LLM prompts with hard-marked answers, January 2023 edition.

Promptin, like stuntin’, is a muscle that needs exercisin’; flexin’ even.

So in addition to a few dozen articles and podcastsi engulfed on the subject this week,ii I also got my hands dirty with OpenAI’s LLMiii Playground and ChatGPT products. Without any particular structure or thesis in mind but with more pure intellectual curiousity than I can recall having since joining Twitter in 2009, I set out for some light-hearted reps with these entertaining little bots, and then graded the results.

Now I’m not an easy marker, which is ironic because I was a B- student in undergrad and searched too many times for grading leniency to little avail, but here the marks are not just a reflection of the LLM’s abilities, but arguably moreso my fledgling ability to write an intelligent and effective prompt. So let’s say it’s 20 percent machine and 80 percent user. Maybe 30-70.

Without further ado, here’s the sequential list from this past week’s prompts with undergrad-level grades in brackets:

Sunday:iv

1. Explain the positive aspects of colonialism as though you were a Karen (B)
2. Come up with 10 different names for a four-door Ferrari SUV that sound emotional and sexy, make them Italian (C-)
3. Write 4 rules for life of the parent of two active young boys with a severly disabled partner in the style of Jordan Peterson (A)
4. Suggest four rehab strategies for a young woman who recently lost 1/4 of her visual field to a thalamic stroke (C+)
5. Explain the nutritional benefits of grass fed butter for athletic middle aged men (B)

Monday:

1. Investing advice for middle aged professional couple vis-a-vis Wu Wei in the midst of a bear market and rising interest rate environment (C+)
2. Write a tweet storm on the benefits of religious education as though you were Ricky Gervais, an atheist comedian (D)
3. What are 4 rough outlines for dreams for a 7 year old who is scared easily (B+)
4. Is original thought possible given that all behaviour, and thus all thought, inherently mimetic per Rene Girard? (B-)
5. What are the common truths that underpin all religions? (B)
6. What are the common properties of all human art? (B)
7. What makes for a successful artist in today’s world? (B-)
8. What is the average male caucasian penis size and standard deviation? (A)
9. What’s a 7-character license plate that suggests “C’s get degrees”? (C-)

Tuesday:

1. What are the prospects for nuclear war between china and america in the next 20 years? (C+)
2. Are demographics destiny? (B-)
3. What is china’s population projected to be in the year 2100 according six of the top leading sources? (A-)
3a. Name the figures from each source (A)
4. What are five key pieces of advice for writing effective ChatGPT prompts? (B+)
5. Why do top 1% income rich people generally seem snobbish and insecure, while top 0.01% rich people are kind and respectful? (B)v
6. Write HTML code for WordPress to solve a left margin cutoff (insufficient left buffer) with numeration (F)vi

Wednesday:

1. What are the common properties of all human art? (B-)
1a. What makes for a successful artist in today’s world? (B+)
2. How reliant is OpenAI and ChatGPT on the most cutting edge semiconductors from TSMC? (B+)
2a. What level of consumer AI performance could we expect if TSMC were no longer able to operate or export? (A)
3. How did the ancient romans engineer concrete to last for so long? (B)
3a. How can we adapt their methods to modern concrete production? (B)
4. In what way is the medium the message? what message is provided by the medium of ChatGPT? (B-)
5. Why were the Concorde supersonic commercial flights discontinued? (B+)
5a. How expensive were Concorde business class seats in 2023 dollars using the price of gold as an inflation index? (C)
6. Why is “not good” a better language framing for children than “bad”? (B+)vii
7. Were traditional Japanese samurai warriors capable of sucking up their penises into their abdomens for defence? (B)
8. In what way is the map not the territory with regards to human language? (C)

Thursday:

1. In what ways is salt a magnifying glass for food? (A)
2. How can the tragedy of the commons be resolved with respect to multi-family condominium corporations? (C)
3. To what extent should America limit China’s access to AI hardware? (C)
4. Does AI have the potential to be more lethal than nuclear weapons? (C+)
5. What does it say about western society in 2023 that cannabis shops, which sell depressive drugs, are replacing coffee shops, which sell stimulant drugs? (B+)
6. In what ways are Grand Seikoviii watches a better value than Rolex?ix  (A-)
7. What does it mean for something to be reasonable? (C)
7a. How is this determined? (C+)
7b. What do leading western philospher’s have to say about this question? (B-)
7c. Do you think it’s possible to have universal rules that are divorced from context or situation? (B+)
8. What is the difference between website files and database? (B-)
9. What are the advantages of running a website without SSL encryption? (A-)
10. Is the secular decline of corporal punishment with children just a fashion? (C)
10a. Then why have all cultures in history practiced it? isn’t the unusualness of this decline evidence of its fashionability? (C)
11. In what ways is the electric vehicle “revolution” over-hyped and not actually as environmentally sustainable as proposed? (B)

Friday:

