Devin AI - Is it already happening?

[FreeDelphiPascal 12]

I expected that the transition to AI will be a bit more gradual, but it seems, Devin is already able to create a full application from a prompt. And I mean the proper way, not ChatGPT style....

So, all kind of questions are coming to my mind. For example, since we know that Microsoft is using GitHub to train its AI, for which later I will have to pay a shitload of money to use it, should I retract my projects from GitHub?

Also, what is Embarcadero doing to stay on top? Will we continue to write code line by line in Delphi IDE, while other programmers will have no IDE anymore but a small chat box where they can enter a prompt?

Since Devin will cost a load of money, how will independent developers (and even small companies) be able to compete? Will we all work for megacorporations that afford to buy the AI?

 

 

[Stefan Glienke 1974]

The last thing I know of is that Devin AI was fake (that info is from April 2024 when the entire developer community was all over the demo they showcased). Do you know more?

[Brandon Staggs 231]

10 hours ago, FreeDelphiPascal said:

Also, what is Embarcadero doing to stay on top?

If this is all for real, why would you care? 

 

Anyway, I am unaware that Embarcadero has any AI development at all, other than people who work for them with side projects for copilot-like integrations of other LLMs. This is a long way off from magically making your app for you, but if that is real, then what difference does it make who makes the tool you use? Because at that point, you aren't even using an IDE.... 

 

Personally, I think we are a LONG way off from any non-trivial software being programmed entirely from AI prompts. And given how Embacadero can hardly keep up with mobile platform updates, I am quite certain if that ever does happen, it'll be from someone else.

[FreeDelphiPascal 12]

17 hours ago, Brandon Staggs said:

I am unaware that Embarcadero has any AI development at all

I am not talking about building their own AI, but IDE-AI integration as copilot...

[Die Holländer 33]

Delphi IDE ChatGPT

My feeling is that a real kind of Copilot will also come for Delphi..

[Lajos Juhász 270]

3 hours ago, Die Holländer said:

My feeling is that a real kind of Copilot will also come for Delphi..

 

Not in near future. Embarcadero made the decision to not integrate any AI yet due to possibility of security issues. Not every customer would be happy if an AI could see their source code. The code could hold sensitive data or proprietary code owned by the company or its client.

 

The current AI systems anyway can help you only to write or check trivial code. You can achieve that using ide plugin or copy paste to AI.

[FreeDelphiPascal 12]

4 hours ago, Lajos Juhász said:

The current AI systems anyway can help you only to write or check trivial code. You can achieve that using ide plugin or copy paste to AI.

I assure you, Copilot is doing a lot more than that !!!!!!

Edited July 25 by FreeDelphiPascal

[hsauro 37]

1 hour ago, FreeDelphiPascal said:

I assure you, Copilot is doing a lot more than that !!!!!!

It does more than trivial code. Even with trivial code, if it a long bit of trivial code, the AI is a time saver.

[corneliusdavid 183]

10 hours ago, Lajos Juhász said:

Embarcadero made the decision to not integrate any AI yet due to possibility of security issues. Not every customer would be happy if an AI could see their source code

I don't understand why this would be a concern. Certainly, any AI built in would just be an interface to utilize your own AI account somewhere, right? Nothing would get automatically uploaded unless you hook it up and start using it. Just like Delphi has integration for git but you don't have to use git inside the IDE unless you tell it to and give it your account information.

 

Am I missing something?

[Joseph MItzen 246]

1 hour ago, corneliusdavid said:

I don't understand why this would be a concern. Certainly, any AI built in would just be an interface to utilize your own AI account somewhere, right? Nothing would get automatically uploaded unless you hook it up and start using it. Just like Delphi has integration for git but you don't have to use git inside the IDE unless you tell it to and give it your account information.

 

Am I missing something?

It all depends on the implementation. I believe JetBrains is offering their own LLM service in their IDEs now, but they've also added a completely local LLM that can perform whole-line auto completion. If hypothetically Embarcadero was offering their own LLM service, it could be turned on by default and people might not be happy about that.

 

But I agree, as an excuse it seems a bit flimsy. I'm not sure why waiting longer would somehow lead to anything being more secure by design.

[Lajos Juhász 270]

9 hours ago, corneliusdavid said:

Am I missing something?

 

Microsoft might or might not using Github for train AI. How can you be sure that when you send with a prompt a piece of code it is not going to be used for training the AI system. If that code contain some top secret detail. The owner of the source code could sue the developer. (I daily work on code bases that is owned by a client of the company I am working for)

Edited July 26 by Lajos Juhász
(Edit. In theory at least now every AI is also learning from the request they get)

[FreeDelphiPascal 12]

I hope (at least in EU) they will introduce some regulations around AIs.

