🗣️ Post 1: Brandon Sanderson invents ChatGPT? The Story of a Prompt Engineer [video]

As posted by: nevster  |  🔥 Points: 2

🔗 https://www.youtube.com/watch?v=RtkmFN26VuU

💬 Summary

[No content available]

🗣️ Post 2: Show HN: Turn any study material into practice questions with one photo

As posted by: e_patjas  |  🔥 Points: 2

🔗 https://www.lexielearn.com/en

💬 Summary

Built this because kids 10-15 can't prompt AI effectively. They'd type "make quiz plz" and get confused when ChatGPT asks for context.

Lexie takes 46 steps of AI prompting, context explanation, and output formatting and makes it one step: point camera, get quiz. The photo constraint isn't just better UX, it's the only way to get usable output from that age group.

This demographic gets the worst deal in edtech. Too advanced for baby apps, too young for adult tools. They get gamification theater instead of real learning tools.

I stripped all that out. Photo your geology notes, get instant flashcards and quizzes that expose what you don't actually understand.

No ads, no trackers, photos stay on device. Making money from subscriptions, not from kids.

Available on iOS, Android coming soon. Works in English and in Finnish .

Personal motivation: got tired of watching my kids waste time highlighting and re-reading when I knew there was a better way to study.

🗣️ Post 3: Ask HN: Do you use the chat "Share" link as a save/load snapshot?

As posted by: TXTOS  |  🔥 Points: 1

🔗 https://news.ycombinator.com/item?id=44969763

💬 Summary

i noticed something simple that’s been surprisingly useful. most chat UIs have a Share button. if you treat that shared link as a snapshot of the current state, you can later paste the link and reload the exact tuned persona/config without re-priming. feels like “save slot” for AI chats.

how i use it

quick start: 1) tune until behavior is right → 2) press Share, copy link → 3) paste next time to boot the same state

works for: ChatGPT, Gemini, Claude, Perplexity, Grok

not true snapshots (in my tests): Mistral, Kimi (their “share” seems to export text, not state)

why it helps

reproducibility for A/B prompts and eval

faster incident response for RAG/OCR/agent pipelines (no re-tuning)

stable voice for long-form writing or T2I workflows

safe red-team vs blue-team comparisons in parallel tabs

open questions for HN

have you observed state fidelity differences across providers? which ones actually restore the same behavior?

any known privacy or retention pitfalls with shared links? do teams treat them as secrets?

tips to reduce drift after reload? warmup lines, pinned rules, or version tagging?

operational patterns you like? e.g., “master seed link → clone per task”, “creator/editor twin links”, audit tables, rotation cadence.

where does this fail? model/version swaps, truncation, context size, org policies, url lifespan?

sample use cases (compact)

RAG triage: one link per fault family (indexing, vector drift, routing)

prompt-injection lab: attacker link vs defender link, run side-by-side

writer flow: one “voice+outline” link, one “line-edit” link

SRE postmortem: template link with timeline + five-whys

i’m curious if others already do this, or if there are better patterns. what have you seen work or break?

🎯 Final Takeaways

These discussions reveal how developers think about emerging AI trends, tool usage, and practical innovation. Take inspiration from these community insights to level up your own development or prompt workflows.