Equipment Reliabilty, Automation and AI

[Bruce Cloutier]

My most recent thermal experience has been with DDR5 ram sticks. I don't think I'll ever buy Crucial memory again (at any price and they are, typically at the top tier of the models they make). I had computers crashing left and right with them. Swapped in some Amazon special "A-Tech" sticks...problem solved. Stress test it for hours...no problem...cool to the touch even. Same speed RAM, same size, same everything.

I don't know how TL/Eprad is doing in their non-cinema industries but I suppose okay. They've always targeted smaller production custom design/build sort of manufacturing. We have not had problems getting products from them (automation or dimmers). They did, semi-recently, rev the CPU board (to version 2), at least partly at my request, for features (which, effectively doubles, if not more, things like Macros, RDIs, and the number of instructions).

I don't want to get too far off topic on this thread. It should be mostly about Q-SYS rather than a particular manufacturer. Eprad has come up here due to the plugin. Not vice-versa. Furthermore, part of the plugin discussion is that the use of AI (Claude) was key in both of the Dolby DSS Server and the Eprad plugins. They were the very first time I ever tried it. The speed at which i've been able to orchestrate the plugin, adding features or making the user experience better (by my perception) has been pretty remarkable, in my opinion.

Breaking this out from the Q-SYS thread - especially since you mentioned AI...

I have been working with ChatGPT. I have trained it with knowledge of the JNIOR and a lot of the history there. That LLM is pretty amazing. AI is absolutely useful. I still feel that it is dangerous. Only because I don't trust that the human side of the interaction can maintain the required grip on reality and not be sucked into it. But as a search tool, capable of finding exactly what you meant to have searched for (not blatantly as entered) it is incredible.

It can make the coding process very efficient. I am not sure that the code it generates is necessarily efficient. It doesn't matter what language you are coding and how good the IDE might be, you still draw blanks when you need to remember what methods are needed to do a thing. AI comes to the rescue. You can't fault it for that. I let it remind me or show alternative approaches, but I can't bring myself to blindly cut and paste. Not anything significant. I think ultimately that will get us into trouble.

Basically we shouldn't give AI responsibilities or enable it to take action on its own. That might work for a while. But it will go horribly wrong. I am not talking about the Terminator. Although you can see how that is a feasible outcome now. It is more like things where it offers customers a free car replacement for their problems or leads a youth to make a disastrous choice.

It also writes better than I can. I write an article and feed it to AI for "opinions". It's critiques are dead on it seems. I then have trouble not agreeing with it and so do the editing. It not only can clarify the points I am making but remind me of some things that I really should have included. Ugh. It starts to not sound like me. I won't just ask it to write the article from scratch. It does suggest good outlines tho.

Anyway...

Controls business seems to be in a recession. I think things at the start of the year threw everyone into shock. Projects and investments were delayed pending everything. At the same time the supply chain has been disrupted both by the economy and the international situation. Add the effect that AI has on the demand for hardware and things that should be forever parts in electronics are just going out of production (EOL). We're going to have to revise JNIOR PCBs when production picks back up. Some parts I won't be able to get. Some parts I won't be able to cost-justify. I have always held the line on price increases. That ship has sailed.

I know that we can't be political here. It might be helpful to get others perspective on things however. We may need to cope as a group.

[Harold Hallikainen]

On AI, I've been using Google's Gemini AI since it comes with my subscription to a big Google Drive. I've been having it help write an HTTP client for a PIC32MZ and a WINC1500 Wifi module. Gemini DOES know a lot about the WINC1500 interface code. I've been copy pasting debug output back into Geminii. It comes up with reasonable explanations as to what is going wrong and give me the "final ultimate fix." I've now had several dozen of those. I also have to keep reminding it of things like we're using HTTP1, don't worry about skipping HTTP headers, etc. It is truly amazing. It has read all the documentation that I have not yet read. But, I told it at one point that it appeared to be throwing code at the wall to see what would stick.

[Steve Guttag]

Yeah...AI is both remarkable and scary and, for sure, there needs to be guardrails on it. I don't know if you looked at my screen shots...but on the Setup page, I put a notice that it was built with Claude AI. I'm not going to take credit for work I didn't do. However, that would change, I would think, if I were selling the plugin as a product. Then, Claude, in my case, is just a "work for hire" just like if I sub-contracted out to someone to code. In the end, the company/business' name is on the product and the responsibility for supporting it.

AI, at this point, definitely is not perfect. There is a fair bit of "me" in the plugins . It would get stuck on things and I would need to suggest how to make it work...and like pushing a car that is stuck in snow...it gets going again and you work through the problem. It is VERY iterative. But, if I had a coder that I was paying...it would be iterative too. So, in that case, I don't see any difference (aside from cost...in AI's favor). Additionally, AI is always ready when you are (and whatever you budget for its time...I'm on the thrifty side).

But, aside from Q-SYS...my mind is a blaze with the possibilities of what I could do with it. Where I really think AI could do the most good would be in the field of medicine. Something that can work 24/7 on a problem...never get tired (unless the power goes out). Also, long-term planning on a global scale.

I do fear the total dependence on it. You know how to code so you can work WITH it to improve things. What happens when nobody learns to code because, well, AI does it? How good are people's math skills, really, since calculators/cell phones do it all? I guarantee that the current generation is worse than mine as we are worse than my parents.

Is writing about the information or the personality? I think both. But if people have AI do the writing...then, pretty soon you'll write it with AI. I'll have my AI read it and then give me the TL;DR version.

The ability of AI and AI like things to impersonate people, the likeness/sound needs to be put to a stop with incredibly severe penalties too. The amount of harm that could be done with that cannot be overstated.

[Marcel Birgelen]

At this point in time, AI is a power tool for engineers who know how stuff works. You can easily outsource the grunt work to an AI, as long as you check what the thing produces. The bottleneck though is often the latter, as AI is pretty prolific and the amount of code it produces can easily overwhelm the best engineer or developer.

AI works best when hold tight to a leash: when you tell it exactly what it needs to do and check if it actually did what you wanted it to do. If you don't do that, the thing will hallucinate complete APIs and specifications for you, alongside the modules to talk to those sand castles.

For those who think they can code now, without a single knowledge about the systems and technology involved, this whole AI thing can be a shot in both feet simultaneously... I've heard many times already: I don't need any developers or engineers anymore, I've got <insert vibe coding thingy here>.. Wait until the thing breaks the first time and your favorite AI thingy is stuck in a loop for answers...

Another thing I'm really worried about is security. While AI is pretty good at writing stuff that eventually does what it should do, one major problem is security. I've already seen major issues with this, where a piece of vibe coded software worked beautifully, but the API simply allowed everybody to elevate themselves to admin level...