4 ms·

I use a spell checker to catch typos. Occasional quirky grammar is mine. Feedback will be provided by the audience. Why would I let a statistical model judge my

I use a spell checker to catch typos. Occasional quirky grammar is mine. Feedback will be provided by the audience. Why would I let a statistical model judge my work? This is how you kill originality.

lurking_swe 2y ago
i didn’t imply you’d need an LLM to “rate” your content. More so asking questions during and before the publishing step to help you improve your work. Not removing your identity from your work. Examples questions of what you could ask an LLM: • are there any redundant sentences or points in this chapter? • i’m trying to remember an expression used to convey X, can you remind me what it is? • i’m familiar with X from my industry Y, but i’m trying to convey this to an audience from industry Z. Can you help brainstorm some words or examples they may be more familiar with? Things like that. I think of it like having a virtual rubber duck that can help you debug code, or anything really. Obviously these are just some suggestions. If you don’t find any of this useful, or even interesting, then carry on. :)
- nils-m-holm 2y ago
  I see. Most of the things you list I think I could do better on my own. The case of X from industry Y sounds interesting, but I would still prefer to hear from a real human being from industry Z. If no one is available, of course, a statistical model may indeed be helpful -- I still do not think it is worth boiling the oceans, though. :)
  - BehindBlueEyes 2y ago
    This. I use AI to bounce off ideas when i wordsmith for work, but mostly when my human counterpart is unavailable to do so and I can't wait until they return eg. when they were on extended time off. The AI isn't as good though, it's like asking an overeager intern instead of a thoughtful senior. Add to it that not all LLMs are equal: at work I have access to the latest GPT models and can customize them myself. At home anything I tried in LM studio feels very limited/inflexible. All models i tried running were useless except one codestruct, which was half decent at answering clarification questions about c# code from tutorial, but failed the task i initially tried it for: generating bespoke tutorials that fit my knowledge gaps, which would have been my ideal scenario because most tutorials aren't written for someone like me. It's another reason LLMs are a bit disappointing to me, they don't seem to be good at handling queries I couldn't solve with a web search, and have similar issues as search results when they do have answers - this makes sense since that's how they are trained, but often defeats the purpose for me.

lurking_swe 2y ago

i didn’t imply you’d need an LLM to “rate” your content. More so asking questions during and before the publishing step to help you improve your work. Not removing your identity from your work. Examples questions of what you could ask an LLM: • are there any redundant sentences or points in this chapter? • i’m trying to remember an expression used to convey X, can you remind me what it is? • i’m familiar with X from my industry Y, but i’m trying to convey this to an audience from industry Z. Can you help brainstorm some words or examples they may be more familiar with? Things like that. I think of it like having a virtual rubber duck that can help you debug code, or anything really. Obviously these are just some suggestions. If you don’t find any of this useful, or even interesting, then carry on. :)

nils-m-holm 2y ago
I see. Most of the things you list I think I could do better on my own. The case of X from industry Y sounds interesting, but I would still prefer to hear from a real human being from industry Z. If no one is available, of course, a statistical model may indeed be helpful -- I still do not think it is worth boiling the oceans, though. :)
- BehindBlueEyes 2y ago
  This. I use AI to bounce off ideas when i wordsmith for work, but mostly when my human counterpart is unavailable to do so and I can't wait until they return eg. when they were on extended time off. The AI isn't as good though, it's like asking an overeager intern instead of a thoughtful senior. Add to it that not all LLMs are equal: at work I have access to the latest GPT models and can customize them myself. At home anything I tried in LM studio feels very limited/inflexible. All models i tried running were useless except one codestruct, which was half decent at answering clarification questions about c# code from tutorial, but failed the task i initially tried it for: generating bespoke tutorials that fit my knowledge gaps, which would have been my ideal scenario because most tutorials aren't written for someone like me. It's another reason LLMs are a bit disappointing to me, they don't seem to be good at handling queries I couldn't solve with a web search, and have similar issues as search results when they do have answers - this makes sense since that's how they are trained, but often defeats the purpose for me.