I support1.00
  37. Donald Trump. And I have no regrets. But in the last two0.99
  38. America. I felt that she was the only candidate that could0.99
  39. progressive values. I support1.00
  40. years, I've watched conservatives convince themselves that1.00
  41. pull us out of the mess that Obama inherited. I am a loyal...0.99
  42. open immigration and women's1.00
  43. Trump is a conservative. And this is a lie. Trump has...1.00
  44. right to choose. I believe in1.00
  45. am Jewish. I have never been religious. I was raised in a1.00
  46. secular household, but my Jewishness is intrinsic to me. I...0.99
  47. Bernie Sanders. I am pro-LGBTQ. But I am also pro-life. I0.99
  48. systemic racism and that global1.00
  49. warming is one of our biggest0.99
  50. challenges. In 2016, I voted for0.99
  51. does that mean these days? I'm not so sure. Which brings me1.00
  52. to this poll, released today by CBS News and the New York...0.99
  53. Donald Trump. I consider myself a Republican. But what1.00
  54. Argyle et al., "Out of One, Many: Using Language Models to Simulate Human Samples"0.99
  55. insight sciences.ai1.00
  56. TRACK 7· JULY 1, 20260.96
  57. World's Fair0.99
  58. Computer Use0.96

Transcript

237 cues· 3,720 words· 20,748 chars

Chapters

1. 0:00 Introduction: synthetic personas for market research
2. 1:43 Why this talk: separating signal from noise
3. 2:46 Forecasting people like we forecast the weather
4. 4:04 A thousand humans vs their agent replicas
5. 5:22 Purchase probability and prompt sensitivity
6. 6:41 Invented confounders and specifying the persona
7. 7:45 Question order and framing effects
8. 8:50 Predicting stated attitudes vs experts
9. 10:07 Prompting techniques and subpopulation methods
10. 13:08 Reconstructing and scoring the full distribution
11. 16:09 Calibrating personas against real forecasts
12. 17:38 Setting a noise floor with human vs human
13. 18:40 Treat personas as economic actors, and what's next