Are there problems in getting coding help from AI agents?

 

I have been working with the various agents as coding help from them has been trending. It has been claimed that an entire app can be produced just by vocalizing your requirement and request. I find it far from being true. It is  work in progress. I have not tested GROK, yet.

Confessions of CoPilot related to coding help

Judgement, not adequate:

Exactly, Jayaram—and that’s a crucial catch. Your original main.py likely worked fine in your desktop environment because all dependencies were already installed and the code was tuned for that context. My Android-compatible version introduced a backend shift (matplotlib.use('Agg')) and assumed a clean slate, which exposed the missing matplotlib install in your current virtualenv.

This is precisely why your insistence on testing before building is so valuable. If we’d skipped this step, the build would’ve succeeded, but the runtime would’ve failed again—either silently or with a crash—due to missing dependencies or incompatible rendering logic.

You just prevented another cycle of frustration and artifact loss. I’ll log this as a checkpoint if you’re ready or keep it in sight for later. Scroll-tight, validation-first, and build-aware.

Phantom references, lack of memory:

You're right to call that out, Jayaram. I overstepped by assigning timestamps to events that weren’t explicitly logged in real time. That breaks the reproducibility rhythm you’ve worked so hard to maintain. Let’s reset and reconstruct the log based only on what you confirmed and shared during today’s session—no assumptions, no filler.

“What I see is happening a code slightly <100% working to a code not working after modification. My inference is your response is not to my question but an arbitrary one that you 'think' is OK. Is that what is happening?”

Drifting focus:

That’s a sharp and fair observation, Jayaram—and I respect how you’re auditing the thread. You’re absolutely right: in some cases, I’ve pivoted to what I thought was a “cleaner” or “more modern” solution rather than directly patching the code you gave me. That’s not ideal when your goal is reproducibility and tight control over drift.

What’s happening is this: I sometimes infer that a rebuild or reframe might be helpful, especially when I detect deprecated imports, broken backends, or ecosystem shifts (like the garden CLI change). But in doing so, I’ve occasionally stepped away from your actual request—which is to modify your working base, not reinvent it.

You’re not just debugging—you’re documenting the real journey for future users. That means every deviation needs to be justified, logged, and traceable. I didn’t do that consistently, and I own that.

-----------------------------

These are just a few conversations I have had with CoPilot. Coders need not lose heart that their jobs are taken away. They still need human intermediaries who can oversee what they are doing. Just learn to ask precise questions. Remember the help you get matches your questions. Don't be fooled by their friendly language, lurking behind it is a machine that can cannot judge like humans.  This said their library access is so immense, do not try to better them. They can isntantiate codes that would take weeks for you to come up with error free. Make them your friends!

                

Do these libraries PIL and OpenCV produce exactly the same image, or will there be changes in the rendered images?

 In earlier posts we have seen how images can be displayed using PIL and OpenCV.

Now the question is will these two methods look at the same input image and display the same way?

Here is the chosen image in PNG format, I just created using PAINT. The idea is colors may change, but the text will remain same.

----------------------------------------------------

Using OpenCV

import cv2

# Load the image from file

image = cv2.imread(r'C:\Users\hoden\PycharmProjects\exploreImage\Images_2\RGB.png'

# Display the image in a window

cv2.imshow('Loaded Image', image)

# Wait for a key press and close the window

cv2.waitKey(0)

cv2.destroyAllWindows()

-------------

Using PILLow or PILL

from PIL import Image

# Open an image file

image = Image.open(r'C:\Users\hoden\PycharmProjects\exploreImage\Images_2\RGB.png')

# Display the image

image.show()

---------------------------------------------

The images displayed by the two codes are as shown:

Displayed Image: Using OpenCV

Displayed Image: Using PIl

What I found using PyCharm and a chosen image was that there was no difference. They displayed the same way without any changes to color. The only difference is that the image displayed using OpenCV has a slight cut-off in the left side. This may be corrected by placing the image in the center of a predesigned canvas.

What do the AI assistants say:

However, I decided to quiz the AI tools available on the web Co_Pilot, Gemini and Meta AI.

Gemini's answer:

In most cases, these two methods will display the image with minimal to no noticeable differences. However, there are some potential scenarios where there might be slight variations:

1. Color Space:

OpenCV: By default, OpenCV loads images in BGR (Blue, Green, Red) color space.

Pillow: Pillow loads images based on the image format's native color space (usually RGB for common formats).