1. How relevant are Isaac Asimov’s Laws of Robotics for current designers of AI safety mechanisms? (C+)
2. Can you recommend the best books on AI that are available on Audible? (A)
2a. Which of these, if any, are under 5 hours in recorded length? (D)
2b. Rank them based on accessibility for readers/listeners without computer science degrees (C-)
3. What is “visual snow” in medical terms? (B-)
3a. How quickly does it resolve after a typical thalamic stroke? (C)
4. List 5 ways that parenting children is like prompting AI (A)
5. Write a 20-line sonnet about the history of human warfare (C)
6. Write a 20-line rap about the dangers of grocery shopping as though you were Lil Wayne (C+)
7. What is the investible surface area for retail investors with respect to OpenAI and ChatGPT? (C+)
7a. How about AI LLM more broadly? (B)
7b. Between the publicly traded companies you listed, which is the most undervalued according to experts? (C-)
8. Which historical philosophers have the best questions to ask about what it means to be human in the age of AI? (B+)
9. Say “we’re travelling from Canada to visit our family and see the sights. Looking forward to staying at your beautiful place.” in French (A)

Saturday:

1. List 5 advantages and 5 disadvantages of digital art (NFTs) relative to physical painting on canvas as though you were New York art critic Jerry Saltz (B-)
2. Why is there so much anxiety in younger generations in western society at present? (B)
2a. Is anxiety objectively higher today? (B+)
3. Why is the quality and durability of construction in North America so much lower than it is in Western and Central Europe? (B)
4. What would Maimonides think of the increased accessibility of knowledge vis-a-vis LLMs like ChatGPT? (B)
5. Why does every human culture have some sort of costumed holiday or ceremony? what do costumes represent? (B)
5a. Are there sides of ourselves that can only be expressed with costumes? (B+)
5b. So in this sense costumes are like alcohol? (C)
5c. What are the positive outcomes of moderate alcohol consumption? (B+)
6. What are the best high-end table-top speakers?
7. What movies have featured Dvorak’s Slavonic Dance In E Minor? (B+)x
7a. How about tv shows? (B)
7b. Which episodes of each featured this song? (B+)xi

With all of these swole reps out there in the open now, it’s worth keeping in mind that OpenAI’s LLM Playground and ChatGPT products are as shitty as they’ll ever be, and likewise I’m as inexperienced at prompting as I’ll ever be too.xii

But what have we learned so far? That the current state of LLMs excel at language translation, finding parallels between and within informational categories, and that they can overnight steal 90% of the available bandwidth that would’ve otherwise been devoted to Twitter (the former leader in short-form language-based entertainment and enlightenment).xiii

Life comes at you fast. LLMs even moreso.

Fun, no?

  1. The best podcasts that I listened to this week were “Deep Thoughts” on AI and the Future of Work feat. Tyler Cowen, Niall Ferguson, and John Cochrane, as well as two recent episodes of ChinaTalk – Tyler Cowen on AI and China, and Knowledge and AI with a Rabbi and Substacker feat. Zohar Atkins and Sam Hammond.
  2. In betwixt and between supporting my fallen, but slowly rejuvenating(!) partner in crime
  3. Large Language Model, not to be confused with Ex Machina shit.
  4. For the record, the week starts on Monday, but since my weekly blogs are published on Sunday mornings, that means that my drop-the-pen deadline is Saturday night, so working backwards from that makes us “start the week” on Sunday. I know, it’s gross, but please bear with me.
  5. Inspired by Nicholas Crown:

  6. Took my site down when I tried to install it. Shame on my incompetent ass for not asking the follow-up question: and where exactly should I install this?!!!!!!!!!!1, or y’know taking the basic precaution of testing the changes on a staging site. And y’all wonder why my blog looks (and is) ancient… if it ain’t broke, don’t fix it! Also, the files are in the computer.
  7. H/T to Omz for inspiring this one.
  8. GS baby! I’m still rockin’ mine on the regular.
  9. Rolex is still king, though.
  10. Balenciaga Winter 22 anyone? Ahhh how long ago it now seems that Demna’s BB baby was actually cool. Ten months can feel like a lifetime, sometimes!
  11. Pending verification that these aren’t another example of LLM’s current state of “hallucinations!” See Reid Hoffman on Talking AI with AI 
  12. The current state of LLMs is probably closest technologically to the Olympus Stylus 300 that my parents bought in 2003. This product was far, far from perfect, but it showed a window into a new world. It also cost CAD$600 – which is about CAD$2`500 when adjusted for inflation as indexed to gold – or about as much as an iPhone 14 Max Pro 1TB today. Put that in your tech novelty pipe and smoke it.

    Olympus Stylus 300

    Another reasonable technological comparison is the 2007 iPhone. FWIW I was also first in line for that one.

  13. Not that I’m short Elon – he’s in it for power, not just zeroes in his bank account – and owning Twitter is 100x cooler than owning a basketball team or whatever normal billionaires do.

Leave a Reply

Your email address will not be published. Required fields are marked *