Reducing algorithmic influence
Two necessary, unremovable, features exist that warrant this quick sitting header:
1. Language learning models, by the feature of their construction via matrix math & transformer architecture, confabulate outputs. The philosophically correct and extremely precise technical term for this is that LLMs are bullshit generators. They are in no way shape or form related to the anatomical structure of human hallucinations, and the term 'hallucination' for it is entirely 100% incorrect.
2. Human brains prioritize negatively valence information as a survival mechanism.
_____
Therefore,
- in an architecture that prioritizes capturing your attention
- as a necessary consequence of the features of LLMs & how brains work
Forms of online text media are bullshit that makes you stressed out/upset, and it is unprofitable/unrewardable/structurally-disincentivised for this to exist in any other way. Fortunately, some organizations prioritize algorithmic transparency.
It's probably helpful for you to conceptualize your LLM bullshit category error as asking a thesaurus for a boolean statement. Human neural circuitry is predictive in its capacity insofar as it optimizes energy usage within/via pyramidial neurons (there's tradeoffs for substance usage: increase in divergent creative problem solving ability can save time vs a lowering of convergent brute-force predictive capacity). This human architecture is completely, utterly, unrelated to any form of this computational architecture.
P.S. the negative effects of social media are due to the design of the platforms, so even a completely transparent algorithm does not change engaging with any form of social media being a net negative to your brain health.
Additional Bullshit Material
Book: On Bullshit by Harry G. Frankfurt
Lecture series: Calling Bullshit in the Age of Big Data by UW iSchool
Paper: ChatGPT is bullshit
Podcast: Partially Examined Life Ep. 327: Harry Frankfurt on Bullshit and Authenticity