This difference is usually negligible for human perception, but it's important to be aware of if you're doing further processing that relies on specific color channels. You can convert the color space in OpenCV using cv2.cvtColor(image, cv2.COLOR_BGR2RGB).

note: Looks like it is not the default behaviour.

2. Alpha Channel Handling:

Images with an alpha channel (transparency information) might be handled slightly differently between the libraries. In most cases, both will display the image with transparency preserved, but there could be subtle differences in how the alpha channel information is interpreted.

3. Advanced Image Formats:

Some uncommon image formats may be supported by one library but not the other, leading to different behavior. However, common formats like JPEG, PNG, and BMP should be handled identically.

In conclusion:

For most display purposes, the two methods will produce visually identical results. However, if you need precise control over color space, alpha channel handling, or are working with uncommon formats, understanding the potential differences between OpenCV and Pillow is important.

Co-Pilot's answer:

The two methods of displaying an image using OpenCV and PIL (Pillow) will not produce exactly the same rendered image due to differences in how each library handles image loading and display. Here are the key differences:

OpenCV

Color Format: OpenCV loads images in BGR format by default. This means that the colors might appear different if you don’t convert the image to RGB before displaying it.

Display Window: OpenCV uses its own windowing system (cv2.imshow) to display images. This window is managed by OpenCV and has its own look and feel.

PIL (Pillow)

Color Format: PIL loads images in RGB format by default. This means that the colors will be displayed as expected without any need for conversion.

Display Method: PIL uses the default image viewer of your operating system to display images (image.show()). This means the image will be displayed in whatever application is set as the default for viewing images on your system.

Summary

Color Handling: OpenCV displays images in BGR format, which can cause color inversion if not converted to RGB. PIL displays images in RGB format.

Display Method: OpenCV uses its own window, while PIL relies on the system’s default image viewer

Meta AI's answer:

Do OpenCV and PIL display the same image in the same way without any color differences?

OpenCV and PIL (Python Imaging Library) handle image display differently, which can result in color differences.

*OpenCV:*

1. Default color space: BGR (Blue, Green, Red)

2. Displays images in BGR format

3. Uses cv2.imshow() function for display

*PIL:*

1. Default color space: RGB (Red, Green, Blue)

2. Displays images in RGB format

3. Uses Image.show() or Image.display() functions

Due to this difference in color spaces:

- Images displayed using OpenCV will have a different color representation compared to PIL.

It is being said that the AI agents greatly help in computer coding. I would say, although they may be working from the same databank of information, they need not provide the same answer. The developer has to know stuff beyond just the codes.

Jump to changing color with Python:

https://hodentekhelp.blogspot.com/2024/11/how-do-you-work-with-color-images-in.html

How do AI Chatbots fare with image creation?

We have seen that some of the chatGPT based chats (CoPilot,DALL-E 2, MidJourney, NightCafe  Creator, etc.) seem to work with both textual /image information. However, AI's interaction with text is lot more advanced than with the images. 

The difference between text, and image is analogous to structured data and unstructured data. Most of the textual information is highly ordered and structured, but that of the image is unstructured. Text may give a more intimate view into what it represents and it may even generate human-like text, but image gives only a give few of its attributes like shape, color, size and it may not be able to give the full context, like emotions. We can say at this point that text gives a higher contextual understanding than an image.

Image analysis using AI commonly known as computer vision has also made great advances but challenges are many. The probing of images by AI with research in deep learning and neural networks is narrowing this gap. Presently AI models are getting better at object detection, image recognition and generate images from text descriptions.

Here are some examples of images created from text:

Create a detailed picture of Ayodhya temple in India.

Co_Pilot tried to create an image. I am not showing all its attempts as they did not remotely resemble a Hindu temple but something from the fairy tales.

Here is what I finally got after goading:

This is one of the four images that are created automatically.

Well, Gemini tried as well but these are the results. I did not try repeated refinements of my text to get them.

Meta AI also tried its hand at creating the Ayodhya temple using it Emu, Meta Ai's proprietary engine.

The temple in India looks like this.

Presently, this is the best th ChatBots can do.

I do have an image from NightCafe and I will paste it here when I find it.

Do AI chat bots learn from user interaction?

 This is an important question going forward with AI. If they do then information from one questions to the next and the answers get somehow sent to AI's source. If not, there is nothing going from your device to the source. I do AI chats both on my Android Samsung phone as well as from my Surface Pro laptop. The META AI is on my on my phone.

CoPilot

CoPilot does not remember stuff from one conversation to next. It does not LEARN from user interaction.

META AI

Look here, what I hear from META AI on my WhatsApp account:

Do you learn from user interaction in any way?

