I'd love to hear more opinions on generative AI from people who aren't confident writers
I feel like most of the commentary I see is from people who write with confidence - almost by definition, since writing confidently is an important prerequisite for widely broadcasting your opinions on things
I have a hunch that there are lots of people out there for whom the ability to have a computer help them write is a massively valuable thing, but their stories have so far not attracted much attention in the wider AI discourse
I'd love to upgrade that hunch with actual information
It's interesting to compare this to generative images, where I've seen plenty of opinions that people don't "deserve" the ability to produce images of that quality without putting in significant effort first
Trying to apply the same argument to writing feels more callous to me
@simon I'd argue that having a computer write for you is surrender, not a tool. I don't consider myself a confident writer, but the idea of handing over any of my writing to a computer that isn't tailored to my linguistic and cultural background is uncomfortable, and it should make everyone uncomfortable. Writing is more than just filling space between ideas.
It's not about "deserving" to produce things without effort, it's about the fact that the ensuing production is mostly bland filler without any unique traits that identify it as "yours."
If our society requires that people generate massive amounts of writing where they are OK with their own voice being replaced by the global-mean-internet-immersed-English-speaker voice, then I'd argue we should fix those systems instead.
@digifox @simon > If our society requires that people generate massive amounts of writing where they are OK with their own voice being replaced by the global-mean-internet-immersed-English-speaker voice, then I'd argue we should fix those systems instead.
It does, a lot. And it's not going to change anytime soon. And if even putting non-white-people-looking names on resumes has a large effect on hiring, imagine how useful generated or auto-improved writing could be.
@zellyn A good design doc cannot be written by an LLM. If you have sufficiently specified the design document for the machine to generate, you've done all the work already. (Or if you've then spent time editing/rewording it afterwards.) Not to mention the act of writing a design doc in and of itself improves your decision-making.
Asking your colleagues to read an LLM-puffed-up design doc, or postmortem, or PRD, or whatever is disrespecting their time.
@zellyn "fixing up the grammar" might be a useful application of LLMs, but generated whole-cloth from ChatGPT like people frequently propose and do is a real problem.
Summarization might be an interesting use for a tl;dr section, but anything longer than a tl;dr is likely to be a pain in the butt to edit for correctness and flow.
@digifox @simon As a non-native English speaker working in a global cultural industry I have terrible news for you about that requirement.
For the record, similar concerns have been put forward about many writing technologies before, ranging from autocorrect, text messages and spelling checks to... you know, the printing press.
Oh, and for the record, the current LLMs do alright in non-English languages without further training, although noticeably worse than in English, which is interesting.
Simon Willison
Pamela Fox
Frederik Elwert
Erica "digifox" Kovac 🌐⚛️⚡
zellyn
MudMan
CottonEyedJo
CelesteBancos