4 ms·

> My experience with it is that the more knowledgeable I am at the domain, the more useful it is. Exactly right. I made this argument in an unrelated thread:

> My experience with it is that the more knowledgeable I am at the domain, the more useful it is. Exactly right. I made this argument in an unrelated thread: I gave an example of testing ChatGPT by asking it to write CRC calculation code in ARM assembler. The code worked. It wasn’t as good as mine (slower, inefficient), yet, as you said, as a starting point, it would have been fine. “This is today, mid 2023. Give it a year or two (maybe less?) and it will be a tool to contend with. People who like to blame everything else rather than their lack of knowledge and experience will not do very well in that world. Why would I pay someone to do <X> when they bring nothing special to the table? Here's the huge paradigm shift (at least for me): I could not care less what someone knows or does not know. I care about the range and breath of their experience and how they approach learning that which you do not know. Someone like that can use any available tool, including AI tools, to deliver value in almost any domain. Someone who blames others (tools, people, the system, whatever), cannot. We might just be entering an era in which experience will be shown to have serious value.”

ghaff 4y ago
That's my experience with writing text as well. If I know the area well enough that I don't really need to do a lot of fact checking that isn't off the top of my head--and I can correct any areas of nuance it doesn't get quite write--it's easy enough to get a bunch of words down on paper and rework them as needed. I haven't found it to be a big help but it can save me time.
passion__desire 4y ago
Programming languages, being precise in description and semantics, means it should be possible, for an automated system, to create new programs by unfolding and expanding already created programs to achieve new solutions. These solutions will be then be memoized and become building blocks for complex problems. A system like GPT-4 should be able to do it, sort of like generating your own training data through self-play like AlphaZero does.
ilaksh 4y ago
But if we find any web page or document that has your knowledge of what makes ARM code fast and feed that into GPT4 (up to say 80kb of assembly with short comments) along with the task then we don't need you to do that anymore. AI in fact can use any tool, including AI tools, to deliver value in almost any domain. So the problem with the comments here in this thread is that the obvious conclusion is that we need some kind of basic income as a floor as well as some fundamental adjustment of the system to handle the new realities of AI. In fact due to the short-sightedness of comments like yours I am tempted to use the AI programming tool I am developing to try to build an ARM assembler expert business and sell it as Robo Martin the Master ARM Assembly Bot. Since GPT4 can write the code for me, it is possible to launch this business within the next two days.
- surgical_fire 4y ago
  Do it. I've been challenging everyone that LARPs about the capabilities of GPT to start a business now, and use it in place of developers. Hell, for a $20 account you get the perfect developer, that is extremely productive, never gets tired, never gets sick, and has no PTO. What is stopping you?
  - ilaksh 4y ago
    I am building it. Have a version deployed for static pages and working on backend/VMs. See my profile.
    - surgical_fire 4y ago
      Nice! Good luck!
- robomartin 4y ago
  > In fact due to the short-sightedness of comments like yours I am tempted to use the AI programming tool I am developing to try to build an ARM assembler expert business and sell it as Robo Martin the Master ARM Assembly Bot. First rule of having a conversation between humans: You can't have a conversation if you resort to insults. Not only insults, but some sort of weird expression of vindictive intent? Pipe down. OK? Evolution is what got humanity here. And, yes, evolution will produce AI that will be far more capable than us at many levels and across a wide range of domains. That is inevitable. However, there will be boundaries for quite some time. Areas where AI will have trouble inserting itself into. Don't ask me what these areas might be. I have some ideas and thoughts on this, yet, I don't really know. In other words, I feel like that conversation would be pointless...just like trying to predict the future. There might also be lots of areas where AI will not be trusted for a range of reasons. As a general statement, I will still hold to the belief that one of the key differentiators as AI becomes more widely useful will be the experience and knowledge of the humans wielding it. We might also see an advantage in the realm of creativity. Today? Absolutely. Ten years from now? Don't know. Can someone create a better ARM assembler writer, one that might beat most experienced programmers across a range of criteria? Sure. Of course. These are mechanical things in the end. There are books, tons of them, written about how to write better code in assembler. If someone did, it better not be a commercial enterprise, making money from writing assembler code isn't likely to be a thing, particularly when one can already do a decent job of it for free. My prediction is that the divide will have experienced and widely knowledgeable people on one side and everyone else on the other. Part of engineering is knowing what questions to ask, when to ask them and when they might be relevant. This becomes particularly complex the more multidisciplinary a project might be. This is where, for now, experience wins.

