Are you using AI?

This is the part of AI, I don't find helpful. Investors and founders who think they've found the next unicorn.  As for fl*x trying to flex... Good luck to them. Let them burn their cash...

No, not really. – Not yet.

Why? – I hate data theft and identity theft.

It's so much easier with any kind of AI.

Just record a few words you say, and you can use those stolen words to conduct business or commit other crimes.

And what if your face was also recorded? Well, then you're done for – the damage done can never be repaired.

Then there are arguments here – "Help with data analysis, help with creating tables, help with writing programs, etc..."

Help with writing programs?  – Well, then you might not be able to check if the AI ​​is programming a "black box," and

suddenly you have a nasty virus in your program that will never be detected. Cool! 

Okay, okay… one more question: Can AI really distinguish between true and false information?

“Yes, AI has a lot to learn.” – Sorry, but that argument is a bit dubious.

Yesterday, the news on TV reported that AI is already being used in pharmaceutical companies.
But the authorities are currently prohibiting its use because the AI ​​is mixing and suggesting poisons.

So, if no one checks what the AI ​​is actually doing, then medications will be produced that don't cure the sick, but instead cause even worse problems…

I really don't want to discuss this argument to death here.

So, YOU - the human must have every time the control over the AI.

The argument "faster" is not really the 1. argument.

Security and live is the domain.

Regards

Gerald

---

I find AI does help with sub-circuits (text, I don't ask it to draw circuits since I don't believe that capability is there yet), and also to compare specs and to suggest parts so I can dig deeper.

Just saw this unfortunate thing: https://news.ycombinator.com/item?id=48368121

No idea what it's about, but the theory there seems to be Adafruit may have tried a tool, and a firm doesn't seem to be happy about something. I do feel sorry for Adafruit.. personally I can't imagine they deserve legal threats over perhaps an AI tool but as mentioned I don't know the detail (and wouldn't be wise speculating much here, just leaving the link there in case anyone wishes to follow it there).

Most of this has been done with the GH Copilot integration  for VSCode with the model set to "auto" (which means that it will select the model based on my request, from the models available to me, which currently include several versions of Claude Sonnet and Opus and GPT-4 and 5).

I've tried selecting specific models for testing though, without really noticing much of an improvement.

Wouldn't this be another case where a written documentation would have been more useful in the long run?
A video is not searchable nor printable

I have code magazine subscription as well, it was packaged as a part of visual studio subscription. Unfortunately they don't mail the physical copy to India. That publication is useful for learning about dotnet/c# updates.

Which AI are you using for coding "assistance" ?

Thanks.

MK

Initially I watched a couple of videos... but afterwards, rather than rewatch I asked ChatGPT how to do individual tasks.

At work we've been encouraged to use AI for a while now, and in fact part of what I do at work is to leverage AI to build systems that automate analysis, create concise reports or centralize access to different repositories of information to make search and retrieval easier. And I know all too well the limitations and issues of most models.

But I'm not a fan of it. None of the solutions we have implemented with AI is perfect. They never go beyond "It works to some extent" and no amount of fine tuning, either by me or other engineers have led to consistent performance. Sometimes the responses are great, sometimes they are trash.

I've also been testing it for solving coding problems or helping with debugging issues and my results have also been mixed. Most of the time I already know the solution before I ask the AI, and half of the time it has spotted the problem right away. A few other times it has required guidance. But it has also wasted a lot of time throwing around random hypotheses that had nothing to do with the actual issue, and testing fixes that obviously did not work.

When tasked with writing a piece of code, it keeps adding methods and features that are not needed, or making changes to the code that I didn't request, and then it fails to revert them when asked. It constantly switches between "I'll ask for permission on every step along the way" to "I'm going to do this bunch of changes without even telling you". Maybe this is acceptable to some people, but for me it feels like babysitting someone and cleaning up their output to ensure it meets standards and it's not breaking anything.

Apart from coding I have also tried using AI to summarize and run comparisons between technical specs of different products or configurations, and I regularly have to correct it because it starts mixing up specs, or making things up. This has happened with several agents and models, so it's not like I just need to use a "better" one.

In the end, I don't think AI has any value for me. It still feels like a toy. And for all the rather unacceptable compromises, socially, morally, and environmentally that it's built upon, I don't feel like using it. Hopefully they can address those issues eventually, so it uses less resources and it's trained on properly authorized data instead of stolen intellectual property, but I doubt it. I don't think there's any incentive for AI companies to do that. They just want to build larger scale agents and market this to every person and company alive.

I know that for some people AI is tremendously useful, but that happens with tools in general. What is useful to some, might be completely pointless and a waste of time to others. It is all a matter of what people need, expect, what they can or can't (or care to) do themselves, and the kind of output they consider acceptable.

I'll keep testing different models on different tasks from time to time to assess their usefulness as objectively as I can. It is my line of work after all, and I need to know what's the state of it is. But so far I don't feel a drive to use AI in any capacity in my day to day life or at work, beyond the systems that I'm tasked with building.

So far it hasn't been a time saver, nor it has helped me in any way that an internet search (without AI) wouldn't have.

Coding forces us to balance readability against efficiency. Take passwords as an analogy. If I used a random string like "Q17u1}?3f'#P" as my password, it might look highly secure to some because of it's randomness, but a much longer plain-text sentence like "Hey!I'm_using-this_really-long_password+because_I_can" can actually be harder for a computer to crack due to its sheer length.

The lesson for coding is that human-friendly structure matters, not for security or sheer efficiency of performance or for the compiler, but because we need it to understand the logic and ascertain whether it will achieve the desired outcome or not.

As an AI agent reminded me this morning, modern compilers don't care where or how a variable is declared; clean code is entirely for the human architect.