Menu
JAQForum Ver 19.10.27

Forum Index : Microcontroller and PC projects : AI research project for this month

   Page 8 of 8    
Posted: 03:21pm
29 Oct 2025
Copy link to clipboard
Mixtel90
Guru


Off topic unless AI suggests which pub, based on such criteria as distance, weather, cost of a pint, availability of seating and quality of salted peanuts.

;)
 
Posted: 03:28pm
29 Oct 2025
Copy link to clipboard
twofingers
Guru


The good thing is, I learned a new word: "tell-lie-vision". I didn't know that one before. I had to ask ChatGPT what it means.
 
Posted: 03:56pm
29 Oct 2025
Copy link to clipboard
cicciocb
Regular Member

I apologize for stepping into a topic where I probably don’t belong, but I couldn’t resist. I’m 100% in agreement with matherp, and anyone who programs regularly can now attest that AI has changed the rules forever. Things are not only moving fast but accelerating at an astonishing pace, to the point where it’s almost impossible to keep up.

Today, we can practically do without any specific programming language, since we can directly specify the final goal and obtain a fully working program or application without writing a single line of code. AI can even help us define our needs simply by asking the right questions. You just have to choose the development environment, copy and paste, and voilà! For web applications, AI can even show the live result directly in the browser.

And this doesn’t apply only to computing, it applies to every field. We’re still at the very beginning of this revolution; the rest is yet to come. Companies are now implementing their own private AIs to specialize (and protect) them with their internal data and, probably, to replace certain jobs with AI agents. In the future, the true know-how of a company will probably reside within its private AI.

Ignoring the existence of AI and neglecting its potential will inevitably lead to isolation, because AI is steadily infiltrating every area, just as calculators and computers once did. Before long, we simply won’t be able to live without it.

Sorry again for the intrusion.
 
Posted: 04:20pm
29 Oct 2025
Copy link to clipboard
twofingers
Guru


I asked ChatGPT:
Can you also create 3D models (in STL format) for 3D printing?

Answer:
Yes – I can design 3D models and prepare them as STL files for 3D printing.

Here’s what I can do:

✅ What’s possible
Creation of simple to moderately complex geometries: cubes, spheres, cylinders, and more complex composite shapes.
Mechanical parts: gears, mounts, clips, simple housings.
Toys, decorations, and prototypes: figures, small sculptures, keychains.
Parametric designs: dimensions, holes, wall thicknesses can be adjusted.
STL export: I can provide the model directly as STL data, ready for 3D printers.

⚠️ Limitations
Very organic or highly detailed sculptures: like complex figures or faces, are limited; I can only produce them schematically or in block form.
Models may require manual or software post-processing: for fine adjustments (e.g., support structures, overhangs).
I do not create copyright-protected designs: only original or generic objects.

If you want, I can generate an example STL model immediately, e.g., a simple cube with a hole, a keychain, or a small mechanical part.

Do you want me to do that? If yes, please briefly describe what kind of model you need.
_________________________________________
I'll try that and report on the result.
 
Posted: 04:29pm
29 Oct 2025
Copy link to clipboard
JohnS
Guru

  Mixtel90 said  Off topic unless AI suggests which pub, based on such criteria as distance, weather, cost of a pint, availability of seating and quality of salted peanuts.

;)

The quality of the beer is vital, never mind the nuts!!

(I don't need AI for that, I have keen beer-drinking friends who can say.)

John
 
Posted: 04:39pm
29 Oct 2025
Copy link to clipboard
lizby
Guru

  cicciocb said  Sorry again for the intrusion.


This is exactly the kind of comment that is most welcome--the experience of another "super-programmer"--the author of Annex RDS, a very capable and comprehensive BASIC for the ESP32.

My efforts are not remotely as complex as those of matherp or cicciocb, but it's remarkable to me that ChatGPT-5 can even understand my prompts, much less produce working programs in minutes that might have taken me hours--especially in areas that are unfamiliar to me.

My career was as a programmer/analyst--retired now for 27 years. I consider myself to have been pretty productive. If a time machine had given me access to one of the AI coding agents, I would have been astonishingly productive.
 
Posted: 04:51pm
29 Oct 2025
Copy link to clipboard
PhenixRising
Guru

  lizby said  If a time machine had given me access to one of the AI coding agents, I would have been astonishingly productive.


