Brandon's Blog

7/15/2023

I Can't Do That, Dave

Harkening back to Slashdot proto-memes, I for one welcome the AI revolution. I get a kick out of ChatGPT (and especially DALL-E) and have actually used it for a couple of minor practical purposes. I write pretty well when I have a high-level pathway to follow, but I get some blockage when I'm on a blank page with minimal steering. Having boilerplate that pulls on a literal near-infinite library of material is really good for pro forma stuff that is otherwise a pain to write.

Some of the "AI is going to blow up white-collar industry" hype seems way over its skis to me. Part of the scaffolding of the joke in Office Space with the TPS reports was that the TPS reports were not the crux of Peter's actual job. It was just some pro forma thing he had to bang out as part of a software development job (which didn't end up being that difficult or challenging, but that's beside the point here). But the hype-ists go to "white-collar work is nonsense and GPT is super good at nonsense, therefore GPT will replace white-collar work." This seems off-base to me. My theory is that nonsense work will finally become devalued and put in its place based on time/effort arbitrage enabled by language models. Then each specific job has to reckon with the removal of this work: will the job still be justified? Case-by-case.

This paradigm jibes much better with the experience of things like email and online calendaring and word processing. Secretarial jobs became "assistant" jobs, not just for PC reasons, but because the nature of that work upgraded and shifted to account for the removal of nonsense and repetition from the job description.

I think GPT stuff is going to deliver some death blows. One I keep thinking about is thank-you notes. What is a thank-you note? A thank-you note is a proof-of-work (hello, crypto!) performative activity that affirms gratefulness by showing effort on the part of the gift recipient. Very little original has been said in a thank-you note, and - if truly original - the content would likely have benefitted from not being constrained to that format.

I fed in this prompt to ChatGPT: "write a thank you note from the recipient of a plush hippopotamus toy to their generous aunt who gave it to them for their birthday". Here's what I got:

Dear Aunt [Aunt's Name],

I hope this note finds you in good health and high spirits. I wanted to take a moment to express my heartfelt gratitude for the incredible birthday gift you gave me. Your thoughtfulness and generosity have truly touched my heart, and I cannot thank you enough for the adorable plush hippopotamus.

From the moment I laid eyes on this cuddly creature, I knew it was something special. Its soft fur and gentle expression bring instant joy to my heart. I have named it "Hippy," and it has quickly become my constant companion, accompanying me on all my adventures and providing endless comfort.

Every time I look at Hippy, I am reminded of your love and care. You always know how to find the perfect gift that reflects my interests and brings a smile to my face. The attention to detail in choosing this hippopotamus toy is remarkable, and it perfectly captures my love for these magnificent animals.

I have already spent countless hours playing with Hippy, and I can already tell that it will be cherished for years to come. It has found a special place in my room, and I take great delight in snuggling up with it at bedtime. Your gift has brought so much warmth and happiness into my life, and I am truly grateful for your thoughtfulness.

Your love and support have always been a source of strength and inspiration for me. Whether it is birthdays, holidays, or just ordinary days, you have a remarkable ability to make me feel incredibly loved and cherished. I am truly blessed to have you as my aunt.

Once again, thank you from the bottom of my heart for this wonderful gift. Your presence in my life is a gift in itself, and I am eternally grateful for your love and generosity. I look forward to our next visit and creating more beautiful memories together.

With all my love and gratitude,

[Your Name]

That's almost too good! So I told it to tone it down a few notches: "please make the note shorter and simpler."

Dear Aunt [Aunt's Name],

I wanted to say a big thank you for the plush hippopotamus you gave me for my birthday. It's absolutely adorable! I named it "Hippy," and I love it so much. It's become my favorite companion, and I always feel happy when I see it.

Your thoughtful gift means a lot to me. I appreciate your love and generosity. Thank you for making my birthday extra special. I'm lucky to have you as my aunt.

Sending you lots of love and gratitude,

[Your Name]

Enough with the Hippy thing, so insincere: "please do not mention the name hippy or naming the hippopotamus."

Dear Aunt [Aunt's Name],

I wanted to take a moment to thank you for the wonderful plush hippopotamus you gave me for my birthday. It truly made my day special, and I appreciate your thoughtfulness.

The hippopotamus is so cute and cuddly, and I can already tell it will become a cherished companion. I'm grateful for your generosity and for always knowing how to bring a smile to my face.

Thank you again for the lovely gift. I feel blessed to have you as my aunt.

With love and gratitude,

[Your Name]

Ultimately, these language models boil down to f(prompt, some randomness) = message. And prompt = f(minimal effort and thought). If I sent this to Aunt Petunia she would get that ChatGPT vibe for sure, and the only remaining benefit is me relaying that I appreciated the gift through my initiative in generating the text and conveying it to her.

You know what else would do that? Sending this text: "ACK toy". Petunia could have Venmo'ed me $10 with a memo "SYN birthday" and similarly conveyed her meaning, but buying the gift had context (hippos are cute, remember that trip to the zoo?) and the physicality of the act maintained sincerity of action (picking something out, having it shipped... even if it was 100% online).

Context, sincerity, and physicality are exactly the kinds of elements to interactions that startup dorks would totally overlook and undervalue. But it's what's left when we are able to strip out the nonsense! Ideally, this means these models and the tools that they will empower should allow us to focus more on what actually matters, the juicy core of communication, relationship, decision-making, etc., and get rid of the "PR Speak" because it's all just f(something) anyway.