[corneliusdavid 183]

On 7/26/2024 at 3:10 AM, Lajos Juhász said:

Microsoft might or might not be using Github to train AI

First of all, I don't care if my code is used to help train an AI--in fact, I'd be honored! Second, I don't put top-secret stuff in code--or if I do (like a user-requested back-door) I either encrypt it first or (more often) keep it completely separate in external files not included with the source repository.

 

Third, from what I understand, only publicly available repositories (open source) can be legally used for analysis and AI training. And if you do include sensitive code, GitGaurdian can watch your code and immediately warn if you upload secrets or vulnerabilities that shouldn't be there.

 

There's a lot of different opinions to what I've said here; some people are very concerned about privacy and security (and I commend their concern and read their blogs), others don't know or care. I'm kinda in the middle: wary and take steps to protect obvious holes in my process, but I also don't bog myself down with trying to hide everything in constant fear of some large corporation knowing everything I do. I personally feel there's a lot more advantage  in utilizing tools that save me time than there is hiding everything I do for fear it might possibly be used inappropriately. In other words, I have decided to trust policies that state my private code is private.

 

Besides, a vast majority of anything that needs to be kept private is in the database, not the code--and databases are never put on Github, of course.

[Dalija Prasnikar 1382]

On 7/26/2024 at 3:06 AM, corneliusdavid said:

Nothing would get automatically uploaded unless you hook it up and start using it. Just like Delphi has integration for git but you don't have to use git inside the IDE unless you tell it to and give it your account information.

 

Am I missing something?

Once something is integrated, there is always a possibility of bugs. You have a setting that says you are not allowing some feature, but the bug creeps in and the feature ends up enabled behind your back.

 

Another problem is that with AI, companies have incentive to make "bugs" and use anything they can get their hands on for training. Your data is the product. So while you may be fine with some parts of your code being used for training, there will be parts you will want to keep secret for security reason (if nothing else, various API keys and similar), but if you have AI integrated in the IDE, you can never be sure which parts of your code will be used for training and which ones are not. 

 

For some people and companies the security concern is real and even slight possibility that some parts of their code or other information can leak through AI can be a huge risk.

[Vincent Parrett 706]

Here we are talking about AI integration in the IDE - meanwhile, code insight, code completion, code navigation etc - ie the basics of a modern IDE are still pretty much completely broken after how many versions of delphi?  

[Lajos Juhász 270]

8 hours ago, corneliusdavid said:

I don't put top-secret stuff in code--or if I do (like a user-requested back-door) I either encrypt it first or (more often) keep it completely separate in external files not included with the source repository.

Usually the source code of the closed source application is a top secret. That might contain algorithms that are fine tuned, most products are developed and fined tuned for years. I doubt that nobody would buy any product if it was available for free to write an AI prompt:

 

Write and compile an application that behaves like application X that is 200% more efficient and consumes less memory.

 

8 hours ago, corneliusdavid said:

In other words, I have decided to trust policies that state my private code is private.

 

Beside as Dalija wrote there is a chance that a large company will ignore the law. I believe there were some law suits on that matter for example the Volkswagen diesel scandal. I doubt that nobody noticed that they were breaking laws. Similairly tech companies from time to time get sued for various mistakes when they break the law.

[Lajos Juhász 270]

5 hours ago, Vincent Parrett said:

Here we are talking about AI integration in the IDE - meanwhile, code insight, code completion, code navigation etc - ie the basics of a modern IDE are still pretty much completely broken after how many versions of delphi?  

I believe the last perfect IDE was Turbo Pascal?

[Vincent Parrett 706]

4 hours ago, Lajos Juhász said:

I believe the last perfect IDE was Turbo Pascal?

Who mentioned anything about perfect - just pretty good would be an improvement. That is where I would rather my subscription $$$ go towards fixing than more AI distraction. 

 

 

Edited July 29 by Vincent Parrett

[corneliusdavid 183]

8 hours ago, Dalija Prasnikar said:

Your data is the product. So while you may be fine with some parts of your code being used for training, there will be parts you will want to keep secret for security reason (if nothing else, various API keys and similar)

I don't put data into the code repository--nor are there API keys. Those are always kept totally separate and private. THAT is what I consider top secret.

 

8 hours ago, Dalija Prasnikar said:

For some people and companies the security concern is real and even slight possibility that some parts of their code or other information can leak through AI can be a huge risk.

