“It is important to understand that it can take a long time for a computer’s ability at a task to go from 0 to x, but then take very little time to go from x to 10x.”
I spent a few weeks focusing on non-coding work and when I came back to code again, GPT was noticeably much better at it. This is already a major labor saver for people who take advantage.
GPT was better at the coding work? I just tried to use ChatGPT to write some code and it was frustratingly annoying to get it to do exactly what I wanted. Maybe I just know what I want and can do it better than the machine. What the machine did well was write an individual function. What it didn't do as well was everything else around it.
That being said, I imagine the machine could do a very good job in something like refactoring some code base, particularly if there are tests making sure the code is right before and after. It could be much less error prone than a human in tedious jobs like that.
If you’re coding in Python, you probably want to use GitHub Copilot rather than ChatGPT. For other languages, I’m still experimenting but I did get a lot of mileage using ChatGPT for R.
My approach was to explain my big picture goal at the outset (along with the obligatory “remember you are an expert R programmer...”) and the basic steps I wanted to put together. Then I say, here is the basic structure/shell I have to start with, can you modify to complete this task?
Had to iterate from there to tweak/explain modifications of course but I found it much better than google/stackoverflow and a lot of the code I could paste directly into my file.
Probably you’re a better coder than me, so this may not move the needle for you just yet, but it made a big difference for me. [And the earlier versions were already useful for ditching tedious busy work...]
I haven't tried GitHub Copilot yet, just ChatGPT*. I've got probably fifteen years experience writing R code, but it's a focused set of knowledge. In my little corner, I'm pretty good, but there's a lot I don't know where some AI might be helpful. Your way of asking it to do a first draft and then modifying might be a better approach than expecting it to produce exactly what I want.
* My use of ChatGPT was with a less well known language, which may have contributed to the poorer results. I would guess that you get better results with languages that it is able to use more information to train on.
My son with 5+ years coding experience has used ChatGPT to write code in a language rather new to him. He had to restructure it a bit but it saved him from lots of tedium getting the syntax right.
I managed the entire Zvi article, but took about four or five sittings to get to the end!
I am teaching a new module after Easter and am trying to collate as much information as possible to use AI effectively and encourage students to do the same. Exciting times!
'And what it accomplished was superhuman. I will go through the details in a moment, but, in 30 minutes it: did market research, created a positioning document, wrote an email campaign, created a website, created a logo and “hero shot” graphic, made a social media campaign for multiple platforms, and scripted and created a video. In 30 minutes.'
So what's the marginal value of a good that is in infinite supply? Its still zero right?
Oh goody, ChatGPT greatly reduces the time it takes to design a marketing and social media campaign aimed at persuading me to buy crap I don't want. An unending supply of annoying ads interrupting my online videos, most of which are already unnervingly targeted at my demographic (for example, anti-aging creams for elderly women). In Mollick's case, the product being marketed is a game "designed to teach leadership and team skills on a fictional mission to Saturn." Ugh. The proliferation over time of management and team building training programs, and the time and money wasted on them, was one of the things I noticed about my last federal government job. I boycotted team building exercises involving hypothetical scenarios. In my experience, team building evolved spontaneously on an as-needed basis out of the project we were assigned to do, provided the team members had the experience, skills and intelligence to figure out what to do. It can't be taught separate from the work, using hypotheticals. My theory was that the proliferation of training programs and associated funding was fueled by some mysterious rent-seeking mechanism.
Is it just me or would a little more context be useful? It took me a bit to figure out what the first quote was about. I still have no idea what the second one is about.
“It is important to understand that it can take a long time for a computer’s ability at a task to go from 0 to x, but then take very little time to go from x to 10x.”
I spent a few weeks focusing on non-coding work and when I came back to code again, GPT was noticeably much better at it. This is already a major labor saver for people who take advantage.
GPT was better at the coding work? I just tried to use ChatGPT to write some code and it was frustratingly annoying to get it to do exactly what I wanted. Maybe I just know what I want and can do it better than the machine. What the machine did well was write an individual function. What it didn't do as well was everything else around it.
That being said, I imagine the machine could do a very good job in something like refactoring some code base, particularly if there are tests making sure the code is right before and after. It could be much less error prone than a human in tedious jobs like that.
If you’re coding in Python, you probably want to use GitHub Copilot rather than ChatGPT. For other languages, I’m still experimenting but I did get a lot of mileage using ChatGPT for R.
My approach was to explain my big picture goal at the outset (along with the obligatory “remember you are an expert R programmer...”) and the basic steps I wanted to put together. Then I say, here is the basic structure/shell I have to start with, can you modify to complete this task?
Had to iterate from there to tweak/explain modifications of course but I found it much better than google/stackoverflow and a lot of the code I could paste directly into my file.
Probably you’re a better coder than me, so this may not move the needle for you just yet, but it made a big difference for me. [And the earlier versions were already useful for ditching tedious busy work...]
I haven't tried GitHub Copilot yet, just ChatGPT*. I've got probably fifteen years experience writing R code, but it's a focused set of knowledge. In my little corner, I'm pretty good, but there's a lot I don't know where some AI might be helpful. Your way of asking it to do a first draft and then modifying might be a better approach than expecting it to produce exactly what I want.
* My use of ChatGPT was with a less well known language, which may have contributed to the poorer results. I would guess that you get better results with languages that it is able to use more information to train on.
Definitely hit or miss, depending on the language. GPT is AWFUL at STATA.
My son with 5+ years coding experience has used ChatGPT to write code in a language rather new to him. He had to restructure it a bit but it saved him from lots of tedium getting the syntax right.
I managed the entire Zvi article, but took about four or five sittings to get to the end!
I am teaching a new module after Easter and am trying to collate as much information as possible to use AI effectively and encourage students to do the same. Exciting times!
'And what it accomplished was superhuman. I will go through the details in a moment, but, in 30 minutes it: did market research, created a positioning document, wrote an email campaign, created a website, created a logo and “hero shot” graphic, made a social media campaign for multiple platforms, and scripted and created a video. In 30 minutes.'
So what's the marginal value of a good that is in infinite supply? Its still zero right?
Oh goody, ChatGPT greatly reduces the time it takes to design a marketing and social media campaign aimed at persuading me to buy crap I don't want. An unending supply of annoying ads interrupting my online videos, most of which are already unnervingly targeted at my demographic (for example, anti-aging creams for elderly women). In Mollick's case, the product being marketed is a game "designed to teach leadership and team skills on a fictional mission to Saturn." Ugh. The proliferation over time of management and team building training programs, and the time and money wasted on them, was one of the things I noticed about my last federal government job. I boycotted team building exercises involving hypothetical scenarios. In my experience, team building evolved spontaneously on an as-needed basis out of the project we were assigned to do, provided the team members had the experience, skills and intelligence to figure out what to do. It can't be taught separate from the work, using hypotheticals. My theory was that the proliferation of training programs and associated funding was fueled by some mysterious rent-seeking mechanism.
Is it just me or would a little more context be useful? It took me a bit to figure out what the first quote was about. I still have no idea what the second one is about.