3 ms·

The reason jobs aren't coming back is because startups can't borrow free money. ChatGPT is nowhere near ready to be correct enough to replace humans for importa

The reason jobs aren't coming back is because startups can't borrow free money. ChatGPT is nowhere near ready to be correct enough to replace humans for important tasks.

johnwheeler 2y ago
It doesn't matter that it is wrong often. Consider this: When I first used GPT last year, I asked it to "Create me a content delivery network on AWS" GPT began outputting a fairly comprehensive CDK, which is Code-as-configuration, a layer ontop of AWS SAM, which is another layer over AWS CloudFormation. The script created all the buckets, certificates, policies, and the CloudFront distribution and linked everything together. It was in JavaScript, and I asked it if they had a Python SDK, and it converted the script into Python. Then GPT gave me instructions on how to go into my domain provider to setup a CNAME for a CloudFront subdomain of my choice. THIS IS WHAT HAS BEEN REPLACED: I would have performed a Google search for "AWS CDN How" or something like that. Next, I would read AWS documentation until I felt comfortable enough about the components involved. I'd then go into AWS Console and set one up manually by creating the buckets, policies, certificates, and the CloudFront distribution. I would have wanted a custom domain, so I'd figure out how to setup the CNAME to point to the CloudFront distribution. This would've taken at least two hours for me personally. Knowing nothing about DNS, I'd want to get my bearings and I'd waste time down that rabbit hole. Burn an extra hour with stupid issues like not having a '.' at the end of the CNAME target. That has actually tripped me up in the past. Next to make sure I really understood what was going on-- I would tear the CDN down, build it back up, and take notes and put them in the company Confluence. Afterwards, I would try to automate it the process. Being a generalist and not a DevOps person, I'd probably start searching for some older tech I had heard about but never used: "Ansible AWS", "Puppet AWS" or "Chef AWS". I'd quickly learn about a thing called CloudFormation and start reading up on it. I'd realize I had zero interest in learning CloudFormation because it is tedious and error-prone. I would begin to Google for something on top of it. I'd likely get dumped into a SAM tutorial Next, I'd start building out a prototype in SAM, and by the time it was half-way done and the week was almost over I'd watch a YouTube video and find out about: CDK By then, I would have had enough! As a generalist engineer, I'd tell myself, "you're not qualified to setup a CDN. You need DevOps." But then I would pick myself up and push through for another week in more frustration sorting out the rest of the odds and ends. And this is what my career has been like my whole life. And now GPT is giving me a Python script I can put to use in seconds. It wasn't perfect, and I had some hiccups along the way, but by the end of the day, I had the basics of the CDN I'm using for my product, and I feel I understand it better because it was delivered in such a bite-sized compact script vs me having to cross-reference my memory. This revelation blew my fucking mind! So, I don't use GPT for writing code because that's not what it's good at. I use GPT to automate the very expensive, very time-consuming, highly compensated function of specialized Google Search
- SoftTalker 2y ago
  You have to know what you want, and knowing that you need a "content delivery network on AWS" is something of specialist knowledge, I'll grant that. But if you can type a few words into ChatGPT and have a solution for that, so can anyone. The "very expensive, very time-consuming, highly compensated function of specialized Google Search" was a thing. It's not anymore, and it won't be highly compensated for long because now anyone can do it.
  - johnwheeler 2y ago
    I'm not conveying correct. Prompt engineer is the real thing. I think I'm very good at it - people are worried that it can't write code correctly and they're missing the darn point! I am building a business, my app is complex and polished (of course could be better), but I know I wouldn't have been able to do that hadn't ChatGPT cut through all the hard knowledge work. So I know what it can do. This is my stack NextJS for SSR running on Digital Ocean App Platform React / ChakraUI for UI components Python FastAPI for API routes AWS Cognito and Google OAuth for authentication PostgresQL on Digital Ocean Docker for dev environment and continuous deployment AWS Parameter store for secrets AWS Lambda with FFMpeg layers for video processing Pydantic for JSON marshalling SQLAlchemy for ORM AWS Cloudfront for CDN AWS S3 for storage Chrome Extension API and WXT framework for extension development Hubspot for CRM Gmail for email Namecheap for domain and DNS Figma, Adobe After Effects (check out my product video https://www.youtube.com/watch?v=IJuiTbTjlAo https://www.youtube.com/watch?v=IJuiTbTjlAo) Adobe Photoshop Notion to manage everything, todo lists, etc I'm a full stack engineer, and I know my limits pre-ChatGPT. There's no way I could have done this myself. It helped me with all that.
- dbdifncosmsbxv 2y ago
  It sounds like you replaced learning with trusting some code the internet gave you. Which sure, chatgpt is much better at, but this kind of tech is only a threat if your value is pasting in code you found on the internet. Most serious employers pay for your ability to learn.
  - 2y ago
threatofrain 2y ago
> for important tasks. Yeah, but a lot of people do menial tasks like human OCR, just like once upon a time humans were literally called computers and had such a job.