True. And I can appreciate that. For the most part, keeping code in a private repository is good enough security for the projects I work on (small business database apps); the databases and API keys are what are always kept out. It may very well be a completely different scenario for others--and for many good reasons as you've stated.

1 hour ago, Lajos Juhász said:

Usually the source code of the closed source application is a top secret. That might contain algorithms that are fine tuned, most products are developed and fined tuned for years. I doubt that nobody would buy any product if it was available for free to write an AI prompt:

This is the part I guess could vary a lot for different developers/companies. Like I mentioned above, I mostly write small-business database desktop apps containing grids, edit forms, reports, and so forth. There's nothing terribly tricky about the code except the way the forms are designed or business rules that are implemented. Even embedded SQL statements won't reveal more than how the tables are structured.

 

Now, if the customer databases or sales history were to get out, that is something that would be very bad. So for me, those are the "top secret" part of my projects. But if an AI were to train on my proprietary code, all it would get was how I implement logging or store configuration settings or create/destroy forms and show error messages and how data is accessed--but no usernames or passwords.

 

I'm not at all arguing that no code should be private or even that trusting private online repositories or AI privacy policies should be fine for everyone--obviously not. But they're good enough for the micro-apps I build. And I was surprised to see the resistance to including AI in the IDE; but now I understand. (And I'd also like to see bugs fixed first, like @Vincent Parrett said.)

[Dalija Prasnikar 1382]

51 minutes ago, corneliusdavid said:

I don't put data into the code repository--nor are there API keys. Those are always kept totally separate and private. THAT is what I consider top secret.

I am not talking about what you put in your code repository and where and how you host it. I am talking about full AI integration with IDE where you may open security sensitive code in the IDE which then might be sent to AI and end up in training data without your knowledge. And you don't even have to open it. It may just be a part of your project where AI will go through your complete data to be able to give you relevant explanations and completions. 

 

Similar to https://economictimes.indiatimes.com/news/international/us/is-google-gemini-ai-accessing-your-google-drive-files-heres-what-you-can-do-about-it/articleshow/111788643.cms?from=mdr

 

[Stefan Glienke 1974]

Quote

"It is much easier to apologize than it is to get permission" (Grace Hopper)

 

[Anders Melander 1705]

11 hours ago, corneliusdavid said:

Third, from what I understand, only publicly available repositories (open source) can be legally used for analysis and AI training.

Open Source does not mean that the code can be used freely. There are still conditions that must be observed (attribution, restrictions on usage, etc.). An AI (or the people that train it) doesn't care about that.

 

11 hours ago, corneliusdavid said:

In other words, I have decided to trust policies that state my private code is private.

I think that is naive. Have the people who wrote those policies earned your trust? In my POV, that whole industry has shown from the start that they can't be trusted.

 

3 hours ago, Lajos Juhász said:

Write and compile an application that behaves like application X that is 200% more efficient and consumes less memory.

I don't think you understand how these things work; We are nowhere near being able to do anything like that. In fact that task would require a completely different type of  AI. All we have now is one that is good at emulating stuff.

 

3 hours ago, Lajos Juhász said:

I believe there were some law suits on that matter for example the Volkswagen diesel scandal. I doubt that nobody noticed that they were breaking laws. Similairly tech companies from time to time get sued for various mistakes when they break the law.

Volkswagen wasn't a mistake or an accident. They knowingly implemented a mechanism to circumvent an environmental test.

 

AI companies use the Science excuse instead; Their purpose and goal is more important than rules and ethics. If they were actually working a kind of AI that could do more than just copy other peoples work then I might understand it - but they aren't.

 

[Stefan Glienke 1974]

It's really embarrassing to see how many people who work in software development don't know how generative AI works and are being fooled by it like an average TikTok teenager.

[Vincent Parrett 706]

Read this the other day - https://www.nvidia.com/en-us/glossary/generative-ai/

 

Interesting that there is no mention of eithics etc - I guess shareholders don't care. 

[Lajos Juhász 270]

4 hours ago, Anders Melander said:

I don't think you understand how these things work; We are nowhere near being able to do anything like that. In fact that task would require a completely different type of  AI. All we have now is one that is good at emulating stuff.

Please read the previous sentence: "I doubt that nobody would buy any product if it was available for free to write an AI prompt:"

 

I wrote if AI would learn everything from our source and would be able to generate such an application. Companies would not be able to sell their products. Everyone would using AI generate the application they want / need. Without buying custom or available programs. Yes I am aware also on various "SPAM" youtube videos how you can get rich using AI using prompts to generate application for stock trading based on AI and similair. I block those videos as soon as Youtube tries to show them.