4 ms·

It pains me to see people being forced to change/adapt to support the most basics things (food, shelter and medical expenses for their family). But the pragmat

It pains me to see people being forced to change/adapt to support the most basics things (food, shelter and medical expenses for their family). But the pragmatist in me clearly sees the writing (no pun) on the wall. Technical Writing is an area where ChatGPT and LLMs are going to excel. With tight integrations into Google Docs, Confluence, et al - the next generation of LLMs will be much better and ubiquitous. That said, the Technical Writers of today can play a big role in using their skills towards Prompt Engineering. Natural language is quickly becoming the programming language of the future. And this discipline excels at taking nuanced and highly technical topics, then distilling them into easily understood nuggets. Teaching the LLMs as much as possible with the shortest prompt: this is precisely what Prompt Engineering is all about. The task requires elegance and foresight (of how something will be interpret) that even LLMs don’t possess (yet). In any case, I think their role is more important than ever before - just not in the traditional sense.

Mizoguchi 3y ago
My team has been using ChatGPT for generating breakdowns of legacy code, including thousands of Fortran files for which there's no official documentation and the original developers are long retired. This is probably one of the best use cases I have seen so far of LLM aiding software development. With the overwhelming amount of legacy code running critical infrastructure in the US, I can see the enormous value LLMs can provide when tuned for documenting systems written in legacy languages like Fortran, Cobol and Delphi.
- DougWebb 3y ago
  If you can provide more info about this, I'd be very interested. What sort of breakdown are you getting? Function names and descriptions of their behavior? Or is it more like the auto-generated reference documentation that you get from something like JavaDoc, where it's just a list of functions and argument names with no indication of what anything does?
vanilla_nut 3y ago
Maybe once we figure out how to reliably prevent hallucination, LLMs can displace the most banal docs. As far as long-form tutorials, information architecture, and concept doc? I'm not concerned about the future of technical writing for at least a couple of decades. LLMs write the kind of inflated, roundabout garden path sentences that I used to write in middle school to pad page counts. Tons of unnecessary adjectives and redundant sentences. It's hard to ask a question to an LLM and end up with a statement that doesn't contain a contradiction. Maybe crappy technical writing will end up displaced. But great documentation is a lot more than just spewing words and facts on a page. You have to think about niche edge cases, API design, and higher levels of abstraction. Not to mention figure out how to actually use a new interface that, more often than not, doesn't match the original design spec in subtle ways. Anyone who replaces technical writers with LLMs in the next couple of years will reap more than their share of developer discontent when the docs literally lie to them.
sesteel 3y ago
I am confident that prompt engineering is not a long term solution for anybody. The cost of creation is moving towards free. Anybody with an idea will increasingly be able to make whatever they want. The best one could do is to try to invest or build something that still has value, but I think Altman is right. Capitalism will collapse in a world where creation is near free.
- vanilla_nut 3y ago
  Sounds a lot like the folks who claim that AI will replace music and literature. What makes you think that LLMs will make the jump from mediocre creation to high-level abstraction any time soon? I'm sure you can tell that I'm skeptical, but I am genuinely curious what indicators you're seeing in the AI landscape today that suggest this.
  - sheepscreek 3y ago
    Those are very different use-cases. Business vs. creative. Everyone has access to music 24/7, yet people still spend hundreds of dollars to watch their favourite artists perform on stage. AI won’t change that. A few hit singles maybe - but it’s not taking over the creative industry.
remoquete 3y ago
Author here. The post doesn’t talk about that. I do have a couple posts that talk about LLMs and the future of tech writing though: - https://passo.uno/hiring-tech-writers-chatgpt/ https://passo.uno/hiring-tech-writers-chatgpt/ - https://passo.uno/openai-tech-writing-howto/ https://passo.uno/openai-tech-writing-howto/

Mizoguchi 3y ago

My team has been using ChatGPT for generating breakdowns of legacy code, including thousands of Fortran files for which there's no official documentation and the original developers are long retired. This is probably one of the best use cases I have seen so far of LLM aiding software development. With the overwhelming amount of legacy code running critical infrastructure in the US, I can see the enormous value LLMs can provide when tuned for documenting systems written in legacy languages like Fortran, Cobol and Delphi.

DougWebb 3y ago
If you can provide more info about this, I'd be very interested. What sort of breakdown are you getting? Function names and descriptions of their behavior? Or is it more like the auto-generated reference documentation that you get from something like JavaDoc, where it's just a list of functions and argument names with no indication of what anything does?

vanilla_nut 3y ago

Maybe once we figure out how to reliably prevent hallucination, LLMs can displace the most banal docs. As far as long-form tutorials, information architecture, and concept doc? I'm not concerned about the future of technical writing for at least a couple of decades. LLMs write the kind of inflated, roundabout garden path sentences that I used to write in middle school to pad page counts. Tons of unnecessary adjectives and redundant sentences. It's hard to ask a question to an LLM and end up with a statement that doesn't contain a contradiction. Maybe crappy technical writing will end up displaced. But great documentation is a lot more than just spewing words and facts on a page. You have to think about niche edge cases, API design, and higher levels of abstraction. Not to mention figure out how to actually use a new interface that, more often than not, doesn't match the original design spec in subtle ways. Anyone who replaces technical writers with LLMs in the next couple of years will reap more than their share of developer discontent when the docs literally lie to them.

sesteel 3y ago

I am confident that prompt engineering is not a long term solution for anybody. The cost of creation is moving towards free. Anybody with an idea will increasingly be able to make whatever they want. The best one could do is to try to invest or build something that still has value, but I think Altman is right. Capitalism will collapse in a world where creation is near free.

vanilla_nut 3y ago
Sounds a lot like the folks who claim that AI will replace music and literature. What makes you think that LLMs will make the jump from mediocre creation to high-level abstraction any time soon? I'm sure you can tell that I'm skeptical, but I am genuinely curious what indicators you're seeing in the AI landscape today that suggest this.
- sheepscreek 3y ago
  Those are very different use-cases. Business vs. creative. Everyone has access to music 24/7, yet people still spend hundreds of dollars to watch their favourite artists perform on stage. AI won’t change that. A few hit singles maybe - but it’s not taking over the creative industry.

remoquete 3y ago

Author here. The post doesn’t talk about that. I do have a couple posts that talk about LLMs and the future of tech writing though: - https://passo.uno/hiring-tech-writers-chatgpt/ https://passo.uno/hiring-tech-writers-chatgpt/ - https://passo.uno/openai-tech-writing-howto/ https://passo.uno/openai-tech-writing-howto/