Various LLM smells

2026-05-31 · source · source ↗ · raw capture · wiki page

By Shiv, published May 28, 2026 on shvbsle.in (Bear Blog).

Referenced HN thread: news.ycombinator.com/item?id=48313810

Core Argument

The author describes starting a math blog and using LLMs to polish writing, initially finding the output superior. After three months, they recognized recurring patterns ("ai-smells") appearing universally across LLM-assisted work. The piece catalogs tells in two domains.

Section 1: LLM Writing Tells

Way too many punchlines

Short, declarative, quasi-profound one-liners. Examples: - "Humans trust symmetry because it feels like intelligence made visible" - "The Tiger fit the story. Jin-yong fit the physics."

Consecutive short sentences

Rapid-fire fragments used for emphasis: - "Then AlphaEvolve arrived. It had no preference for symmetry" - "These examples are not decorative. They form a distributed argument."

"X is the Y of Z" construction

Example: "Cringe is the visible signature of moving along a gradient you chose."

"Not just X, it's Y" frame

The author flags constructions about solutions that "do not merely satisfy the constraint but satisfy the aesthetic instincts."

Section 2: AI-Generated Website Tells

Recurring visual elements: - JetBrains Mono font - "Step" and bullets in that exact font - Specific button styles - Repeated card component patterns - A blinking-dot badge component

Key Theme

"ai-smell seems like an artifact that emerges across various AI assisted tasks" — recognizable once you've seen enough examples.

Closing Note

"I'm not against LLM/AI usage for creative tasks. This is just me noticing things."

The math blog is noted as "now deleted."