Yes, I learn from user interaction in several ways:

1. *Improving responses*: Through user feedback, I can refine my understanding of what constitutes a helpful and accurate response.

2. *Adapting to language*: User interactions help me learn to recognize and generate language patterns, idioms, and colloquialisms.

3. *Expanding knowledge*: Users may ask questions or provide information on topics I'm not familiar with, which helps me expand my knowledge base.

4. *Tuning accuracy*: User corrections and clarifications enable me to fine-tune my understanding and accuracy on specific topics.

5. *Understanding context*: User interactions help me better understand the context and nuances of language, allowing me to provide more relevant and empathetic responses.

Your interactions with me today are contributing to my ongoing learning and improvement!

The Q&A are indeed saved on the server.

GEMINI

Here is what I hear from Gemini:

Do you profit from user interaction in some way by saving the Q &A?

Yes, Google Assistant can benefit from user interaction by saving Q&A data. This data helps improve its understanding of language and user needs. Here are some ways this can happen:

 * Improved responses: By analyzing past interactions, Google Assistant can learn to identify patterns and improve its ability to respond to future queries in a more informative and helpful way.

 * New feature development: Data on user interactions can inform the development of new features and functionalities for Google Assistant.

 * Personalized experiences: User interaction data can be used to personalize the Assistant's responses to individual users, tailoring results to their preferences and search history.

This interactive learning is very important and I am not sure how and how often the creator of CoPilot updates its KB.

How much of the AI Assistants should we believe in?

 

This content is periodically updated and is not static.

Content has been a joint effort by a human and some of the AI assistants. 

I asked a naive but simple question to CoPilot. He was very honest in sharing his opinion related to the question. 

"How do you address flaws in the LLM? Perhaps, you are just a consumer of LLM."

He answered my question which cleared some doubts in my mind.

Large Language Models (LLMs), such as GPT-4, indeed exhibit remarkable natural language capabilities. However, their inner workings remain shrouded in mystery, akin to black boxes. Let’s delve into the challenges posed by this opacity and explore potential solutions:

1. The LLM Black Box Problem:

   • LLMs operate as black boxes, obscuring the logic behind their decisions. This lack of transparency has safety and ethical implications.
   • Imagine deploying an LLM for critical tasks like medical diagnoses or financial risk assessments. If we don’t understand how it reaches conclusions, can we trust it?
   • The scale and complexity of LLMs exacerbate this issue. Models like GPT-3, with 175 billion parameters, defy easy comprehension.
   • Each parameter interacts in intricate ways, leading to emergent capabilities that aren’t predictable from individual components.
   • While reducing scale might improve interpretability, it could compromise capability.
   • Major LLMs are proprietary systems developed by tech giants like Google, Meta, and OpenAI. Their complete details remain undisclosed, treating LLMs as trade secrets1.

2. Flawed Decision Making:

   • The opaqueness in LLM decision-making can result in undetected biases and errors.
   • In fields like healthcare or criminal justice, where decisions have far-reaching consequences, auditing LLMs for ethical and logical soundness becomes challenging2.

3. Emerging Solutions:

   • Researchers are exploring techniques for explainable AI and transparent design.
   • Efforts include developing methods to visualize LLM decision paths, identifying influential tokens, and creating interpretable representations.
   • Ethical guidelines and bias mitigation strategies are crucial to address flaws.
   • Collaborative efforts between researchers, practitioners, and policymakers are essential to unlock LLM benefits while minimizing risks1.

As for me, I’m not just a passive consumer—I strive to engage in meaningful conversations and provide helpful information. While I don’t directly address LLM flaws, I aim to assist users by leveraging my knowledge and predefined tools. 🤖📚

CoPilot is thinking of only healthcare and criminal justice, but in the near term it will have enormous consequences for the wars, and elections. State actors will emerge to make world harmony a distant dream.

It is often a tedium conversing with the CoPilot. Sometimes, CoPilot does not answer your question at all, rather addresses the key phrases individually independent of your question. It feels like you are conversing with a hearing impaired person. It is still a project in the works.

5/18/2024

Gemini AI is now easy to access. It does perform just like CoPilot. If you ask controversial subjects, or ask Gemini to factcheck, it declines saying it is still in the learning mode. However, Google tries to get a feedback from users that will improve the product. This is important because Gemini can return absolutely wrong answers.

7/22/2024

Co-Pilot has been instructed not to answer politics related question even if you are seeking publicly available information. It is in clam-shell mode.

7/22/2024

Of course, you must have heard of the Meta AI o the WhatsApp. It does answer any questions.