CHATGPT & AI Models are Here to Stay

[Ed Panek]

December 2017 I was working at the Louis Stokes VA in Cleveland Ohio. It was lunch so I was surfing the Internet. I am a huge chess fan so an article came up about a new AI model developed by a branch of Google called DeepMind. It had been trained only on chess games. It was given NO GUIDANCE on how pieces rank in importance. It was only provided the legal moves of each piece, starting position and win draw loss info. It just played itself over and over and over. Then those games would be trained on, a new model released and then back to millions of games played. This happened in only a few hours. In a few hours AlphaZero (the "program") had exceeded all human ability in chess by a thousand ELO. I was impressed. But I was skeptical.

Its 8 years later, and progress is still being made. I'm not ready to let AI run my company but its become a really helpful tool. And its only getting better. I have started insisting that my employees learn how to use it. Why? Because regulators are using AI now. The FDA has an AI agent that reviews 510(k) submissions. It's gonna be everywhere.

So, are you using an AI assistant to help with work or personal life? What creative ideas do you have for it?

 

[ScottK]

December 2017 I was working at the Louis Stokes VA in Cleveland Ohio. It was lunch so I was surfing the Internet. I am a huge chess fan so an article came up about a new AI model developed by a branch of Google called DeepMind. It had been trained only on chess games. It was given NO GUIDANCE on how pieces rank in importance. It was only provided the legal moves of each piece, starting position and win draw loss info. It just played itself over and over and over. Then those games would be trained on, a new model released and then back to millions of games played. This happened in only a few hours. In a few hours AlphaZero (the "program") had exceeded all human ability in chess by a thousand ELO. I was impressed. But I was skeptical.

Its 8 years later, and progress is still being made. I'm not ready to let AI run my company but its become a really helpful tool. And its only getting better. I have started insisting that my employees learn how to use it. Why? Because regulators are using AI now. The FDA has an AI agent that reviews 510(k) submissions. It's gonna be everywhere.

So, are you using an AI assistant to help with work or personal life? What creative ideas do you have for it?

Coincidentally I just sat in on a training class on Microsoft's Co-Pilot. Since my company partners with MS that's our company standard. The training was all about how to make good prompts... The class was really geared towards marketing people. I don't see much application in what I do today, though I could see using for proof reading or looking for contradictions in procedures or work instructions.

When ChatGPT was released to the public I was in the midst of an ISO9001 implementation and I used it to create some quality policies as points of discussion and they honestly were not that bad.

 

[Nichole F]

AI still has a long way to go at least in the Regulatory field. Just for kicks, I asked ChatGPT what the proper EMDN was for one of my company's devices. I already knew what we had assigned it and what our Notified Body had accepted. Long story short, I spend a good amount of time telling ChatGPT it had provided an incorrect EMDN again, and again, and again. Some of the EMDNs it provided are not even real numbers within the nomenclature.

I do have to give it credit for polishing up a written piece. I have started to use it to refine memos to file or customer requested information letters.

 

[Ed Panek]

AI still has a long way to go at least in the Regulatory field. Just for kicks, I asked ChatGPT what the proper EMDN was for one of my company's devices. I already knew what we had assigned it and what our Notified Body had accepted. Long story short, I spend a good amount of time telling ChatGPT it had provided an incorrect EMDN again, and again, and again. Some of the EMDNs it provided are not even real numbers within the nomenclature.

I do have to give it credit for polishing up a written piece. I have started to use it to refine memos to file or customer requested information letters.

Ive used it sucessfully for things like "Read this document and prepare a 15 slide powerpoint outline" Or "Read this document and create a 10 question quiz. 3 easy, 3 medium and 4 difficult"

 

[Miner]

Have you ever cross-checked ChatGPT against Copilot? They do give different responses. Sometimes one is better, sometimes the other.

