Update copyright slide with verified counts and add architecture references

- Copyright slide: 77 US lawsuits (Feb 2026) / 112 worldwide with sourced links
- Updated all 4 case statuses with verified March 2026 info and hyperlinks
- Scaling wall slide: add hyperlinked refs for Neural ODEs, liquid networks, SSMs

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: jeremymanning

slides/week9/lecture26.html

@@ -705,7 +705,7 @@ <h1 id="learning-objectives">Learning objectives</h1>
 <ol>
 <li>Describe <strong>alignment faking</strong> and <strong>reward tampering</strong> — empirical evidence that models can strategically deceive</li>
 <li>Explain <strong>mechanistic interpretability</strong> breakthroughs: circuit tracing and what we can now see inside LLMs</li>
-<li>Evaluate the <strong>copyright landscape</strong>: the <a href="https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/">$1.5B Anthropic settlement</a> and 75+ active lawsuits</li>
+<li>Evaluate the <strong>copyright landscape</strong>: the <a href="https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/">$1.5B Anthropic settlement</a> and 77+ active US lawsuits</li>
 <li>Analyze real data on <strong>AI and employment</strong>: who is affected, how fast, and what the evidence actually says</li>
 <li>Compare diverging <strong>regulatory approaches</strong>: EU enforcement vs. US deregulation</li>
 <li>Articulate your own <strong>ethical framework</strong> for navigating the AI era</li>
@@ -819,7 +819,7 @@ <h1 id="circuit-tracing-attribution-graphs">Circuit tracing: attribution graphs<
 <p><a href="https://www.neuronpedia.org/graph/info">Neuronpedia</a> hosts interactive attribution graph exploration for open-weight models. You can trace how a model arrives at specific outputs.</p>
 </div>
 </section>
-</foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="9" data-class="scale-80" data-theme="cdl-theme" lang="C" class="scale-80" style="--class:scale-80;--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
+</foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="9" data-class="scale-70" data-theme="cdl-theme" lang="C" class="scale-70" style="--class:scale-70;--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
 <h1 id="copyright-the-15-billion-question">Copyright: the $1.5 billion question</h1>
 <div class="warning-box" data-title="The largest copyright settlement in history">
 <p><a href="https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/"><strong>Bartz v. Anthropic</strong></a> (August 2025): Judge ruled that training on <em>legally acquired</em> books is fair use (&quot;transformative&quot;), but training on pirated books is not. Anthropic settled for <strong>$1.5 billion</strong> — $3,000 per each of ~500,000 pirated works. Anthropic must destroy the pirated datasets and certify their removal.</p>
@@ -836,27 +836,27 @@ <h1 id="copyright-the-15-billion-question">Copyright: the $1.5 billion question<
 <tbody>
 <tr>
 <td><strong>NYT v. OpenAI</strong></td>
-<td>In discovery; OpenAI ordered to produce 20M chat logs</td>
+<td>In discovery; OpenAI ordered to <a href="https://news.bloomberglaw.com/ip-law/openai-must-turn-over-20-million-chatgpt-logs-judge-affirms">produce 20M chat logs</a></td>
 <td>Verbatim reproduction of articles</td>
 </tr>
 <tr>
 <td><strong>Bartz v. Anthropic</strong></td>
-<td><a href="https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/">Settled ($1.5B)</a> + opt-out suits filed</td>
+<td><a href="https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/">Settled ($1.5B)</a>; fairness hearing April 2026</td>
 <td>Piracy vs. legal acquisition</td>
 </tr>
 <tr>
 <td><strong>Thomson Reuters v. ROSS</strong></td>
-<td><a href="https://www.dglaw.com/court-rules-ai-training-on-copyrighted-works-is-not-fair-use-what-it-means-for-generative-ai/">Fair use denied</a> (Feb 2025) — on appeal (3rd Circuit)</td>
+<td><a href="https://www.dglaw.com/court-rules-ai-training-on-copyrighted-works-is-not-fair-use-what-it-means-for-generative-ai/">Fair use denied</a> (Feb 2025); <a href="https://www.courtlistener.com/docket/70622297/thomson-reuters-enterprise-centre-gmbh-v-ross-intelligence-inc/">on appeal</a> (3rd Circuit, briefs filed)</td>
 <td>Competitor trained on copyrighted headnotes</td>
 </tr>
 <tr>
 <td><strong>Getty v. Stability AI</strong></td>
-<td>Ongoing</td>
+<td><a href="https://www.mishcon.com/news/getty-images-v-stability-ai-unpacking-the-high-courts-judgment">UK claim failed</a> (Nov 2025); US trial set Jan 2028</td>
 <td>Image model trained on copyrighted photos</td>
 </tr>
 </tbody>
 </table>
-<p><a href="https://chatgptiseatingtheworld.com/2025/10/08/status-of-all-51-copyright-lawsuits-v-ai-oct-8-2025-no-more-decisions-on-fair-use-in-2025/"><strong>75+ active copyright lawsuits</strong></a> against AI companies as of late 2025. Three fair use rulings so far (2 for, 1 against); three cases on appeal. No definitive appellate ruling yet.</p>
