About author
Hi, I’m Dennis. I started making electronics from my first year in university and then continued working as Electronics engineer (EE). I went from making hobby robotics PCBs to the automotive grade ECUs. I have about 4 years of commercial electronics development experience by now. I believe, I gained decent understanding of the EE processes, and I see how some of them can be improved. My writing is focused on electronics field. It might not apply to your field of expertise. But it might, so it still might be insightful for you.
Preface
ChatGPT is a thing for 2 years already. Let’s start big and ask why is it still not doing all engineering jobs. From what I saw at work there is the hype, there are attempts, but no successful deployments so far. Why? In my opinion, because you can’t trust the thing. It does make a lot of mistakes and in electronics unlike programming you can’t copy/paste and clik the run button to see if it works. Cost and time-expense of a prototype run is high and what EEs (electronic engineers) do is improving the chance of a project to work the first time, or fix it quickly by rework if it dosent. Introducing hallucinating tool in the loop is not helping here.
How do we reduce number of mistakes in design
EEs have developed a set of practices over the years on how to detect and avoid design mistakes. They are some rules of thumbs, components libraries, simulations, calculations, analisys documents. As design itself is done by human it’s impossible to avoid mistakes, so we do paperwork to reduce their amount. The bigger the project (the bigger the failure cost) the more of such paperwork is being created.
Where is the source of truth
Luckily for this writing the source of truth for all of the analysis, documentation, and simulations we do as EEs is in electronic components datasheets, and application notes from microchips manufacturers. It’s not like EE is a blessed person providing you with error free experience. It’s just a person who is effective at reading the datasheet and using this info to make and check the design.
How AI can help
I think the first step in adopting AI assistance by EEs is gaining trust in the tool.
Whatever the LLM outputs should be based on the holy grail of knowlege – source of truth pdf document. ChatGPT at present is able to reference the entire document but this is not helpful in quickly checking the LLM output correctness
Introducing FactFrame
This tool uses special PDF parsing and LLM prompting techniques to allow LLM to remember where exactly in the document the information is taken from.
FactFrame technology can be usefull for any industries where the cost of mistake is high and the knowlege is mostly stored in the “source of truth” documtents.
It allows to speedup “paperwork production” by LLM while still leaving humans full control and visibility over what had been generated.
This lays a foundation to use LLMs in electronics engineering.
Let’s discuss particular applications and tools in the next article.
Leave a Reply