Has AI Changed Software Development?
-
Comments posted to this topic are about the item Has AI Changed Software Development?
-
My employer blocks all AI, so we can't use it for anything.
On my own, at home, I've used both Bing Chat and ChatGPT to solve programming problems. Especially for something I've never used before, like some configuration of Git repos in GitHub. I've found it to be very useful in helping me to find what to do. I'll take the suggestion, very its efficacy, then make whatever necessary changes I must to the approach given, to make it work.
I'd say GenAI would be fantastic to use, as an assistant, in developing code. I only wish I could use it at work.
Kindest Regards, Rod Connect with me on LinkedIn.
- Rod at work wrote:
My employer blocks all AI, so we can't use it for anything.
On my own, at home, I've used both Bing Chat and ChatGPT to solve programming problems. Especially for something I've never used before, like some configuration of Git repos in GitHub. I've found it to be very useful in helping me to find what to do. I'll take the suggestion, very its efficacy, then make whatever necessary changes I must to the approach given, to make it work.
I'd say GenAI would be fantastic to use, as an assistant, in developing code. I only wish I could use it at work.
The company I work for has implemented ChatGPT 4 on its internal servers, actively promoting its usage among all employees. Given its deployment on company servers, there is also a strong encouragement to utilise it for code improvement without concerns about the security implications of pasting code onto the internet.
-
This was removed by the editor as SPAM
-
While I haven't used it for any technical work, I have used ChatGPT to help me write business cases and other project documents -- not in the sense of "do this for me," but in the sense of "this sentence isn't very good, how would you rewrite it?"
- Jonathan AC Roberts wrote:
The company I work for has implemented ChatGPT 4 on its internal servers, actively promoting its usage among all employees. Given its deployment on company servers, there is also a strong encouragement to utilise it for code improvement without concerns about the security implications of pasting code onto the internet.
Interesting, my company (just me, for now) has a similar approach. Not sure what "implemented ChatGPT 4 on its internal servers" means tho. Are you building custom GPT(s)? Are you using the latest GPT4 (gpt-4-1106-preview)? Gpt-4-1106-preview has training cutoff of 4/2022 and the context size is 16K (double the standard Gpt4). For an interactive interface we're only using the Playground (https://platform.openai.com/playground?mode=chat). We're also using the openai .NET library (https://github.com/betalgo/openai) for certain things.
The thinking behind using GPT without limit is the code we get is imperfect and never the final product. Often we discard it at some point after the fact. To piece together something coherent based on observing our prompts seems unlikely. OpenAI would have to judge what we're doing as having a better time/value payoff than developing their own business. Seems unlikely. Other reasons are reputation/trust, legal/ethical, sheer volume they'd have to sift through, time it saves us vs OpenAI, etc.
Aus dem Paradies, das Cantor uns geschaffen, soll uns niemand vertreiben können
-
Same, corporate has blocked access to any generative AI website for legal/ethical/copyright reasons. Personally, I have used it create logic shells for languages I am less familiar with that I then fine tune to meet my needs. I would agree that most generative AI output is at a junior level and needs a SME to determine if it fits the need. However, it is quite helpful it you don't have access/resources to hire a junior level developer 🙂
- Steve Collins wrote:Jonathan AC Roberts wrote:
The company I work for has implemented ChatGPT 4 on its internal servers, actively promoting its usage among all employees. Given its deployment on company servers, there is also a strong encouragement to utilise it for code improvement without concerns about the security implications of pasting code onto the internet.
Interesting, my company (just me, for now) has a similar approach. Not sure what "implemented ChatGPT 4 on its internal servers" means tho. Are you building custom GPT(s)? Are you using the latest GPT4 (gpt-4-1106-preview)? Gpt-4-1106-preview has training cutoff of 4/2022 and the context size is 16K (double the standard Gpt4). For an interactive interface we're only using the Playground (https://platform.openai.com/playground?mode=chat). We're also using the openai .NET library (https://github.com/betalgo/openai) for certain things.
The thinking behind using GPT without limit is the code we get is imperfect and never the final product. Often we discard it at some point after the fact. To piece together something coherent based on observing our prompts seems unlikely. OpenAI would have to judge what we're doing as having a better time/value payoff than developing their own business. Seems unlikely. Other reasons are reputation/trust, legal/ethical, sheer volume they'd have to sift through, time it saves us vs OpenAI, etc.
They've hosted it on Google Cloud Platform (GCP). It has: OpenAI's GPT-3.5 turbo, GPT-4, GPT-4-32K-0613, gpt-mistraland Google VertexAI's PALM 2 text-bison, chat-bison, chat-bison-32k, and codechat-bison, codechat-bison-32k. All this from one web page that looks very much like the ChatGPT page.
-
"Gpt-4-1106-preview has training cutoff of 4/2022 and the context size is 16K (double the standard Gpt4)" Oops that was not correct. The cutoff for Gpt4-Turbo is 4/2023 and the context size is 128K (which was a surprise because they just announced 16K a few months ago). This is the table (from today: 1/8/2024) at OpenAI.com
Aus dem Paradies, das Cantor uns geschaffen, soll uns niemand vertreiben können
-
This was removed by the editor as SPAM
-
Thank you, Steve Collins, for the link to the OpenAI .NET library! I was unaware of that.
Kindest Regards, Rod Connect with me on LinkedIn.
-
This question still has me thinking. How has development changed because of AI? There's more than one trend happening at once imo. For SQL the impact for me has been minimal with some exceptions. One exception is AI knows the sys tables really well and can crank out meta data queries asap. For everything else (C#/git/Powershell/etc) AI makes accomplishing tasks smoother overall. More thoroughly too because to rely on AI output directly is impossible. For C# AI has nudged me into following TDD or Test Driven Development, which I hadn't done before because it seemed slow and overly formal. TDD means first you get a unit test working and then you use the result once the test assertions are confirmed. The idea "AI will write all the code in the future" goes against another big trend in C# which is compiler written methods and code generators. Code from AI works maybe 60-80% of the time whereas code from the C# compiler ipso facto works 100% of the time. In C# there is something called "minimal API endpoints" where you assign API endpoint attributes to a .NET web application (in C#) and the compiler writes the endpoint methods (in Intermediate Language). Which way is preferable to generate code? Not AI imo. Imo the bigger trend in C# is compiler/language optimizations and not AI. AI helped me figure this out tho and now we have unit tests which improve code maintenance and management 🙂
Aus dem Paradies, das Cantor uns geschaffen, soll uns niemand vertreiben können
-
Interesting. I think AI works better, in general, with more generic things, or with some code prompting, which is easier in C# than SQL.
I'm doing some testing with SQL Prompt + (w/ AI) and because it sees my schema, it can do a few things. I'm hoping it will help me setup/write tests quicker as well and do more TDD.
-
Interesting Steve. I've not used AI to generate much C# code, but I have used it to help me configure something. Like I used it to help me do some configuration in GitHub organizations. When it comes to C#, I can easily find what I'm looking for. When it comes to something like configuring GitHub organizations settings, I've no clue where to start. And I'm basically on my own, at my job.
Kindest Regards, Rod Connect with me on LinkedIn.
Viewing 14 posts - 1 through 13 (of 13 total)
You must be logged in to reply to this topic. Login to reply