Notifications
Clear all
Artificial Intelligence
12
Posts
4
Users
4
Reactions
216
Views
Topic starter
2026-07-21 1:21 pm
I've never really, intentionally sat down to use AI. At my age, its more important to use my brain than taking the easy way. My slow adoption has mainly been a result of Google jumping in when I post a question on Chrome.
My background in software development was a good acid test. I'd pose it to write some code I'd already written for a specific purpose. It did a respectable job even to the point of using proprietary functions for Arduino IDE C/C++, Microsoft C# Window Forms and Linux C/C++.
This gave me a false impression when I turned to the hardware side of projects. Here, it would have destroyed hardware if I hadn't dug a little deeper.
So beware! Do some reality checks. Come to DroneBotWorkshop and get confirmation. 😎
My hardware knowledge is rudimentary. I can handle P=Vi, V=ri and can generally keep the "red" and "black" sides separated without making smoke. I try to not ask the same questions twice of the hardware experts on this forum and have learned or at least bookmarked specialized circuits. But, in general, I tend to do cookbook, monkey-see-monkey-do type hardware design. The Internet is replete with examples. Some good, some downright damaging. I'm guessing AI using this wealth of information has the same problem with what's good and what's bad.
I started posting a Google search:
Using SCT-013-020, ADS1115 and ESP8266 for house AC current
It initially said I could connect the passive current clamp (SCT) directly to the analog to digital converter (ADS), connect this to the ESP8266 and program it to use differential sampling. I have to admit, it did work great, but I was very lucky. I was testing it out with clothes iron that draws about 8 Amps. I started reading the ADS datasheet and realized the negative portion of the 60Hz wave goes below the allowed ADS limits and would have destroyed it, if I had a load near the 20A limits.
On the good side, if you leave the browser page open and continue asking questions, it remembers what you previously asked... even over days. Pretty cool! I pointed out this design flaw and AI admitted:
You are right! Here is the revised circuit that will flawlessly serve your needs.
Well now! That's not very comforting that AI is relying me (or some other newbe) to correct it.
I have problems with capacitors. They don't intuitively make sense to me. I know small ones can be used to filter out noise and large ones can store electrical power to handle surges, but knowing what size to use when and where doesn't jump out at me like it does to the hardware experts here on the forum. I asked the same AI browser page the general question:
What capacitors do I need for making this more robust and reduce noise
It gave me a new circuit and extensive explanation adding approximately 14 capacitors to the circuit that weren't there before. I'll post this circuit in another topic about the project soon, but for this topic, I noted two flaws in the circuit. Here is one portion of the circuit running between the ESP8266 powering the ADS's:
This seemed wrong to me, like, what would the capacitor do running in parallel to the power line?!? So, I posted this back to AI:
I want to clarify VCC is the 3.3V coming from the ESP8266 and VDD is the input on the ADS1115? Also, the 100nF power supply decoupler is on the line, not running to the GND wire?
Here is AI's response:
"Your observation about the 100nF decoupling capacitor is critical—the text schematic was a bit ambiguous there. The decoupling capacitor must connect across the power line and the ground line. It does not sit inline (in series) with the power wire."
Hmmm! 🤔 I don't know if I'd used ambiguous here? Again, I'm struggling with, what little I know is barely enough to correct the AI and I'm concerned about what I don't know about what I don't know! 😆
You should too! Come to DroneBotWorkshop and get a reality check.
robotBuilder reacted
2026-07-21 7:28 pm
Welcome back to the forum it has been quiet recently. Although AI can answer a lot of technical questions and is friendly and "takes an interest" in your project it can get it wrong and you need to navigate that reality. I have found it helpful in coding and hardware but also in many other areas of my life with good advice and chasing up answers for me about anything.
Inq reacted
2026-07-21 11:16 pm
There have been plenty of times that I have to correct the AI i am using. So much so that I am having to change the way i use it. It seems to be able to offer up ways to troubleshoot problems really well in ways that i don't. There are times when I have to remind it that we had already discussed a subject. Or times that I told it something and it forgot it and i have to repeat it. It will come back with "yes, you already told me that and I should know this" type of response. So I don't know. I am much less confident in it or it may be i am not using it correctly. I only use it in chat mode as kind of a mentor.
It does help but I feel it sometimes slows me down and I still need to read the docs so I can catch the times it gets it wrong. I think maybe its good for figuring different ways to do the same thing and its up to me to pick which way to go.
I still think it is an overall advantage and I am trying to figure out why it seem to miss the obvious and so helpful at other things.
Inq reacted
2026-07-22 6:35 am
Prompts are the search queries into existing knowledge so the quality of answers you get depends heavily on how you ask your questions. You will find lots of articles and videos on how to write effective prompts.
Topic starter
2026-07-22 10:41 am
There are times when I have to remind it that we had already discussed a subject.
