3 ms·

> Management believes they can replace most of these employees with LLM-based automation In the writing world, too. I’ve yet to hear of attempts at it being an

> Management believes they can replace most of these employees with LLM-based automation In the writing world, too. I’ve yet to hear of attempts at it being anything but a wasteful distraction. Probably works fine if you don’t care about accuracy or quality (and some use cases for writing don’t! Spam, astroturfing, paid trolling, mass-message scams) but otherwise LLMs aren’t close to replacing 80% of your content department. Maybe 5% at best. But that doesn’t seem to have dampened enthusiasm for trying, yet.

thelastparadise 3y ago
> but otherwise LLMs aren’t close to replacing 80% of your content department. Maybe 5% at best. LLM alone, no. Well-engineered LLM-based distributed system with good prompting and input data? Absolutely.
- jay_kyburz 3y ago
  You just replaced a cheap writer with an expensive AI prompt and input data manager.
acdha 3y ago
One way to think about it: how good does it need to be for the managerial class types who are visibly salivating at mass layoffs to convert current professional jobs into 1099 “temporary” LLM quality control gigs? I’m betting we’re close to the level where some large places are going to try.
- vundercind 3y ago
  I do think it likely we’ll see a lot of organizations decide they just don’t care that much about (much of) their writing being any good. Like rushing to replace your first-line support with a chat LLM: sure it’s terrible, but you don’t give a shit, so that’s fine (until you get stuck with the obligation to keep its promises…) Things will get a little worse all over, GDP and productivity will go up, and we’ll claim it’s better. That’s not all of what’s been happening since about the 1970s, but it’s a lot of it, so, nothing new there.
- marcosdumay 3y ago
  How good does it need to be to get any passable result? Well, as good as the people writing it nowadays. The kind of quality you get from writers is not the same as you get from editors. Or did you mean how good it must be before mangers actually start replacing people? That one I have no idea, and I don't expect it to have any relation with how good content people expect to read or how good content the LLMs actually produce.

thelastparadise 3y ago

> but otherwise LLMs aren’t close to replacing 80% of your content department. Maybe 5% at best. LLM alone, no. Well-engineered LLM-based distributed system with good prompting and input data? Absolutely.

jay_kyburz 3y ago
You just replaced a cheap writer with an expensive AI prompt and input data manager.

acdha 3y ago

One way to think about it: how good does it need to be for the managerial class types who are visibly salivating at mass layoffs to convert current professional jobs into 1099 “temporary” LLM quality control gigs? I’m betting we’re close to the level where some large places are going to try.

vundercind 3y ago
I do think it likely we’ll see a lot of organizations decide they just don’t care that much about (much of) their writing being any good. Like rushing to replace your first-line support with a chat LLM: sure it’s terrible, but you don’t give a shit, so that’s fine (until you get stuck with the obligation to keep its promises…) Things will get a little worse all over, GDP and productivity will go up, and we’ll claim it’s better. That’s not all of what’s been happening since about the 1970s, but it’s a lot of it, so, nothing new there.
marcosdumay 3y ago
How good does it need to be to get any passable result? Well, as good as the people writing it nowadays. The kind of quality you get from writers is not the same as you get from editors. Or did you mean how good it must be before mangers actually start replacing people? That one I have no idea, and I don't expect it to have any relation with how good content people expect to read or how good content the LLMs actually produce.