The reason jobs aren't coming back is because startups can't borrow free money. ChatGPT is nowhere near ready to be correct enough to replace humans for important tasks.

johnwheeler 2y ago

It doesn't matter that it is wrong often. Consider this: When I first used GPT last year, I asked it to "Create me a content delivery network on AWS" GPT began outputting a fairly comprehensive CDK, which is Code-as-configuration, a layer ontop of AWS SAM, which is another layer over AWS CloudFormation. The script created all the buckets, certificates, policies, and the CloudFront distribution and linked everything together. It was in JavaScript, and I asked it if they had a Python SDK, and it converted the script into Python. Then GPT gave me instructions on how to go into my domain provider to setup a CNAME for a CloudFront subdomain of my choice. THIS IS WHAT HAS BEEN REPLACED: I would have performed a Google search for "AWS CDN How" or something like that. Next, I would read AWS documentation until I felt comfortable enough about the components involved. I'd then go into AWS Console and set one up manually by creating the buckets, policies, certificates, and the CloudFront distribution. I would have wanted a custom domain, so I'd figure out how to setup the CNAME to point to the CloudFront distribution. This would've taken at least two hours for me personally. Knowing nothing about DNS, I'd want to get my bearings and I'd waste time down that rabbit hole. Burn an extra hour with stupid issues like not having a '.' at the end of the CNAME target. That has actually tripped me up in the past. Next to make sure I really understood what was going on-- I would tear the CDN down, build it back up, and take notes and put them in the company Confluence. Afterwards, I would try to automate it the process. Being a generalist and not a DevOps person, I'd probably start searching for some older tech I had heard about but never used: "Ansible AWS", "Puppet AWS" or "Chef AWS". I'd quickly learn about a thing called CloudFormation and start reading up on it. I'd realize I had zero interest in learning CloudFormation because it is tedious and error-prone. I would begin to Google for something on top of it. I'd likely get dumped into a SAM tutorial Next, I'd start building out a prototype in SAM, and by the time it was half-way done and the week was almost over I'd watch a YouTube video and find out about: CDK By then, I would have had enough! As a generalist engineer, I'd tell myself, "you're not qualified to setup a CDN. You need DevOps." But then I would pick myself up and push through for another week in more frustration sorting out the rest of the odds and ends. And this is what my career has been like my whole life. And now GPT is giving me a Python script I can put to use in seconds. It wasn't perfect, and I had some hiccups along the way, but by the end of the day, I had the basics of the CDN I'm using for my product, and I feel I understand it better because it was delivered in such a bite-sized compact script vs me having to cross-reference my memory. This revelation blew my fucking mind! So, I don't use GPT for writing code because that's not what it's good at. I use GPT to automate the very expensive, very time-consuming, highly compensated function of specialized Google Search

SoftTalker 2y ago
You have to know what you want, and knowing that you need a "content delivery network on AWS" is something of specialist knowledge, I'll grant that. But if you can type a few words into ChatGPT and have a solution for that, so can anyone. The "very expensive, very time-consuming, highly compensated function of specialized Google Search" was a thing. It's not anymore, and it won't be highly compensated for long because now anyone can do it.
- johnwheeler 2y ago
  I'm not conveying correct. Prompt engineer is the real thing. I think I'm very good at it - people are worried that it can't write code correctly and they're missing the darn point! I am building a business, my app is complex and polished (of course could be better), but I know I wouldn't have been able to do that hadn't ChatGPT cut through all the hard knowledge work. So I know what it can do. This is my stack NextJS for SSR running on Digital Ocean App Platform React / ChakraUI for UI components Python FastAPI for API routes AWS Cognito and Google OAuth for authentication PostgresQL on Digital Ocean Docker for dev environment and continuous deployment AWS Parameter store for secrets AWS Lambda with FFMpeg layers for video processing Pydantic for JSON marshalling SQLAlchemy for ORM AWS Cloudfront for CDN AWS S3 for storage Chrome Extension API and WXT framework for extension development Hubspot for CRM Gmail for email Namecheap for domain and DNS Figma, Adobe After Effects (check out my product video https://www.youtube.com/watch?v=IJuiTbTjlAo https://www.youtube.com/watch?v=IJuiTbTjlAo) Adobe Photoshop Notion to manage everything, todo lists, etc I'm a full stack engineer, and I know my limits pre-ChatGPT. There's no way I could have done this myself. It helped me with all that.
dbdifncosmsbxv 2y ago
It sounds like you replaced learning with trusting some code the internet gave you. Which sure, chatgpt is much better at, but this kind of tech is only a threat if your value is pasting in code you found on the internet. Most serious employers pay for your ability to learn.
- 2y ago

threatofrain 2y ago

> for important tasks. Yeah, but a lot of people do menial tasks like human OCR, just like once upon a time humans were literally called computers and had such a job.