ghaff 4y ago

That's my experience with writing text as well. If I know the area well enough that I don't really need to do a lot of fact checking that isn't off the top of my head--and I can correct any areas of nuance it doesn't get quite write--it's easy enough to get a bunch of words down on paper and rework them as needed. I haven't found it to be a big help but it can save me time.

passion__desire 4y ago

Programming languages, being precise in description and semantics, means it should be possible, for an automated system, to create new programs by unfolding and expanding already created programs to achieve new solutions. These solutions will be then be memoized and become building blocks for complex problems. A system like GPT-4 should be able to do it, sort of like generating your own training data through self-play like AlphaZero does.

ilaksh 4y ago

But if we find any web page or document that has your knowledge of what makes ARM code fast and feed that into GPT4 (up to say 80kb of assembly with short comments) along with the task then we don't need you to do that anymore. AI in fact can use any tool, including AI tools, to deliver value in almost any domain. So the problem with the comments here in this thread is that the obvious conclusion is that we need some kind of basic income as a floor as well as some fundamental adjustment of the system to handle the new realities of AI. In fact due to the short-sightedness of comments like yours I am tempted to use the AI programming tool I am developing to try to build an ARM assembler expert business and sell it as Robo Martin the Master ARM Assembly Bot. Since GPT4 can write the code for me, it is possible to launch this business within the next two days.

surgical_fire 4y ago
Do it. I've been challenging everyone that LARPs about the capabilities of GPT to start a business now, and use it in place of developers. Hell, for a $20 account you get the perfect developer, that is extremely productive, never gets tired, never gets sick, and has no PTO. What is stopping you?
- ilaksh 4y ago
  I am building it. Have a version deployed for static pages and working on backend/VMs. See my profile.
  - surgical_fire 4y ago
    Nice! Good luck!
robomartin 4y ago
> In fact due to the short-sightedness of comments like yours I am tempted to use the AI programming tool I am developing to try to build an ARM assembler expert business and sell it as Robo Martin the Master ARM Assembly Bot. First rule of having a conversation between humans: You can't have a conversation if you resort to insults. Not only insults, but some sort of weird expression of vindictive intent? Pipe down. OK? Evolution is what got humanity here. And, yes, evolution will produce AI that will be far more capable than us at many levels and across a wide range of domains. That is inevitable. However, there will be boundaries for quite some time. Areas where AI will have trouble inserting itself into. Don't ask me what these areas might be. I have some ideas and thoughts on this, yet, I don't really know. In other words, I feel like that conversation would be pointless...just like trying to predict the future. There might also be lots of areas where AI will not be trusted for a range of reasons. As a general statement, I will still hold to the belief that one of the key differentiators as AI becomes more widely useful will be the experience and knowledge of the humans wielding it. We might also see an advantage in the realm of creativity. Today? Absolutely. Ten years from now? Don't know. Can someone create a better ARM assembler writer, one that might beat most experienced programmers across a range of criteria? Sure. Of course. These are mechanical things in the end. There are books, tons of them, written about how to write better code in assembler. If someone did, it better not be a commercial enterprise, making money from writing assembler code isn't likely to be a thing, particularly when one can already do a decent job of it for free. My prediction is that the divide will have experienced and widely knowledgeable people on one side and everyone else on the other. Part of engineering is knowing what questions to ask, when to ask them and when they might be relevant. This becomes particularly complex the more multidisciplinary a project might be. This is where, for now, experience wins.