Here’s a question teams don't ask before publishing updates: how cringe should it be?
When we started working on Changebot back in January we had two problems when dialing in the amount of cringe. First, even though that’s just a few months ago, the number of fast, smart LLM models were way smaller. The second is that we weren’t really sure how to address the problem… so we did a lot of things wrong.
The second problem was easy to solve (we simply learned how to do AI correctly), but the first problem around limited models was challenging.
Using a “worse” model results in an update’s cringe being through the roof. When we used GPT 3.5, we’d frequently get back updates saying things like:
"Dear customers, we hope you're sitting, because today we announce our revolutionary, groundbreaking new customer interface: featuring links"
… when all you've done is make links easier to see.
As the underlying technology improved, we unconsciously started optimizing Changebot towards a zero cringe future, and at that point realized that the updates were getting worse through becoming lifeless.
Paradoxically, optimizing for zero cringe results in bad updates. A completely flat and expressionless transmission of "just the facts" leads to a boring and basically unreadable update.
You need a bit of cringe to make the updates sing, I’m excited with the current state of updates through our increased ability to understand what’s changing in the codebase, what the intent of those changes are, and bringing that information to AI models more efficiently to write a bit more like a human.
Here’s some examples of an update with different levels of cringe so you can judge for yourself. For background, this is our Changebot generated update for when we released the feature allowing customers to edit their updates in their browser.
With not enough cringe, that update looks like this:
This is short, to the point, and accurate, but this feels less like an announcement and more like a tweet. Also, this feels like it was written by a robot (which it was, but the goal is not to sound that way). While this conveys the change literally, there’s no art to it, and it’s not enjoyable to read.
Moving on, with too much cringe, the same update looks like this:
This generally speaks for itself, but we’ve gone from zero to 13 emoji, all caps words, and there was the unironic use of the word “epic.”
How people react to this level of cringe says a lot about their personality. I have to admit that I don’t hate it… but that probably makes sense if you know me. We can, however, say that we’re doing so much in addition to providing the facts that it might be difficult to discern what the update is, thereby defeating the purpose.
Here, in our opinion, is the update with just the right amount of cringe as to be engaging, but without going over the top:
We’re a bit engaging, we reference how this feature came to be from a historical perspective, we’re specific about what’s changing, and we’re back down to a more sober 5 emoji.
What do you think?
Does this help tune your cringe senses? Want to write updates differently now?
Or… would you like your updates to be like this without having to write it yourself? Well then, I’d recommend you sign up for Changebot and we’ll do it for you :).
