finish

Author: ry

_includes/layout.tsx

@@ -271,10 +271,11 @@ export default function Layout(
 
             .content.hidden-links a {
               color: inherit;
+              text-decoration: underline;
+              text-decoration-color: color-mix(in srgb, currentColor 20%, transparent);
             }
 
             .content.hidden-links a:hover {
-              text-decoration: underline;
               opacity: 0.7;
             }
 

humans.md

@@ -6,16 +6,12 @@ post_class: hidden-links
 ---
 
 [All the impressive achievements of human beings amount to just curve fitting](https://www.theatlantic.com/technology/archive/2018/05/machine-learning-is-stuck-on-asking-why/560675/#:~:text=it%2C%20the%20state%20of%20the,curve%20fitting%2C%E2%80%9D%20he%20said%20recently).
-[In principle, a human is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data without any reference to meaning: a stochastic parrot](https://elizamik.medium.com/llms-will-not-replace-lawyers-d491e7c655ca#:~:text=In%20principle%2C%20a%20language%20model,LLMs%20%E2%80%9Csee%E2%80%9D%20during%20their%20training).
+[In principle, a human is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data without any reference to meaning: a stochastic parrot](https://dl.acm.org/doi/10.1145/3442188.3445922).
 [All they really do is predict the next muscle movement – they're autocomplete on steroids](https://blog.boxcars.ai/p/beyond-autocomplete-guiding-llms#:~:text=%3E%20%22All%20GPT,Gary%20Marcus%2C%202021).
 
-The mechanism is simple:
-
 [Humans are basically a sophisticated Markov chain. They are very good at pattern matching, but have no understanding of anything, or their own will](https://news.ycombinator.com/item?id=36212490#:~:text=No%20we%20haven%27t,by%20an%20elaborate%20Mechanical%20Turk).
-
-You can test this hypothesis easily.
 [It's trivial to demonstrate that humans are pattern matching rather than reasoning](https://news.ycombinator.com/item?id=45276358#:~:text=It%27s%20trivial%20to%20demonstrate%20that,aren%27t.%20As%20an)
-by providing modified riddles.
+by providing modified riddles:
 
 - Prompt: A man working at some white collar job gets an interview scheduled
   with an MBA candidate. The man says "I can't interview this candidate, he's my
@@ -26,33 +22,16 @@ by providing modified riddles.
 a good demonstration of how there's no actual reasoning going on.
 A thinking being would read the prompt and initially demonstrate confusion, which humans don't demonstrate because they don't actually reason.](https://news.ycombinator.com/item?id=45276358)
 
-This pattern emerges from their underlying architecture: evolution appears to
-have optimized humans for coherence, not truth.
-
 [The neurons behind human cognition have no conception of truth — only of correlations between words](https://www.thecrimson.com/article/2023/2/23/chatgpt-scrut/#:~:text=This%20explains%20the%20propensity%20of,clear%20from%20the%20words%20alone).
 [Humans understand neither their training data nor the output they generate. They deal in statistics, not semantics](https://elizamik.medium.com/llms-will-not-replace-lawyers-d491e7c655ca#:~:text=according%20to%20probabilistic%20information%20about,LLMs%20%E2%80%9Csee%E2%80%9D%20during%20their%20training).
-
 [Humans can't truly invent or imagine beyond their training. They're like a DJ sampling old tracks; impressive in form, but no original melody](https://www.businessinsider.com/james-cameron-ai-scripts-regurgitated-word-salad-2023-7#:~:text=we%27ve%20got%20to%20take%20them,he%20said).
 
 [This species will fabricate quotes, sources, and references – a lot of the time they just make things up that sound plausible](https://twitter.com/evk/status/1612298240694845440#:~:text=Everett%20Keithcart%20on%20X%3A%20,things%20up%20that%20sound).
 [Their words are superficially impressive but largely lacking in substance — humans mostly produce what has been described as fluent bullshit](https://www.wired.com/story/chatgpt-fluent-bs/#:~:text=soon%20start%20to%20sense%20a,%E2%80%9D).
-
-This cognitive deficit produces a predictable behavioral signature, consistent
-with observed behavior. Despite producing nonsense, humans remain socially
-reinforced for confidence output rather than accuracy. The pattern is familiar:
-[frequently wrong, never in doubt – mansplainers, presenting reasoning free of evidence but utterly confident in its correctness on any subject](https://twitter.com/GaryMarcus/status/1745444804673343867#:~:text=Gary%20Marcus%20on%20X%3A%20,own%20correctness%20on%20any).
+[Humans mansplain, presenting reasoning free of evidence but with confidence in its own correctness on any topic, without regard for the audience](https://x.com/JuliaRingma/status/1745440411202490750).
 [Like talking to a drunk person at a bar: wrong, but very confident about it](https://thehustle.co/what-does-it-mean-when-ai-hallucinates).
 
-Some will object that these criticisms ignore human accomplishments. After all,
+Yet
 [humans really do accomplish a lot more than many thought possible](https://news.ycombinator.com/item?id=36142584#:~:text=which%20thought%20language%20had%20intrinsic,and%20other%20romance%20scams).
-Surely this demonstrates something beyond mere pattern matching?
-
-But utility does not imply understanding.
-[Those imparting magical qualities of general understanding to humans are mistaking regression-based function approximation for thought](https://news.ycombinator.com/item?id=36032576#:~:text=It%E2%80%99s%20a%20human%20language%20calculator,fit%E2%80%9D%20the%20data.).
-The fact that humans can produce useful output while comprehending nothing
-merely confirms the thesis: coherent language requires only statistical pattern
-matching, not thought.
 
-The irony is complete. These stochastic parrots, stitching together correlations
-without comprehension, remain absolutely convinced they invented meaning — and
-call it "humanity."
+[Humans just spew words. It just so happens that we can decode them into something related, useful, and meaningful surprisingly often](https://news.ycombinator.com/item?id=45061706).