ChatGPT

@Desig

As for your question of how Gemini Advanced or ChatGPT Plus has been amazing for me, that's fairly easy to answer: they dramatically cut down the time needed to do things as a developer.
  • They are phenomenal at planning. Specifically, having an open-ended chat with about architectural options for a new project.
  • They can generate high-quality boilerplate for almost anything.
  • Sometimes they can cut down the time it takes to build something, like a WordPress plugin, from days to hours, or from a full day to 2-4 hours.
  • Their ability to QA code is insane. They can catch major and obscure vulnerabilities so easily.
  • I can plug in URLs and get summaries for almost any article
  • Even with videos now, thanks to Gemini Advanced
  • They can generate images, very handy for placeholder content
  • Way more stuff. You're limited by your imagination.
The $20 pm cost is such a drop in the ocean compared to the value I get. There is no comparing how productive you can be in many industries with and without a tool like ChatGPT Plus or Gemini Advanced.

And again, don't use the free versions.

For programming, chatgpt has taken tasks for me which used to take me several days and crammed it down to a few hours. It's great.
 
Not LM Studio, but I have / had some local LLMs running on OOGA BOOGA, but it wasn't particularly useful without internet connectivity.
Interesting, I've only tested out the CodeLlama 7B Instruct and CodeLlama 7B Python models (Meta), running completely offline. So far, so good. I suppose Apple Silicon does have its benefits.

Using it to review / optimize code I've written and writing unit tests.

I have used GitHub Copilot and Amazon CodeWhisperer in the past. I use VS Code and installed the "Continue" extension. I've also opted out of collecting and reporting usage information with Continue.

I don't see myself paying a monthly subscription, there's always Stack Overflow
 
Interesting, I've only tested out the CodeLlama 7B Instruct and CodeLlama 7B Python models (Meta), running completely offline. So far, so good. I suppose Apple Silicon does have its benefits.

Using it to review / optimize code I've written and writing unit tests.

I have used GitHub Copilot and Amazon CodeWhisperer in the past. I use VS Code and installed the "Continue" extension. I've also opted out of collecting and reporting usage information with Continue.

I don't see myself paying a monthly subscription, there's always Stack Overflow
Copilot has come a long way in the GPT-4 era, with workspace context awareness and CLI support. I've seen a gazillion LLM comparisons in some of the dev communities I'm in. It doesn't look like anything comes close to ChatGPT Plus and Gemini Advanced.

Stack Overflow isn't apples to apples. Instead of adapting someone else's code to your needs - which may be of dubious quality anyway or out of date - an LLM can provide you with quality code personalised to your issue.

You can try Gemini Advanced out for free for two months. Just head to gemini.google.com, be signed in with a personal account and start your trial. They'll email you a week before billing starts, and you might be able to cancel immediately anyway and continue the trial.

I've been subscribed to ChatGPT Plus since the day it launched here, but I think I'll be switching to Gemini Advanced. For my needs, which are primarily JS, PHP and discussing ideas, platforms and architectures for new projects, Gemini Ultra has been easily superior to GPT-4 for me.

I'll still cancel my free trial, because having access to my personal emails and files isn't very useful to me, but for sure I'll subscribe once the Gemini for Workspace launches. Being able to easily interact with my business files, emails and calendar will be crazy convenient.
 
Would love to see how capable any of these AIs are with coding up a problem. I suppose it can only generate code for one thing at a time, or a whole lot of stuff that is common out there. I doubt that it will be able to do some domain specific stuff. No ways is it going to do the huge refactor i am busy with as well as bring all tests back in line. Its taken several days so far. This is where I would love to use AI tools
 
Copilot has come a long way in the GPT-4 era, with workspace context awareness and CLI support. I've seen a gazillion LLM comparisons in some of the dev communities I'm in. It doesn't look like anything comes close to ChatGPT Plus and Gemini Advanced.

Stack Overflow isn't apples to apples. Instead of adapting someone else's code to your needs - which may be of dubious quality anyway or out of date - an LLM can provide you with quality code personalised to your issue.

You can try Gemini Advanced out for free for two months. Just head to gemini.google.com, be signed in with a personal account and start your trial. They'll email you a week before billing starts, and you might be able to cancel immediately anyway and continue the trial.