So far, I've never run into that issue. If I keep the same browser page open, it has retained what was said earlier... even over days. I haven't explored much longer than 3 or 4 days, but if I close the page and open a fresh one, it starts completely over. I remember when I was doing web programming professionally, having a "State Engine" that retains even the last interaction was a hard problem in a generalized way. I have to wonder if they're doing that at some server keeping the whole transaction or are they storing it on my box in session or cookies type memory. Either way, it's a huge problem.
2026-07-22 1:18 pm
When chats become too long i read that AI starts losing information. I ask the AI if the chat is too long and if it is I have the AI write a summary that i copy into the new chat. I use Claude and they have a projects mode where you can upload files which are available to all chats in the project. I am not sure if one chat can access the other chats in the project.
In Claude there is a place you can set up how Claude should respond which can make a big difference in how the AI responds. Basically you can set the personality of the AI. The responses i was getting were too long and I set up in the instructions to be concise. For example here is an instruction to Claude that it should follow :
Core Rule: Be extremely concise. Give short, direct answers. Do not write introductory pleasantries (e.g., "Certainly! Here is the answer:") or concluding summaries.
Format: Avoid multi-page explanations. Use 1-3 bullet points unless asked for more details. Answer the exact question asked without expanding beyond the prompt's scope.
Like anything else the more i use it the more I'll learn how it works.
Topic starter
2026-07-22 1:44 pm
That's pretty cool. I definitely haven't considered all the alternatives to Gemini and ChatGPT or any of the subscriptions.
I do find it fascinating the power of it. One of my project involves using NodeRED. Its predominantly a graphical programming language that can integrate logic, MQTT, database, and GUI browser pages. I was having troubles describing the symptoms in relation to the code to Gemini. I finally gave up and dumped a hundred lines on the "Ask anything prompt". I fully expected it to flip me off. I was floored Gemini with such an obscure language, with multiple sets of 3rd party plug-ins was able to find my error and tell me how to fix it.
But... I can also see why studies are already showing kids having reduced cognitive abilities.
We're heading toward Wall-E or Idiocracy.
2026-07-22 2:38 pm
Yes, AI is really good at separating complexity into what
really matters and tunneling down and
pin pointing where the problem could be. I am normally
shooting in the dark and hoping I get lucky. Kind of nice
to have that as a resource when you don't have a team
behind you.
I have so many gaps when it comes to things i know well and
things i am only familiar with and things I do not know
anything about.
Claude is like my knowledgable programming pal that isn't perfect
but knows a heck of a lot more than me. I think I have learned
at a quicker rate than without AI. That's the way I like to
use it. Trying to keep my brain cells healthy and working :).
I hope i don't get to dependent on AI and take the time to
try to solve things myself before using it and learning from
it rather than AI being my goto problem solver.
2026-07-22 6:27 pm
Yes. AI models make terrible mistakes sometimes. We always need to cross-check the info.
2026-07-22 8:38 pm
I hope i don't get to dependent on AI and take the time to
try to solve things myself before using it and learning from
it rather than AI being my goto problem solver.
Some people like to try and understand everything and solve problems themselves. It is not being smarter than anyone else it is just getting joy out of figuring something out for yourself even if it has been done before or just the joy of understanding how the "magic" is done.
However it is not the fastest way to get an outcome! A complex society would collapse if everyone had understand everything before using it. We just have to know how to use it. We specialize and work together to make, fix and maintain complex systems. We use computer software to amplify our abilities. We can choose our level of understanding according to our needs.
My use of chatGPT for coding is as a tutor. I always ask for an explanation of the code and if I don't understand it I use it as a function like I might use a function from a imported library.
The reality is chatGPT has saved me and others a lot of time and money but that is another story.
Lee G reacted
2026-07-22 10:52 pm
I agree with what you posted. The project i am on currrently has taken me over a 1.5 years i am still not finished. I thought I was a done a few times but there is some thing happening every once in a while where my server takes a long time to respond to a request and I am trying to hunt that problem down and its not in my code.
I am not usually in a hurry and this is just a hobby. If someone was paying me to get something done I probably would be fired by now. The list of projects I want to do is getting longer and I def want to finish this one up just so I can work on something else.
For me its more about getting to know how things work and in this hobby its mostly software. There are so many ways to use AI but I holding off using it at a high level and do all the code for. I don't think it would be that satisfying.
Lucky for us hobbyists we at least have AI to turn too when we are stuck on something or the feeling that we are not on our own as AI can seem like having a buddy with you on the project. I did have AI tell me if I don't start using git it was not going to help me. I thought that was pretty funny.
Cheers, robotbuilder, I hope your making good progress on your robots.
2026-07-22 11:48 pm
I did have AI tell me if I don't start using git it was not going to help me. I thought that was pretty funny.
I remember with my first home computer, a TRS-80, there was a simple chat bot called Dr ELIZA. Providing you kept within its subject matter is seemed very "human" yet the BASIC code was short and simple. I remember my sister laughing at one of its responses. The author of the program had his secretary ask him to leave the room so she could have a private chat with it! People do tend to anthropomorphize machines be it their car or their robot vacuum cleaner.
Cheers, robotbuilder, I hope your making good progress on your robots.
Yes thank you.