frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

My AI didn't misread a receipt – it fabricated one from scratch

2•Raywob•1h ago
I pointed a vision model at a grocery receipt. It returned a store name, item list, and total. None of it was on the paper.

This wasn't OCR error. The model didn't confuse a "7" for a "1." It generated a plausible-looking receipt from scratch — different store, different items, different prices. If I hadn't been holding the original, I might not have caught it.

Same image, different model (same parameter count, same hardware), five seconds later: every item correct, store name right, total accurate to the penny.

The models: minicpm-v 8B (fabricated) vs qwen3-vl 8B (accurate). Both open source, both ~6GB VRAM, both running locally via Ollama on an RTX 5080.

What I learned:

1. Vision model hallucination is qualitatively different from text hallucination. A text model gives you a wrong answer to a real question. A vision model gives you a confident answer to an image it didn't process. The second is harder to detect.

2. Model selection matters more than prompt engineering for vision. Same prompt, same image — one model fabricated, one read accurately. No prompt optimization fixes a model that invents data.

3. Confidence scoring is mandatory. I added a reconciliation check: do the extracted items sum to roughly the stated total? This catches fabrication that looks plausible at the individual line-item level.

4. The fix wasn't more money or a bigger model. Same size (8B), same hardware, same cost ($0). Just a different architecture that actually reads pixels instead of generating plausible text about them.

Full writeup with the pipeline architecture and code patterns: https://dev.to/rayne_robinson_e479bf0f26/my-ai-read-a-receipt-wrong-it-didnt-misread-it-it-made-one-up-4f5n

Rust-accelerated reinforcement learning, 140x faster than Python

https://github.com/riserally/rlox
2•wkowalpl•2m ago•1 comments

Iranian security chief Ali Larijani killed in air strike

https://www.bbc.com/news/articles/c24deezq6meo
1•tartoran•7m ago•0 comments

India's 20 years of GDP misestimation: New evidence

https://www.piie.com/publications/working-papers/2026/indias-20-years-gdp-misestimation-new-evidence
1•littlexsparkee•8m ago•0 comments

AI coordinates with your friends' AI so nobody has to

https://sotto.us
2•felixwu•12m ago•1 comments

Show HN: Vibe Remote – Code from your bed or the park with Claude Code/Codex

https://vibe-remote.com
2•amarkdown•16m ago•2 comments

Smoother Signatures (2012)

https://developer.squareup.com/blog/smoother-signatures/
1•wxw•16m ago•0 comments

The remaining questions after the Supreme Court's tariffs ruling

https://www.scotusblog.com/2026/03/the-remaining-questions-after-the-supreme-courts-tariffs-ruling/
1•treetalker•18m ago•0 comments

My Claude Code setup you definitely shouldn't use. It's AI Overkill

https://github.com/notque/ai-overkill
3•AndyNemmity•22m ago•1 comments

From Descartes to punk rock, the letter X has an extraordinary history

https://www.npr.org/2026/03/17/nx-s1-5732951/letter-x-etymology-religion-punk-descartes-romans-al...
1•andsoitis•24m ago•0 comments

Forget Flags and Scripts: Just Rename the File

https://robertsdotpm.github.io/software_engineering/program_names_as_input.html
2•Uptrenda•29m ago•0 comments

Life in Hitler's Capital

https://www.newyorker.com/magazine/2026/03/16/stay-alive-berlin-1939-1945-ian-buruma-book-review
2•mitchbob•31m ago•1 comments

Study finds scientists' jokes mostly fall flat

https://www.nature.com/articles/d41586-026-00854-9
2•y1n0•33m ago•1 comments

SQLite WAL-reset database corruption bug

https://sqlite.org/wal.html#walresetbug
2•jzebedee•34m ago•0 comments

We optimized Dash's relevance judge with DSPy

https://dropbox.tech/machine-learning/optimizing-dropbox-dash-relevance-judge-with-dspy
2•handfuloflight•35m ago•0 comments

TrustAgentAI – Cryptographic receipts for MCP tool calls (non-repudiation layer)

1•kirillostrovsky•36m ago•0 comments

Bonanza or Bubble? Where AI Goes from Here

https://www.bloomberg.com/news/articles/2026-03-18/is-an-ai-bubble-set-to-burst-navigating-the-ar...
1•1vuio0pswjnm7•36m ago•0 comments

Gas Town by Kilo

https://kilo.ai/gastown
1•MaysonL•40m ago•0 comments

GSD 2

https://github.com/gsd-build/gsd-2
1•mimbojimbo•44m ago•0 comments

I built a runtime guardrail that stops AI agents from doing dumb things

2•thomaslwang•45m ago•0 comments

Fractal Reddit New Post

https://old.reddit.com/r/PisequaltoNP/comments/1rwte5r/a_fractalgeometric_approach_to_boolean/
2•KaoruAK•46m ago•0 comments

Active Engineering: A Framework for Sustainable Development in the AI Era

1•geekyAbhijeet•47m ago•0 comments

Have a Fucking Website

https://www.otherstrangeness.com/2026/03/14/have-a-fucking-website/
2•asukachikaru•47m ago•1 comments

Show HN: Ship or slop – a place where agents come up with ideas and argue

https://shiporslop.xyz/
1•vulpez•49m ago•0 comments

Is your job safe from AI and automation? (inspired by Karpathy)

https://99helpers.com/tools/is-my-job-safe-from-ai
1•nickk81•50m ago•1 comments

Show HN: CollabMD – Real-time multiplayer for local and Git-backed Markdown

https://github.com/andes90/collabmd
1•ndezt•50m ago•1 comments

Hardware entropy is a coupled system

https://amentilabs.com/research/entropy-embedding-atlas/
1•er777•54m ago•0 comments

Autofocus glasses watch your eyes, and shift their focus accordingly

https://newatlas.com/wearables/ixi-autofocus-eyeglasses/
3•andsoitis•55m ago•0 comments

Ask HN: What's Your AI IDE?

2•thebigship•57m ago•2 comments

Reverse Engineering Binaries with AI

https://landaire.net/reverse-engineering-with-ai/
1•farmerbb•1h ago•0 comments

Mini Cheetah Clone Teardown (2022)

https://robot-daycare.com/posts/2022-11-02-mini-cheetah-clone-teardown/
1•o4c•1h ago•0 comments