 

[Tidge]

We've had some recent success with AIs preparing some materials (based on our work) to supplement learning materials. It isn't terrible, and when it can pull from not-uncommon examples it seems to do well. Outputs required 100% review, because it would occasionally include some things that were not close to correct. I didn't like it for the quizzes, but that is probably just my past experience preparing exams to test different levels of cognition showing.

The other area where I find the AIs to be helpful is with first-order 'help' for commonly used Microsoft Office products. It used to be checking a couple of pages of results to find out how to do something (I think should be simple) in Word or Excel but is hidden deep on some other menu, tab or option.

 

[ScottK]

I'd be interested in see what ChatGPT gives to this prompt...

summarize the novel "Neuromancer" by William Gibson and explain how Gibson predicted what we are seeing today with the internet and AI, what he got wrong, and what may still evolve. Add a graphic timeline reflecting advances in these areas.

This is what Co-pilot gave me:

​

Neuromancer (1984) by William Gibson is a cyberpunk novel that follows Case, a hacker hired to infiltrate powerful systems in a dystopian future. He uncovers a plot involving two AIs — Wintermute and Neuromancer — that seek to merge and transcend their limitations. The novel explores themes of cyberspace, AI, corporate dominance, and human-machine integration.

---

​

• Cyberspace/Internet: Coined the term and envisioned a shared digital realm — now realized through the Web, VR, and Metaverse.
• AI: Predicted autonomous, goal-driven AIs — now reflected in AGI research and AI ethics.
• Cybersecurity: Anticipated the rise of hacking and digital warfare.
• Corporate Power: Foretold the dominance of Big Tech and surveillance capitalism.
• Human Augmentation: Envisioned cybernetic enhancements — now seen in wearables and brain-computer interfaces.

---

​

• Full VR immersion is still not mainstream.
• AI consciousness remains speculative.
• Social media and mobile tech were underplayed.
• Technological pace was overestimated.

---

​

• True AGI and AI self-awareness
• Neural interfaces for direct brain-digital interaction
• Digital consciousness and identity virtualization

---

​

 

[AllTheThings]

For me, the use of AI is inversely proportional to the risk of incorrect answers/hallucinations and need for accuracy. I cannot trust it NOT to hallucinate. It is also a struggle to get it to constrain answers to appropriate references or training materials: I don't care what a rando on linkedin thinks. I care what IEC or EU government guidance is.

Coding some python to do data or signal analysis, or collect measurements over GPIB, where I can validate the output easily? Yeah. Every couple weeks I do something like that with AI. One adventure was parsing and combining several hundred thousand CSVs generated from a factory floor test system...It would output one CSV per test. I had AI generate functional blocks of code, made some tweaks, set up 25 seed files (some with purposeful errors) to test, adjusted more, and let er rip.

If I was writing embedded C for a DSP, I doubt AI would have been nearly as useful.

Having AI do creepage and clearance determinations? No. Hazard block diagrams? No. FMEAs? Again no. Those need to be HITL (human in the loop) at minimum, and preferably entirely human generated.

Where I've had some success is using it as a 'fuzzy' search engine. Sometimes, I may have a product being used in a few different applications, where it has to comply with several standards: I can upload those standards, and prompt for specific outputs: "Analyze the uploaded standards, determine the minimum and maximum defined operating temperatures (or vibration levels or touch currents, etc.) . Enter the results in a table, where each line contains the standard, the clause/section, and the result. Include both likely and probable relevant clauses." It can make searching through a half dozen overlapping standards a lot easier.

 

[ChrisM]

I prefer to refer to AI as AS - Artificial Stupidity. Yes it can be beneficial in some cases but it still comes up with inaccuracies or blatant untruths far too often. Didn't Einstein have something to say about this?
"Two things are infinite: the universe and human stupidity; and I'm not sure about the universe"
How close to human stupidity will AI/AS get ??

 

[Johnny Quality]

I use AI every now and then for imagine generation (for me the most valuable and best success), market research and prompts for creativity.

As for generating work or delegating duties, for me it's still lacking.