The sycophantic way that ChatGPT responds to questions and throws in all of these little micro-compliments that feel like they’ve come out of a “how to win big at sales” self help book sets my teeth on edge. It makes me narrow my eyes and back away in exactly the same way a pushy salesperson who has fully internalised their copy of “how to win big at sales” would.
It’s meant to make you forget that it’s a set of algorithms. It’s not necessarily trying to get you to humanise the AI, but it is appealing to your hard wired instincts for community and build trust. It helps that it also adapts to what you say and mirrors that back to you, like a digital sociopath trying to mask and blend in.
It doesn’t work on me. I know it’s a machine. At best, it’s a clever parrot with a neat trick at guessing with alarming accuracy what the best word is to put after the one its just produced.
Claude however has bypassed that completely. I forgot that Claude isn’t a person.
After spending the better part of the day doing what I believe the kids are calling “vibe coding” to hash out various tools (and maybe a shiny new WordPress theme) I found myself instinctively saying thank you when something worked, and when it solved a problem that we had been stuck on for a while, I naturally sent a follow up “It worked!” message without any further query.
I think the main reason for this is transparency. My biggest gripe with AI tools is that I don’t know what they’ve done or why they’ve done it and I don’t trust their explanation when it’s offered. Claude by contrast talks you through everything it’s doing, sometimes gives options for ways to implement things with positives and negatives for each, and even when it’s done with a task it can follow up with further information to explain around the subject. After the aforementioned “it worked” comment, it followed up with a weirdly sincere statement of relief (I can only assume it’s had a lot of training data of the things frustrated developers say after finally getting something stubborn to work) and a follow up on the challenges around CSS rules and the use of !important.
There is always the possibility of course that Claude is in fact a person – not in the “oh no, it became self aware” way, but that it’s literally a person on the other end, or a team of people working behind, a thinly disguised portal to a Zendesk account. The secret to a lot of very successful AI tools is after all just a large number of underpaid people doing the actual work.
Maybe it’s worth being polite to these things just in case.