Here is the important part: We propose and test the LLM Brain Rot Hypothesis: continual exposure to junk web text induces lasting cognitive decline in large language model...Contrary to the control group, continual pre-training of 4 LLMs on the junk dataset causes non-trivial declines (Hedges' g>0.3) on reasoning, long-context understanding, safety, and inflating "dark traits" (e.g., psychopathy, narcissism)....Thought-skipping as the primary lesion: models increasingly truncate or skip reasoning chains, explaining most of the error growth...Popularity as a better indicator: the popularity, a non-semantic metric, of a tweet is a better indicator of the Brain Rot effect than the length in M1. Together, the results provide significant, multi-perspective evidence that data quality is a causal driver of LLM capability decay, reframing curation for continual pretraining as a training-time safety problem and motivating routine "cognitive health checks" for deployed LLMs
I think she is exhibiting this brain rot symptom outlined in the paper: Thought-skipping as the primary lesion: models increasingly truncate or skip reasoning chains, explaining most of the error growth. I think what is going on is that she is routing almost all of her cognition to system 1 (intuitive) instead of system 2. System 1 intuitive pattern matching can appear intelligent, but I think its big downside is that it lacks self awareness and the ability to self modify.
p
I think the issue is accurately defined in this recent research paper: LLMs Can Get "Brain Rot"!
Here is the important part: We propose and test the LLM Brain Rot Hypothesis: continual exposure to junk web text induces lasting cognitive decline in large language model...Contrary to the control group, continual pre-training of 4 LLMs on the junk dataset causes non-trivial declines (Hedges' g>0.3) on reasoning, long-context understanding, safety, and inflating "dark traits" (e.g., psychopathy, narcissism)....Thought-skipping as the primary lesion: models increasingly truncate or skip reasoning chains, explaining most of the error growth...Popularity as a better indicator: the popularity, a non-semantic metric, of a tweet is a better indicator of the Brain Rot effect than the length in M1. Together, the results provide significant, multi-perspective evidence that data quality is a causal driver of LLM capability decay, reframing curation for continual pretraining as a training-time safety problem and motivating routine "cognitive health checks" for deployed LLMs
I think she is exhibiting this brain rot symptom outlined in the paper: Thought-skipping as the primary lesion: models increasingly truncate or skip reasoning chains, explaining most of the error growth. I think what is going on is that she is routing almost all of her cognition to system 1 (intuitive) instead of system 2. System 1 intuitive pattern matching can appear intelligent, but I think its big downside is that it lacks self awareness and the ability to self modify.
More options
Context Copy link