FC components and AI (ideas?)

[MJU20]

We can no longer ignore the fact that AI has become a powerful tool.
This is also true for electronics people

It now seems to me that we can also expect an increase in the creation of components for Flowcode.
Before AI, I used to read datasheets thoroughly to understand how to communicate with a new component using, for example, SPI or I2C.
Now, I simply ask an AI robot.

As an example, I took the datasheet of a (simple) component: the 24AA00T-I/OT I2C EEPROM.

In the past, you’d read through the documentation and note which commands the component expected in order to function.
Now, I paste the datasheet and immediately ask what command I need to send to use the component.

I ask, without too much beating around the bush, which commands I need to use to write a value to the EEPROM or to read one.
AI then gives me a clear sentence listing the commands I need to let Flowcode generate in order to make it work.

What are your thoughts on how we can use AI in Flowcode or in electronics as a job or a hobby?
Where do you see areas for improvement?
Should Matrix use this as a standard feature to quickly create components?
Can we, as customers, use this ourselves?

I just want to hear all your ideas and thoughts.

[SpeedPIC32]

Hello,
I am currently experimenting with Gemini.
It works well if the question is specific enough. You tell the AI that you are working with Flowcode 10xx, which PIC is being used, and which AD converter should be operated, for example, and the nice thing is that you can create macros and the AI gives you, for example, the hex code of the specific commands in the correct order. If there are errors, solutions are offered based on a current data sheet. The company Mikroelektronika offers the Necto Studio programming environment.
AI is integrated into it. You can try it out. It works quite well for the C language.
Best regards to you all, including the AI.


Translated with DeepL.com (free version)

[mnfisher]

There are still some very disturbing 'errors' in the output.

Sometimes the results are very good - and others - not so much. For example - I asked which pin to use for an interrupt - and it gave the wrong one, and then argued that I must be using an out of date datasheet (and old silicon). When I asked it for a link to the 'new' datasheet - it gave a link to the same one and even after I clipped and uploaded the relevant pin-outs section it still maintained that I was using an old datasheet.

So - still need to tread with care

It is getting better remarkably quickly though - and when it's on form it can be impressive!