First screen, captured 1 day ago. Hover to scroll the page, or open the video.

Why is AI generated language so hard to understand?

What HN thinks

A thread about unreadable AI prose that spends half its length arguing whether Google is a 'their' or an 'its'.

Training rewards look, not usefulness
RLHF raters and reward signals favour surface features like lists, headings and rule-following concision, so models hit the target but miss the point of communicating with an actual reader.
Verifiable rewards compress language
Reinforcement learning on math, code and reasoning with length penalties pushes models toward terse, out-of-distribution constructions that save tokens but hurt human comprehension.
No intent or mental model
AI text lacks an author modelling the reader or the physical world, so it can't separate important from mundane and reads as a steady stream of fake main points.

Consensus: Most agree AI prose is tiring because training optimizes for something other than clarity to a human audience, though a minority say they find it perfectly readable.

Recommended in the comments

pstack unslop skill ↗ (tool): Said to do wonders for technical writing and code comments.
pstack technical-writing skill ↗ (tool): Recommended alongside unslop for clearer system design and documentation text.
ASD-STE100 (other): Asking for it in prompts noticeably constrains and simplifies the model's language.

Summarised by Claude from the comment thread 1 day ago.

Points: 194
Comments: 167
Best rank: #5
Posted: 2026-10-24 by canjobear
Category: Ai ml

llm psycholinguistics ai-generated-text readability nlp arxiv

All captures:

Full page

Captured 1 day ago. Click for the whole screenshot.

Looks similar

Screenshot of arxiv.org: The alien space of science: Coherent but cognitively unavailable directions

The alien space of science: Coherent but cognitively unavailable directions (arxiv.org)

195 points by martincsweiss 4 days ago | 80 comments

arXiv abstract page for a NeurIPS 2026 paper proposing a method to sample scientifically coherent research directions that existing researcher communities are unlikely to propose.

research-ideation llm scientific-discovery novelty

Screenshot of arxiv.org: Attention Is Not What You Need (2025)

Attention Is Not What You Need (2025)(arxiv.org)

158 points by westurner 15 hours ago | 53 comments

arXiv paper proposing an attention-free sequence model using Grassmann flows and Plucker coordinates, compared against Transformers on Wikitext-2 and SNLI.

attention-free-architecture grassmann-manifold transformers sequence-modeling