4 ms·

I write a lot of documents for my consulting work - technical due diligence, analyses, that sort of stuff. The kind of stuff where some hallucination I forgot t

by fhd2 3y ago

I write a lot of documents for my consulting work - technical due diligence, analyses, that sort of stuff. The kind of stuff where some hallucination I forgot to double check could cause bad business decisions. Until now it didn't occur to me that I should maybe proactively declare that I did _not_ use LLMs whatsoever in any stage of working on these. Does anybody here do that kind of thing? Declaring usage is one thing, but I think declaring non-usage might be the more valuable thing in the end.

ghaff 3y ago
I use LLMs some for background information, parenthetical detail, etc. It's all stuff I know, double-check, and edit as I see fit. It does sometimes save some time but you raise an interesting point. If using LLMs becomes sort of a scarlet letter, I guess I'd just forgo the tool? I'm not going to lie and the benefits aren't really that great.
- fhd2 3y ago
  I don´t want to come off wrong here: I totally don't mind if people use GenAI to produce text responsibly and carefully. I do find it fair to communicate it however: As a reviewer, I'll look at something entirely written by a human and (even partially) generated by a stochastic model with different eyes. If I trust someone for example, I know they wouldn't knowingly lie to me. If they used an LLM, they might do just that without even noticing. Personally, I love writing - gets me in the flow. I'm weird that way. But I realise lots of people struggle in the early stages of drafting something. If they have a tool that helps them, that's cool. It's just easier to spot mistakes in something someone (or something) else has written than to avoid typing them manually, is all.
  - ghaff 3y ago
    I think if you're going to require disclosure though, most people are either not going to use it or lie because readers will expect the worst even if it was only used to smooth out come wording or stick in a boilerplate explanation of what a particular technology is for.
    - fhd2 3y ago
      I wouldn´t require disclosure, I think that's simply not realistic, for the reasons you describe, and also because people may just forget. I was wondering if I should declare that I did _not_ use any text analysis or generation tools in analyses I produce for clients. I'd wager people declaring that are less likely to lie about it than those using such tools are to not mention it. And someone reading that might find it helpful knowing that they can go over it with less (obviously still a healthy amount of) scrutiny. Declaring it proactively vs requiring disclosure seems like the reverse of the effect opt-out vs opt-in has on user behaviour. My answer was, by the way, yes. This particular document was in German, but roughly translated I wrote in the first section (where I clarified how I understood the assignment and what goals I set myself): "Aside from automated spell checking, the analysis behind this document as well as the act of drafting and writing it did not involve - at any stage - tools for automated text analysis or text generation." I found my peace with this now, but still curious if (and how!) others do this.
      - ghaff 3y ago
        I think it's fine, assuming it's true, and doesn't hurt although the day may come when you realize you're at a disadvantage if everyone else is using LLMs. Somewhat comparable is that in some contexts journalists and industry analysts, for example, make a point of disclosing any client relationships or investments with companies they're writing about. It's also common to disclose if you were given a product or trip you wrote about.
        fhd2 3y ago
        That's a nice analogy! If I feel at a disadvantage (as in, other people produce better work with LLMs), I'll just adopt them too. But I'm really about quality over quantity, and for my kind of work that's not looking like it's becoming that much of a niche any time soon. For other types of work - entirely different story.
gavinhoward 3y ago
I am a programmer/blogger. I refuse to use these "AI" things. (I haven't even accessed them once.) I put "100% AI-free organic content" in the footer of my blog. [1] So people see that on every page. I'd suggest something similar. [1]: https://gavinhoward.com/ https://gavinhoward.com/

by fhd2 3y ago

ghaff 3y ago

I use LLMs some for background information, parenthetical detail, etc. It's all stuff I know, double-check, and edit as I see fit. It does sometimes save some time but you raise an interesting point. If using LLMs becomes sort of a scarlet letter, I guess I'd just forgo the tool? I'm not going to lie and the benefits aren't really that great.

fhd2 3y ago
I don´t want to come off wrong here: I totally don't mind if people use GenAI to produce text responsibly and carefully. I do find it fair to communicate it however: As a reviewer, I'll look at something entirely written by a human and (even partially) generated by a stochastic model with different eyes. If I trust someone for example, I know they wouldn't knowingly lie to me. If they used an LLM, they might do just that without even noticing. Personally, I love writing - gets me in the flow. I'm weird that way. But I realise lots of people struggle in the early stages of drafting something. If they have a tool that helps them, that's cool. It's just easier to spot mistakes in something someone (or something) else has written than to avoid typing them manually, is all.
- ghaff 3y ago
  I think if you're going to require disclosure though, most people are either not going to use it or lie because readers will expect the worst even if it was only used to smooth out come wording or stick in a boilerplate explanation of what a particular technology is for.
  - fhd2 3y ago
    I wouldn´t require disclosure, I think that's simply not realistic, for the reasons you describe, and also because people may just forget. I was wondering if I should declare that I did _not_ use any text analysis or generation tools in analyses I produce for clients. I'd wager people declaring that are less likely to lie about it than those using such tools are to not mention it. And someone reading that might find it helpful knowing that they can go over it with less (obviously still a healthy amount of) scrutiny. Declaring it proactively vs requiring disclosure seems like the reverse of the effect opt-out vs opt-in has on user behaviour. My answer was, by the way, yes. This particular document was in German, but roughly translated I wrote in the first section (where I clarified how I understood the assignment and what goals I set myself): "Aside from automated spell checking, the analysis behind this document as well as the act of drafting and writing it did not involve - at any stage - tools for automated text analysis or text generation." I found my peace with this now, but still curious if (and how!) others do this.
    - ghaff 3y ago
      I think it's fine, assuming it's true, and doesn't hurt although the day may come when you realize you're at a disadvantage if everyone else is using LLMs. Somewhat comparable is that in some contexts journalists and industry analysts, for example, make a point of disclosing any client relationships or investments with companies they're writing about. It's also common to disclose if you were given a product or trip you wrote about.
      - fhd2 3y ago
        That's a nice analogy! If I feel at a disadvantage (as in, other people produce better work with LLMs), I'll just adopt them too. But I'm really about quality over quantity, and for my kind of work that's not looking like it's becoming that much of a niche any time soon. For other types of work - entirely different story.

gavinhoward 3y ago

I am a programmer/blogger. I refuse to use these "AI" things. (I haven't even accessed them once.) I put "100% AI-free organic content" in the footer of my blog. [1] So people see that on every page. I'd suggest something similar. [1]: https://gavinhoward.com/ https://gavinhoward.com/