I've been subscribed to ChatGPT Plus since the day it launched here, but I think I'll be switching to Gemini Advanced. For my needs, which are primarily JS, PHP and discussing ideas, platforms and architectures for new projects, Gemini Ultra has been easily superior to GPT-4 for me.

I'll still cancel my free trial, because having access to my personal emails and files isn't very useful to me, but for sure I'll subscribe once the Gemini for Workspace launches. Being able to easily interact with my business files, emails and calendar will be crazy convenient.

Thanks. I made the leap and signed up for Gemini Advanced. Just a quick plugging some old coding tests into it and I'm happy, churned out some great answers.
 
Well I resubscribed to CHATGPT 4 today, it is superior to Gemini with regards to creative pursuits such as critiquing photos and art (Gemini refuses to do anything with images that have people in them), it's more effective at Powershell and Python scripting, its image generator (Dall-E 3) is better than Google's, it has more plugins and the custom GPT functionality as well as custom instructions give it the edge. Its accuracy also seems slightly better...
 
Would love to see how capable any of these AIs are with coding up a problem. I suppose it can only generate code for one thing at a time, or a whole lot of stuff that is common out there. I doubt that it will be able to do some domain specific stuff. No ways is it going to do the huge refactor i am busy with as well as bring all tests back in line. Its taken several days so far. This is where I would love to use AI tools
I'm a genetics PhD student in a group focused on developing a popular scientific software ecosystem. The software has enough of a presence online that ChatGPT can often do domain-specific tasks quite well, but is prone to hallucinating APIs that don't exist. It has been invaluable for debugging: just paste your error and Valgrind log and it will often spot the problem. It's also been outstanding for making complex Matplotlib plots and fast Numpy algorithms.
 
I'm a genetics PhD student in a group focused on developing a popular scientific software ecosystem. The software has enough of a presence online that ChatGPT can often do domain-specific tasks quite well, but is prone to hallucinating APIs that don't exist. It has been invaluable for debugging: just paste your error and Valgrind log and it will often spot the problem. It's also been outstanding for making complex Matplotlib plots and fast Numpy algorithms.
I think the day that you can let it read your entire codebase and ‘understand’ it, and you can tell it to refactor and update the tests it will be a complete game changer
 
I think the day that you can let it read your entire codebase and ‘understand’ it, and you can tell it to refactor and update the tests it will be a complete game changer
Indeed! Once the context windows become big enough for an enormous codebase it will be become even more useful. There'll still need to be someone to check that it's correct though, so that they can take the blame for mistakes.
 
Chatgpt have shortened the text amount. I have a block of code I could easily put in there 2 weeks ago. Today it says nope message too long.
 

What next ? :cautious:

This is the way it's going. There is a certification as an AI Consultant. Demand is growing fast. Companies want to know how "this AI thing" can help them. If they aren't or not willing they will get left behind.
 
This is the way it's going. There is a certification as an AI Consultant. Demand is growing fast. Companies want to know how "this AI thing" can help them. If they aren't or not willing they will get left behind.
That certification is through ZD Net so it seems credible.
 
That certification is through ZD Net so it seems credible.

I'm not sure. I've only come across a few job advertisements on indeed for AI consultants. Mostly overseas in US. I've not followed up on the actual course/degrees themselves.
 
This is the way it's going. There is a certification as an AI Consultant. Demand is growing fast. Companies want to know how "this AI thing" can help them. If they aren't or not willing they will get left behind.
I already know people making money creating Custom GPTs for companies. It's a genuinely useful thing for a company to have. Can include vast knowledgebases, adapt long-form outputs to a specific brand voice and so on. I'm tempted to make some for myself, but it sounds like a huge schlep.
 
I already know people making money creating Custom GPTs for companies. It's a genuinely useful thing for a company to have. Can include vast knowledgebases, adapt long-form outputs to a specific brand voice and so on. I'm tempted to make some for myself, but it sounds like a huge schlep.
It's all fun and games until the GPT is tricked into an enforceable contract but that seems like something which will be solved/mitigated.
 
Top
Sign up to the MyBroadband newsletter