Automating bullshit - OpenAI ChatGPT removes office worker toil
If ChatGPT generates bullshit, perhaps we should eliminate the bullshit we're generating
Current status: there’s about two weeks left before the work year is over. I’m lucky that I get a lot of time off at the end of year. VMware gives a whole week. The idea of the world being shutdown and, thus, feel guilt free about doing nothing is more appealing than it’s ever been. I’m going to try a slightly new format this issue: “less.”
Automating bullshit - OpenAI ChatGPT removes office worker toil
Well, it’s actually really amazing. The OpenAI ChatGPT is a great toil for eliminating office worker toil. Toil is a concept that comes from Google’s Site Reliability Engineering methodology:
Toil is not just “work I don’t like to do.” It’s also not simply equivalent to administrative chores or grungy work. Preferences as to what types of work are satisfying and enjoyable vary from person to person, and some people even enjoy manual, repetitive work… So what is toil? Toil is the kind of work tied to running a production service that tends to be manual, repetitive, automatable, tactical, devoid of enduring value, and that scales linearly as a service grows. -“Eliminating Toil,” the SRE book.
The OpenAI chat thing is impressive. It has several immediate uses:
Generating - er, writing - all those bureaucratic emails one needs in large organizations. The effort it automates away is writing a good email. Think of how you’d upgrade a long Slack message or thread to a more formal email.
Writing quick and dirty corporate memos.
Putting together a rudimentary business case to at least start a discussion about big enterprise purchases.
Daily toil. You can imagine asking ChatGPT to “order more printer ink and ask for a discount because the last batch had two bad toner cartrages” and it’d generate a human email that gets sent to your local printer sales person. And then you get a response that’s like “we can’t give you a discount” and you reply back “write a reply that says we’re going with a different supplier,” and so on.
Writing those “I think our policy is bullshit” emails, but like, polite about it. Many people are incapable (or unwilling?) to take the supplicant tone needed to point out the obvious, usually that some corporate policy needs to change. They could use help.
Writing filler articles, blog posts, website copy. You might think this stuff is lazy, wrong - even immoral! - but I mean…you gotta pay the bills and get money to buy your growing kids new pants in the winter. Also…it sort of make a distressing statement about how we think about and talk about tech.
In other words, if there’s “bullshit” work you need to do that can be done in text, it’s very good for that. It may not be the end text you use, but it’ll give you start, launching beyond the blank screen. It’ll also help you set the tone and structure you want.
Little Videos
It’s also pretty good at generating short scripts for, like, little videos. I tried two of these: one on how product managers work with Backlogs, and one of golden paths.
The first was really good for a “I want to make a viral video in 30 minutes” sort of thing. This is exactly the kind of video you’re supposed to make to build up subscribers, blah blah.
The second wasn’t exactly right. It didn’t talk about Spotify golden paths, but it did a pretty good definition of generic build pipelines, CI/CD stuff.
It’s worth noting that the ~24 hour returns on these videos are not great. But the point here is that with the ChatGPT thing you’d have time to make, like, 4 or 5 of these a day, find the ones that work, tuning them. And, you could re-use them over different channels and repost them over time (hello Instagram!!!). Just doing two isn’t enough: you’d need, like, fifty or a hundred over a month.
Other uses
I asked it to write an email asking permission to take my kids out of school for Thanksgiving. You could tune this into all sorts of things. What it nails is the tone you need and a good persuasive structure…and thew basic typing of writing the email.
I asked it to write a drive-by corporate strategy memo. I had to do a lot of these when I worked in corporate strategy: an executive gets some fun notion (usually a company to acquire) and wants a memo to send to their boss “in thirty minutes.” The result here wasn’t perfect at all, but it was an amazing start. With subject matter expertise, you could spend about fifteen more minutes and have something good enough. If the other OpenAI stuff could generate Mekko charts for you - chef’s kiss!
You can see a comical version of this in the other direction: a memo from the board chastising the executive team’s strategic planning.
School
Early on in school, I had some kind of learning disability. It was the early 80s, so who really knows what it was? I was given extra time with a teacher who would help me come up with stories and write them down for me. Slowly, I started doing the writing.
I haven’t done much of the interactive chat with the AI, but I’m curious to use it to model learning fundamentals like the five paragraph essay for my kids.
I’ve done this with my son in on my own (that is, I’m the AI)…and…I am not trained as a teacher, I have no idea how to do it and it drives me crazy.
This isn’t exactly that, but you can see how it could be the start of teaching writing by modeling. The point wouldn’t be to turn this in and be done, the point would be to understand the mechanics of writing at a basic level. And, I’d theorize, that if you chatted with the AI, asking it questions about the text, asking it change it and saw the changes it made, you’d learn a lot about writing.
I mean, I took whole writing classes in college that were just that.
Jevons Paradox applied to office workers
There is a lot - a lot - of low-value text like this in corporate life. Automating that text will make office workers much, much more productive. It’s easy think this means they’ll be fired, but what we see in the IT space is that when you automate toil, people just move to different tasks, they “do more with less.”
It’s very funny for a professional writer to admit that his job is to come up with some ideas and then encase them in low-value bullshit that a bot could generate, but I do think he’s right, and that generating necessary bullshit is the worst part of many jobs.
There’s a scene in the Max Headroom show where two lawyers are arguing a criminal case. But, the two lawyers are just floppy disks, maybe zip disks (it was the 80s!). You load them up, some lights flash, and verdict delivered. Clearly, that is scary.
However, most of office life is low stakes bullshit: why not have two floppy disks talk? Then the humans can do the more valuable thing: imagine new things, decide between two impossible/unknowing options, and maybe, just maybe…leave work early to go do something more fun than write more corporate bullshit text.
Three Links That Are Relevant to Your Interests
What if failure is the plan? - Some wonderful 00’s internet history and even better thinking about long term (corporate) strategy. Also, there’s some worrying about what happens when a “public space” disappears, especially with regards to evil uses of what was once private data.
Great talk on the platform engineering, platform as a product, etc. from someone who’s done it at a large organization for several years, Bryan Ross. The community building part is especially tasty. (Also, check out more in platform internal marketing and community building.)
Maybe the Real Tech Debt was the Friends we Made Along the Way - New tech debt definitions just dropped: “According to Amundsen, we see evidence of technical debt when we have code/systems that can’t be easily observed or understood; according to Rapson, we see evidence of technical debt when we have slow development and release velocity.”
Here’s a Thing I Made That I’m Excited to Share With You
The talk I gave on our application modernization practice:
You should get and read the paper Marc and I wrote. It’s good and helpful, and I should know, I was there when I we were writing it!