frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

AI Keeps Getting Better. Why Is Its Writing Still So Bad?

https://greyenlightenment.com/2026/10/21/ai-keeps-getting-better-why-is-its-writing-still-so-bad/
2•paulpauper•46m ago

Comments

retrac•27m ago
Commenting here on HN resulted in something I didn't expect - I've been told I'm a good writer. I had suspected I was above average in my ability to conduct my thoughts through the constraint of the written word into the mind of another person, but it's interesting to have some feedback even if from pseudonymous Internet strangers.

I write oddly. I coin words. I wrote the word "penalizement" the other day. It's not a real word. It's a faulty formation, a confusion of punish and penalize with the -ment suffix. But still, it's a logical formation, and I think it made sense in a sentence: "subject to penalizements". So I used it anyway. Why the hell not? I also have a linguistics background, and so my lexicon is full of terms like "disprefer" which have obvious meaning and are valid words but they are red squiggles in Firefox's spell checker; most people don't use words like that.

My writing is improbable. Statistically it's improbable. Good writing is slightly improbable. It follows that it would be hard to emulate my writing statistically.

> Even in a cathedral, each piece of the architecture serves a role, just as each brick does in a modest building. By comparison, AI writing adds words that offer no value.

I actually think that excessive use of filler words without any specific function is one of the biggest flaws in my style of writing. When I edit myself it's the first thing I attack. Oh, I've done it again: "I consider the greatest flaw in my own style to be superfluous filler words."

Nvidia launches Minerva to catch failing parts in giant AI data centres

https://thenextweb.com/news/nvidia-minerva-aiops-ai-factory-monitoring
2•PicardManeuver•2m ago•0 comments

Electron microscope- What it takes to run

https://www.youtube.com/watch?v=wD1YGVwL418
1•oliculipolicula•2m ago•0 comments

This I Believe

https://www.awakin.org/v2/read/view.php?tid=468
1•ctoth•4m ago•0 comments

Building Laminar: Ramp's Managed Agents Platform

https://builders.ramp.com/post/building-laminar
1•gmays•4m ago•0 comments

Show HN: bonds – not an agent, a friend with superpowers for Gen Z group chats

https://bonds.chat/
1•juanchoabdon•20m ago•1 comments

Reverse Engineered Onshape for MCP Server

https://github.com/godbrigero/onshape-native
2•godbrigero•25m ago•1 comments

Nvidia Ampere GPUs got unlocked/exploited

https://github.com/Zagitta/gpu-unlock/blob/main/docs/How%20Ampere%20was%20unlocked.md
1•Zagitta•26m ago•0 comments

Meta's Hunt for AI Capital Strikes a Raw Nerve on Wall Street

https://www.wsj.com/finance/stocks/metas-hunt-for-ai-capital-strikes-a-raw-nerve-on-wall-street-e...
3•cdrnsf•33m ago•0 comments

We Spoke with Entrepreneurs Living the "7 Vibe 3" Lifestyle

https://cadence.moe/blog/2026-09-10-we-spoke-with-entrepreneurs-living-the-7-vibe-3-lifestyle
1•Cknight70•40m ago•2 comments

Humming: Fast, Flexible Kernels for Quantized Inference

https://vllm.ai/blog/2026-10-21-humming
1•gmays•41m ago•0 comments

Red-Teaming OpenAI's Internal Agent Monitoring Systems

https://blog.redwoodresearch.org/p/red-teaming-openais-internal-agent
1•btown•42m ago•0 comments

Offline FairPlay DRM Decryption

https://github.com/EthanArbuckle/fairplay-decrypt
1•sjtgraham•44m ago•0 comments

AI Keeps Getting Better. Why Is Its Writing Still So Bad?

https://greyenlightenment.com/2026/10/21/ai-keeps-getting-better-why-is-its-writing-still-so-bad/
2•paulpauper•46m ago•1 comments

The State of PyTorch 2026

https://techstrong.ai/articles/the-state-of-pytorch-2026/
1•CrankyBear•48m ago•0 comments

XRP Ledger Fixes Lending Code Before the Feature Goes Live

https://coinmarketcap.com/community/post/380272949/
1•joeymabia1•48m ago•1 comments

Intel tour by Dwarkesh: What will computer chips look like in 2045?

https://www.dwarkesh.com/p/intel-tour
1•osnium123•50m ago•0 comments

Untold: Iran

https://www.untoldart.com/venice
1•linuxkernal•52m ago•0 comments

Musk Giving Tops $445M in Midterm Election Push for GOP

https://www.bloomberg.com/news/articles/2026-10-23/musk-giving-tops-445-million-in-midterm-electi...
9•pulisse•52m ago•0 comments

Quest Diagnostics (Hacked)

https://www.questdiagnostics.com/links/quanum-notice
2•tortilla•53m ago•1 comments

PointerCam, a screen recorder that zooms on clicks and writes guides

https://pointercam.com/
1•rhrahul•54m ago•0 comments

A tale of two PHP remote vulnerabilities

https://calif.io/php2shell.x/
1•wyclif•54m ago•1 comments

Go++

https://github.com/jordan-bonecutter/go-plus-plus
1•jordan_bonecut•57m ago•1 comments

Physicists trap a single antimatter particle for a month For the first time

https://www.yahoo.com/news/science/articles/physicists-trap-single-antimatter-particle-150000058....
1•thunderbong•57m ago•0 comments

Demis Hassabis shortens AGI timeline from five to two to three years

https://the-decoder.com/deepmind-founder-hassabis-shortens-agi-timeline-from-five-to-two-to-three...
1•brandonb•1h ago•2 comments

Peter Thiel breaks with Trump over Iran war: 'We should not have done it'

https://www.politico.com/news/2026/10/21/peter-thiel-trump-iran-war-01131284
4•wslh•1h ago•0 comments

Assure – formal verification for every PR

https://symbolia.ai/
2•Veloc1te•1h ago•0 comments

Show HN: Supastack – Cross-Platform Turborepo for Next.js, Expo, and Tauri

https://supastack.dev
1•Learnmore_smart•1h ago•0 comments

CVE Tracker

https://cve.0xosprey.com/
1•mtweak•1h ago•0 comments

Tech's Enshittification, Public Markets, and Business Schools

https://tomorrowasusual.substack.com/p/techs-enshittification-public-markets
3•tudell•1h ago•1 comments

Show HN: GrillMyProfile – AI roast of your LinkedIn, scored against a job

https://www.grillmyprofile.com
1•luckeybearcreat•1h ago•0 comments