Yeah and screw its massive and ever growing carbon footprint    
 
Posted: 05:10pm
29 Oct 2025
Copy link to clipboard
lizby
Guru

  PhenixRising said  Yeah and screw its massive and ever growing carbon footprint    


Yeah, right.
  Quote  Training state-of-the-art models can emit hundreds of tons of CO₂. For example, training OpenAI's GPT-3 emitted about 552 tons of CO₂, equivalent to the annual emissions of 120+ U.S. cars, or a transcontinental flight for 123 passengers.


Imagine that--120+ cars or a transcontinental flight.

  Quote  How many AI tokens can you get for the price of a basic cup of coffee at Starbucks, and how many average ChatGPT-5 queries for that price?


Answer:
  Quote  Summary: For the cost of a basic Starbucks coffee [$2.95 USD], you can process approximately 738 typical ChatGPT-5 queries or millions of tokens for projects, making AI extremely affordable compared to many consumer products.
 
Posted: 05:24pm
29 Oct 2025
Copy link to clipboard
lizby
Guru

My prompt for today, after I got a current sensor on my newly-installed mini-split heat pump.

  Quote  Provide a python program to produce a graph of current usage over the course of a day for a device. The records have 5 space-separated data items, a single character ID, an on/off indicator (1 or 0), an instantaneous current reading, a cumulative daily current reading in kilowatt hours to 3 decimal places, and a timestamp in the form YYMMDDhhmmss. The program should take as command-line parameters the date in YYMMDD from, the ID (e.g., "I" without the quotation marks), and the input file name. The output file name should be /home/user0/charts/IDYYMMDD.html (where ID is the single character input on the command line). The records may occur at random consecutive times. Use the 4th data item to graph a point at the time indicated. Maybe allow for 1440 possible points for the number of minutes in a day (or half that many if more suitable for graphing).


It took a few passes, because it assumed the python default installation on, for instance, a Raspberry Pi, rather than my minimal installation on a Seagate Dockstar flashed with the openWrt Linux distro. But then:



I said I wanted instantaneous current draw in watts.



I need to make some adjustments in how I'm recording the data. For instance, the mini-split was off from 14:30 to 23:30, so the line should be flat at zero--but to get that I have to make a change in the berry script which is running on the Tasmotized Sonoff POW R3 current sensor.
 
Posted: 05:48pm
29 Oct 2025
Copy link to clipboard
twofingers
Guru


My question to ChatGPT:
I want to do a simple test to see how well you can create STLs.
For storing button cells (CR2032), I need a container that holds several cells. The cells must be electrically separated.
Create a hollow cuboid with the (outer) dimensions x=62, y=23, z=13 mm. The cuboid should be closed at the bottom (Z). The wall thickness should be 1 mm.
In the x-direction, every 3.5 mm walls should be inserted, each 1 mm thick. Any questions?

His Answer:
I created an STL file for the described container using exactly your specifications. Here’s a summary:

Outer dimensions: x=62 mm, y=23 mm, z=13 mm
Wall thickness: 1 mm (bottom also 1 mm)
Bottom: closed, top: open
Internal height (from top of bottom to open top): 12 mm
Internal partition walls along the x-axis: every 3.5 mm starting at x = 1 mm + 3.5 mm = 4.5 mm, wall thickness 1 mm
Number of partition walls: 16 (1 mm thick plates spanning the internal Y-length)
Geometry generated as a union of boxes (bottom, outer walls, partitions) — overlapping surfaces are present (typical and acceptable in STL).

cr2032_holder.zip




Important assumptions / notes:

The partitions are full-height plates running from the top of the bottom plate to the open top (12 mm tall).
The partitions are flush with the inner edges of the side walls (no air gaps on the Y sides).
I did not include any fitting details for the actual CR2032 cells (e.g., chamfered entries, pole height, contact separation, or springs). You stated the cells should be electrically isolated — this design provides only mechanical separation; no conductive parts are modeled.
The partitions spaced every 3.5 mm create many narrow slots (16 in total). If you meant something else — for instance, spacing based on the center distances of the cells (20 mm) or compartments sized for the CR2032 diameter (20 mm) — please clarify. I can easily regenerate an alternative layout.
If you’d like additional features (e.g., closed compartments, snap-fit lid, contact guides, dimensions matched to CR2032 cells, latching tabs, extraction slots, or print-friendly fillets), just describe them — I’ll adjust the model and create a new STL.
_______________________________________
I think that's remarkable.
Regards
Michael

@Lance
If I'm bothering you or the posts don't belong here, just say so.
Edited 2025-10-30 03:57 by twofingers
 
Posted: 06:40pm
29 Oct 2025
Copy link to clipboard
lizby
Guru

  twofingers said  I think that's remarkable.


