1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

+++
title = "Fallacies"
summary = "Random fallacies I use to describe patterns I've seen"
date = 2026-08-19
category = "Other"
+++


### Sonic Fallacy

> Gotta go fast!

The Sonic Fallacy is when someone uses or advocates for a specific language or framework simply because "it's the fastest!" without profiling or otherwise attemmpting to discern whether it's even required.

Of course, no one _wants_ to have slow code. However, the Sonic Fallacy generally describes the choice being made _primarily_ because of speed, regardless of whether the language or framework is easy to use, memory safe, or any other number of trade-offs that can occur in the name of speed.[^1]


### Claude Fallacy

> Everyone writes slop AI code _except me_ because I Know How To Use It Properly:tm:

The Claude Fallacy could be described as "The Dunning-Kruger effect for LLM users".

It is the dichotomy among certain LLM users that LLMs produce slop when other people use them, however their own LLM does **not** produce slop because it is being used in a superior manner. \
While there is some discussion to be had about the efficiency or results of various prompting techniques, harnesses, agents, etc., overall in my (completely non-scientific) experience the average LLM project is the same.


-----

[^1]: I manage to support an entire state using Python/Django. I'm not saying that means Python is some perfect language, I'm just saying your project being in Rust doesn't automatically make it a unicorn.