updated numbers on tokenizer and improved profile

Author: beyondsimulations

images/tobias.png

@@ -11,14 +11,19 @@ format:
 
 # [About this Course]{.flow} {.title}
 
-## About me
-
-:::{.incremental}
-- **Field:** Optimizing and simulating complex systems
-- **Languages:** of choice: Julia, Python and Rust
-- **Interest:** Modeling, Simulations, Machine Learning
-- **Teaching:** OR, Algorithms, and Programming
-- **Contact:** [vlcek@beyondsimulations.com](mailto:vlcek@beyondsimulations.com)
+## About Me
+
+::: {.columns}
+::: {.column width="25%"}
+![](../images/tobias.png){width=3.125in style="display:block;margin:auto;" fig-alt="Portrait photo of the lecturer, Dr. Tobias Vlćek, smiling, with glasses and a dark blazer."}
+:::
+::: {.column width="75%"}
+- **Field:** Optimizing & simulating complex systems
+- **Languages of choice:** Julia, Python and Rust
+- **Interests:** Modeling, simulations, machine learning
+- **Teaching:** OR, algorithms, programming & mathematics
+- **More:** [tobiasvlcek.com](https://tobiasvlcek.com)
+:::
 :::
 
 . . .

lectures/lec_01_introduction.qmd

@@ -74,7 +74,7 @@ The course chatbot and every AI tool you will use run on one:
 
 - Text is broken into [tokens]{.highlight}, chunks of text mapped to numbers
 - Common words are one token; rarer words split into pieces: "un-", "break", "-able"
-- The Mistral models we use have a vocabulary of about **1,000,000** different tokens
+- The Mistral models we use have a vocabulary of about **131,000** different tokens
 - That is also why famous models fail at "count the r's in *strawberry*": they see a token like `straw` plus a token like `berry`, never the letters
 
 ## It guesses the next token

lectures/lec_06_modules.qmd

@@ -69,8 +69,9 @@ $lec-code-inline:  #4B7282 !default;   // 4.61:1
 
 // Quarto's slide grey ($text-muted, body colour lightened 30 %: #7C8998) is
 // 3.51:1 on the page ground. Footer, slide number, slide blockquotes and
-// asides take a darker grey.
-$lec-slide-muted:  #697480 !default;   // 4.70:1 / 4.60:1 on #FCFBF8
+// asides take a darker grey. It has to hold on the section slides too, which
+// sit on `codeline`: #697480 was 4.70:1 on the page and 4.21:1 there.
+$lec-slide-muted:  #646C76 !default;   // page 5.24:1 / 5.14:1, codeline 4.71:1 / 4.63:1
 
 // ── Callouts ──────────────────────────────────────────────────────────────
 $lec-callout-note:      #FCCF9C !default;   // = oneLight
@@ -613,6 +614,24 @@ body.quarto-light {
     color: $lec-slide-muted;
 }
 
+// Footer and slide number while a `.loud-slide` is showing. Both belong to the
+// deck, not to the slide, so the slide's colours never reached them: the grey
+// above is 1.08:1 on the accent ground and the link red 1.68:1. They take the
+// slide's own light text colour, 4.36:1, the most this ground allows. A slide
+// in a stack that is not showing keeps its `.present`, so the stack is checked
+// as well. The printed deck gets the same from create_pdf.py (`on-dark-bg`).
+.reveal:has(.slides > section.present.loud-slide,
+            .slides > section.present > section.present.loud-slide) {
+    .footer, .footer a:any-link, .slide-number {
+        color: $brand-lighter;
+    }
+    // The menu icon is an SVG with the link red painted in (1.68:1 here), so
+    // `color` does not reach it; the filter turns it white.
+    .slide-menu-button .fa-bars::before {
+        filter: brightness(0) invert(1);
+    }
+}
+
 // Document pages without a TOC (quizzes, exams with `toc: false`) get Quarto's
 // `fullcontent` grid, which drops the right margin column and centres the body,
 // so the text jumps right compared with every page that has a TOC. Give those