I did rely heavily on GPT throughout the process, and it’s clear to me now that this is something I need to take more ownership of. Working on that will naturally help me improve how I source and cite materials as well.

I’m also taking your final piece of advice to heart—exams are coming up, and I know I need to shift my focus there for now. Thanks again for your honesty and encouragement. I really appreciate it.

I documented this publicly with reproducible behavior and cleanup proposals: → https://github.com/sks38317/gpt-cache-optimization/releases/...

Highlights from the release: - Token flushing failure during long outputs (e.g., PDF export) - Recursive reuse of failed cache content - Session decay from unpurged content - Trigger-based cleanup logic proposal

Before publishing, I submitted a formal message to OpenAI Support. Here's part of what I wrote:

> “I’ve shared feedback and proposals related to GPT behavior and system design, including: > - Memory simulation via user-side prompts > - Cache-loop issues and PDF rendering instability > - A framework modeling Systemic Risk (SSR) and Social Instability Probability (SIP) > - RFIM-inspired logic for agent-level coordination > > I only ask whether any of it was ever reviewed or considered internally.”

Their response was polite but opaque:

> “Thanks for your thoughtful contribution. We regularly review feedback, > but cannot provide confirmation, reference codes, or tracking status.”

Shortly after, I began observing GPT responses subtly reflecting concepts from the release—loop suppression, content cleanup triggers, and reduced carryover behavior.

It might be coincidence. But if independent contributors are echoing system patterns before they appear—and getting silence in return—maybe that’s worth discussing.

If you’ve had feedback disappear into the void and return uncredited, you’re not alone.

*sks38317* (independent contributor, archiving the things that quietly reappear)

Comments URL: https://news.ycombinator.com/item?id=43741181

Points: 2

# Comments: 0

Also, by having each spin affected by neighboring spins and external random fields, it naturally simulates interaction between variables. So it reflects both local dependencies and external noise, which is exactly the kind of behavior I want in my simulation.

I had been trying to account for interactions between variables, but I ran into a lot of frustration due to extreme or unstable outputs. Thanks to your guidance, though, I was finally able to break through that wall— and I’ve started thinking of ways to restructure and improve the model accordingly.

Let me know if I misunderstood anything.

Thanks a lot.

I'm working on a small system simulation project that explores how instability can emerge from the interaction of various variables over time. (Repo linked in the comments.)

Recently, I’ve been trying to incorporate the RFIM (Random Field Ising Model) into the logic—but I’ve hit a wall. It’s not the math or the concept itself, but rather the total lack of Korean-language resources or beginner-friendly guides that’s been challenging.

I’d really appreciate it if anyone:

has experience with RFIM and would be open to helping via a PR

knows of any well-documented codebases I could learn from

or could point me toward approachable explanations or tutorials

I’m still relatively new to GitHub and trying to improve this simulation step-by-step, so any support would go a long way.

Thanks for reading—and for any help you can offer!

Comments URL: https://news.ycombinator.com/item?id=43726538

Points: 2

# Comments: 6

I’ve recently started looking into system modeling and came across concepts like the Random Field Ising Model (RFIM) and temporal graph structures. I’m still new to this area, and while I’ve been going through English materials, I was wondering:

Are there any Korean-language resources, guides, or explanations on these topics? Even blog posts or translated papers would be helpful.

Comments URL: https://news.ycombinator.com/item?id=43725780

Points: 1

# Comments: 0

It’s based on a fictional timeline from 2023 to 2045, and the structure focuses more on how different variables interact over time, rather than predicting anything specific.

The project is still early and quite basic, and I’m mostly hoping to learn from others who’ve worked on system modeling or similar domains.

Here’s the GitHub repo: Why the System Breaks Before Anyone Notices

I’d really appreciate any feedback on: – whether this kind of framework makes any structural sense – where the logic feels weak or could be better grounded – or if you know of better approaches I should explore

Thanks for your time.

Comments URL: https://news.ycombinator.com/item?id=43722948

Points: 2

# Comments: 1

Repeated PDF failures seemed to create token overload and session slowdowns. So I tried manually analyzing the session, tracking token counts, and testing some user-side “optimizations”—like auto-removing failed outputs and cleaning redundant versions.

I used GPT itself to help write the report and interpret the data. It was a mix of curiosity, frustration, and… maybe procrastination. But it turned into a fun experiment.

I’ve only been exploring GitHub and ChatGPT for less than a month, so there are still many things I’m unfamiliar with.

So if there’s anything I’ve overlooked or could improve, I’d really appreciate your feedback.

Comments URL: https://news.ycombinator.com/item?id=43718292

Points: 15

# Comments: 6