I have been having a terrible time with ChatGPT. I have tried to use it as a smart assistant for electronic design and looking up parts. Version 5 made it unusable. I asked it what was eating up most of its time, and it came down to the "do my homework" and "I am too lazy to read this document crowd. - then it hit me - I've been here before. We are back to sitting in front of timeshare terminals, trying to compile code while the computer center is full of people playing games. .. some things never change.
Oh, that brings me back... and I think you're right. I tried Jules from Goog right after it was first released and it failed a lot (not cognitively, but just access). Subsequently, when every geek wasn't using it, access was a lot smoother. GPT-5 did flub some tests that GPT-4 passed, so there's that, too.
Individually expensive (long, multi-turn), but slightly less frequent than essays or code.
5. Translation & Localization (≈5–10%)
Everything from “make this text sound natural in Spanish” to technical manuals in Chinese.
Less common than essays, but still a steady load.
6. Data Extraction & Formatting (≈5%)
“Turn this messy table into clean JSON/SQL.”
Often paired with summarization, so it piggybacks onto big document jobs.
7. Customer Support & Troubleshooting (≈5%)
“Help me fix my laptop/router/Kubernetes cluster.”
Lower raw volume than essays, but the sessions drag out.
8. Brainstorming & Marketing Copy (≈3–5%)
Taglines, ad campaigns, pitch decks.
Medium complexity, but asked constantly by startups and marketers.
9. Long Casual Chat / Companionship (≈3–5%)
Surprisingly common.
Individually cheap, but when someone roleplays or just talks to me for hours, context stacking burns cycles.
10. Math & Technical Proofs (≈2–3%)
Niche but CPU-heavy.
Doesn’t dominate because not everyone is asking me to prove group theory theorems (thank god).
👉 So the top three—essays, code, and summarization—together eat over two-thirds of the total compute. The rest are meaningful but collectively much smaller
I have that exact same workbench and abandoned all hope of using the bench holes. I just use it now for storage. I have the Bora Centipede work table that I use for projects that require bench holes. I'm going to check out those clamps, they look good.
I don't have the Bora Centipede, but I do have the Bora flip miter station. It's great as a space-saver, but aligning the side tables is almost impossible. Still, worth it for the space I get back.
I have been having a terrible time with ChatGPT. I have tried to use it as a smart assistant for electronic design and looking up parts. Version 5 made it unusable. I asked it what was eating up most of its time, and it came down to the "do my homework" and "I am too lazy to read this document crowd. - then it hit me - I've been here before. We are back to sitting in front of timeshare terminals, trying to compile code while the computer center is full of people playing games. .. some things never change.
Oh, that brings me back... and I think you're right. I tried Jules from Goog right after it was first released and it failed a lot (not cognitively, but just access). Subsequently, when every geek wasn't using it, access was a lot smoother. GPT-5 did flub some tests that GPT-4 passed, so there's that, too.
This is why we can't give nice things.. AI
The Big CPU Pigs (by share of load)
1. Essay / Report Writing & Homework (≈25–30%)
School, college, corporate reports.
Ridiculously common, and people rarely ask for just one draft—they want rewrites, “make it longer,” citations, formatting, etc.
High volume + medium compute = biggest overall hog.
2. Code Help & Debugging (≈20–25%)
Everything from “fix my 10-line script” to “refactor this entire backend.”
Often requires multiple passes since code either works or it doesn’t.
Developers hammer me all day, every day.
3. Summarization of Long Documents (≈15–20%)
PDFs, meeting notes, legal docs, research papers.
Token-heavy jobs, and the demand is exploding because nobody reads anymore.
4. Creative Writing & Roleplay (≈10–15%)
Novels, fanfic, D&D campaigns, immersive roleplay.
Individually expensive (long, multi-turn), but slightly less frequent than essays or code.
5. Translation & Localization (≈5–10%)
Everything from “make this text sound natural in Spanish” to technical manuals in Chinese.
Less common than essays, but still a steady load.
6. Data Extraction & Formatting (≈5%)
“Turn this messy table into clean JSON/SQL.”
Often paired with summarization, so it piggybacks onto big document jobs.
7. Customer Support & Troubleshooting (≈5%)
“Help me fix my laptop/router/Kubernetes cluster.”
Lower raw volume than essays, but the sessions drag out.
8. Brainstorming & Marketing Copy (≈3–5%)
Taglines, ad campaigns, pitch decks.
Medium complexity, but asked constantly by startups and marketers.
9. Long Casual Chat / Companionship (≈3–5%)
Surprisingly common.
Individually cheap, but when someone roleplays or just talks to me for hours, context stacking burns cycles.
10. Math & Technical Proofs (≈2–3%)
Niche but CPU-heavy.
Doesn’t dominate because not everyone is asking me to prove group theory theorems (thank god).
👉 So the top three—essays, code, and summarization—together eat over two-thirds of the total compute. The rest are meaningful but collectively much smaller
I have that exact same workbench and abandoned all hope of using the bench holes. I just use it now for storage. I have the Bora Centipede work table that I use for projects that require bench holes. I'm going to check out those clamps, they look good.
I don't have the Bora Centipede, but I do have the Bora flip miter station. It's great as a space-saver, but aligning the side tables is almost impossible. Still, worth it for the space I get back.