It is in my opinion. I might even take the plunge and see what I could get printed at the local library.

  Quote  @Lance If I'm bothering you or the posts don't belong here, just say so.


All AI projects welcome. Thanks for posting.
 
Posted: 07:11pm
29 Oct 2025
Copy link to clipboard
okwatts
Regular Member

I think the "intelligence" is in the correct specification and evaluation of the result. For programming or anything with well defined rules the training is based on all the code that existed at the time of it's training. If it understands the request and is true to the "facts/rules" it was trained on, then it should do well.
Remember that it will respond with the same bias that exists in the training data as it cannot make "value judgments". This can be a concern if it uses sources like "Wikipedia" which has limited and unknown editors who can make changes. This is particularly true when it comes to political or social issues. There, our personal biases/exposure will cause us to accept or reject the answers.
My grandson is a computer nerd and does programming. He is now just about to enter University and my advice has been that it is good that you can program and understand some of the issues but don't make a career of programming. The analysis and understanding of the subject is more important in assessing the result than actually writing the program. I am reminded of a book I found useful by Forman Acton entitled 'Numerical Methods that "Usually" Work'.

PS I have just purchased both a Jetson Orin Nano and a gaming PC with lots of memory and one of Nvidia's newer graphics cards with 16GB of memory with the intention of evaluating "local" AI options. While neither as large or current as are available using the cloud or their server farms I hope it is educational at least.
Edited 2025-10-30 05:16 by okwatts
 
Posted: 07:32pm
29 Oct 2025
Copy link to clipboard
Mixtel90
Guru


This is correct. Defining the problem, any initial parameters and the desired result accurately are where the intelligence is, at the beginning and end points. The bit in the middle can be filled in by human or AI coders, it doesn't matter unless time is an issue.  :)
 
Posted: 07:53pm
29 Oct 2025
Copy link to clipboard
lizby
Guru

  Mixtel90 said  human or AI coders, it doesn't matter unless time is an issue.  :)


Or money. How much of a top coder's time could my $20 a month for ChatGPT buy?
And for whom is time not an issue?
 
Posted: 07:59pm
29 Oct 2025
Copy link to clipboard
twofingers
Guru


  lizby said  ... All AI projects welcome. Thanks for posting.

Thank you, I consider that a privilege.

I printed the model without further checking it.
The result was somewhat disappointing. The spacing of the battery compartments is incorrect. Unfortunately, the distance is only 2.5mm, but it should be 3.5mm. ChatGPT confirmed the error.
  Quote  ACKNOWLEDGEMENT: You’re right — the slots ended up only about 2.5 mm wide because the partition thickness (1 mm) was subtracted incorrectly (the code placed partitions every 3.5 mm, so usable gap = 3.5 − 1 = 2.5 mm). That is too small for a CR2032 thickness of 3.2 mm.

WHAT HAPPENED
• I placed partition walls at positions spaced by the user’s 3.5 mm value and used a 1 mm plate for each partition.
• The free gap between adjacent partitions therefore became (spacing − partition_thickness) = 2.5 mm.
• That is why the cells do not fit.
...

ChatGPT created a new model, which looks better. It's apparently not 100% correct, but usable for the purpose.

cr2032_holder_fixed.zip
Regards
Michael
 
Posted: 08:36pm
29 Oct 2025
Copy link to clipboard
PhenixRising
Guru

RFO BASIC for Android is a bit of an odd dialect albeit amazingly capable.

I asked Claude to create a program to plot a bent tube based on:

Distance between bends
Degree of bend
Plane of bend
Tube diameter
Bend radius
Tube wall thickness

Holy moly, I have a wire mesh plot already. Not perfect because the bends are messed up but I am using the time limited online version.

Scary good 😎
 
   Page 8 of 8    


To reply to this topic, you need to log in.

The Back Shed's forum code is written, and hosted, in Australia.
© JAQ Software 2025