humans are stochastic

.github/workflows/ci.yml

@@ -13,7 +13,7 @@ jobs:
         uses: actions/checkout@v2
 
       - name: Install Deno
-        uses: denoland/setup-deno@v1
+        uses: denoland/setup-deno@v2
 
       - name: Check formatting
         run: deno fmt --check
@@ -22,4 +22,4 @@ jobs:
         run: deno lint
 
       - name: Typecheck
-        run: deno check main.js
+        run: deno check

_config.ts

@@ -7,8 +7,7 @@ import mdx from "lume/plugins/mdx.ts";
 const site = lume({
   src: ".",
   dest: "_site",
-}, {
-  url: "https://tinyclouds.ry.deno.net",
+  location: new URL("https://tinyclouds.ry.deno.net"),
 });
 
 site.use(jsx());

_includes/layout.tsx

@@ -2,7 +2,7 @@ export default function Layout(
   { title, background, children }: {
     title?: string;
     background?: string;
-    children: React.ReactNode;
+    children: unknown;
   },
 ) {
   return (
@@ -269,6 +269,16 @@ export default function Layout(
               background-color: #1a1a1a;
             }
 
+            .content.hidden-links a {
+              color: inherit;
+              text-decoration: underline;
+              text-decoration-color: color-mix(in srgb, currentColor 20%, transparent);
+            }
+
+            .content.hidden-links a:hover {
+              opacity: 0.7;
+            }
+
             @media (max-width: 640px) {
               .container {
                 padding: 2rem 1.5rem;

_includes/post.tsx

@@ -1,11 +1,12 @@
 export const layout = "layout.tsx";
 
 export default function Post(
-  { title, publish_date, cover_html, children }: {
+  { title, publish_date, cover_html, post_class, children }: {
     title: string;
     publish_date: Date;
     cover_html?: string;
-    children: React.ReactNode;
+    post_class?: string;
+    children: unknown;
   },
 ) {
   const formattedDate = new Date(publish_date).toISOString().split("T")[0];
@@ -14,7 +15,9 @@ export default function Post(
     <>
       <div class="header">
         <a href="/" class="back-button">← Back</a>
-        <button class="theme-toggle" onclick="toggleTheme()">🌓</button>
+        <button type="button" class="theme-toggle" onclick="toggleTheme()">
+          🌓
+        </button>
       </div>
 
       {cover_html && (
@@ -29,7 +32,7 @@ export default function Post(
         {formattedDate}
       </div>
 
-      <div class="content">
+      <div class={`content${post_class ? ` ${post_class}` : ""}`}>
         {children}
       </div>
     </>

deno.json

@@ -25,7 +25,10 @@
   "lint": {
     "plugins": [
       "https://deno.land/x/lume@v3.0.5/lint.ts"
-    ]
+    ],
+    "rules": {
+      "exclude": ["no-explicit-any"]
+    }
   },
   "deploy": {
     "org": "ry",

humans.md

@@ -0,0 +1,37 @@
+---
+title: Humans Are Just Stochastic Parrots
+publish_date: 2025-10-10
+layout: post.tsx
+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://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).
+
+[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).
+[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:
+
+- 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
+  son." How is this possible?
+- Human: Because the interviewer is the candidate's mother.
+
+[This is clearly pattern matching and overfitting to the famous "doctor riddle" –
+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)
+
+[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).
+[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).
+
+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).
+
+[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).