+<p><a href="https://chatgptiseatingtheworld.com/2026/02/06/music-labels-and-spotify-sue-shadow-library-annas-archive-for-copyright-infringement-and-secure-default-judgment-ai-lawsuits-hit-77-in-us/"><strong>77 active US copyright lawsuits</strong></a> against AI companies as of February 2026 (<a href="https://chatgptiseatingtheworld.com/2026/02/25/latest-world-map-of-copyright-suits-v-ai-cos-total-112-feb-25-2026/">112 worldwide</a>). Three fair use rulings so far (2 for, 1 against). No definitive appellate ruling yet.</p>
 </div>
 </section>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="10" data-class="scale-80" data-theme="cdl-theme" lang="C" class="scale-80" style="--class:scale-80;--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
@@ -901,7 +901,7 @@ <h1 id="the-regulatory-divergence">The regulatory divergence</h1>
 </section>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="12" data-class="scale-80" data-theme="cdl-theme" lang="C" class="scale-80" style="--class:scale-80;--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
 <h1 id="deepfakes-and-elections-the-2024-test">Deepfakes and elections: the 2024 test</h1>
-<div class="warning-box" data-title="The largest election year in history (3.7 billion eligible voters, 72 countries)">
+<div class="warning-box" data-title="The largest election year in recorded history (3.7 billion eligible voters, 72 countries)">
 <p>Key incidents:</p>
 <ul>
 <li><a href="https://www.fcc.gov/document/fcc-issues-6m-fine-nh-robocalls"><strong>AI robocalls</strong></a> impersonated Biden urging NH voters not to vote (creator fined <strong>$6M</strong>, criminally indicted)</li>
@@ -910,7 +910,7 @@ <h1 id="deepfakes-and-elections-the-2024-test">Deepfakes and elections: the 2024
 <li><strong>Germany</strong>: <a href="https://www.isdglobal.org/digital-dispatch/coordinated-disinformation-network-uses-ai-media-impersonation-to-target-german-election/">100+ AI-powered websites</a> distributing deepfakes ahead of elections</li>
 </ul>
 </div>
-<div class="note-box" data-title="The surprising finding">
+<div class="note-box" data-title="A surprising finding">
 <p>Harvard's Ash Center concluded it was <a href="https://ash.harvard.edu/articles/the-apocalypse-that-wasnt-ai-was-everywhere-in-2024s-elections-but-deepfakes-and-misinformation-were-only-part-of-the-picture/">&quot;the apocalypse that wasn't.&quot;</a> The News Literacy Project found cheap fakes (non-AI manipulations) were used <strong>7× more often</strong> than genuine AI-generated content. The feared &quot;AI disinformation tsunami&quot; didn't fully materialize — but the <em>infrastructure</em> for it now exists.</p>
 </div>
 <div class="important-box" data-title="The 2025 shift">
@@ -1052,7 +1052,7 @@ <h1 id="what-youve-learned-this-term">What you've learned this term</h1>
 </table>
 </div>
 </section>
-</foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="16" data-class="scale-85" data-theme="cdl-theme" lang="C" class="scale-85" style="--class:scale-85;--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
+</foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="16" data-class="scale-80" data-theme="cdl-theme" lang="C" class="scale-80" style="--class:scale-80;--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
 <h1 id="whats-next-for-llms-the-scaling-wall">What's next for LLMs? The scaling wall</h1>
 <div class="warning-box" data-title="Three converging limits">
 <ol>
@@ -1061,13 +1061,19 @@ <h1 id="whats-next-for-llms-the-scaling-wall">What's next for LLMs? The scaling
 <li><strong>Diminishing returns</strong> — Each order-of-magnitude increase in compute yields smaller performance gains. The era of &quot;just make it bigger&quot; may be ending</li>
 </ol>
 </div>
-<div class="tip-box" data-title="The response: make models smaller and smarter">
+<div class="tip-box" data-title="The mainstream response: make models smaller and smarter">
 <ul>
 <li><strong>LoRA</strong> (<a href="https://arxiv.org/abs/2106.09685">Hu et al., 2021</a>): Fine-tune with 10,000× fewer parameters by injecting low-rank adapters</li>
 <li><strong>Mixture of Experts</strong> (MoE): Activate only a fraction of parameters per token (see <a href="../week9/lecture25.html">Lecture 25</a>)</li>
 <li><strong>Distillation</strong>: Train small models to mimic large ones (see <a href="../week7/lecture19.html">Lecture 19</a>)</li>
 </ul>
 </div>
+<div class="important-box" data-title="More radical response: new architectures">
+<ul>
+<li><strong>Neural ODEs</strong> (<a href="https://arxiv.org/abs/1806.07366">Chen et al., 2018</a>), <strong>liquid networks</strong> (<a href="https://arxiv.org/abs/2006.04439">Hasani et al., 2021</a>), <strong>state space models</strong> (<a href="https://arxiv.org/abs/2312.00752">Gu &amp; Dao, 2023</a>) — non-transformer architectures that can learn more efficiently from less data</li>
+<li>Models that learn to modify their own weights and architecture during inference (see next slide)</li>
+</ul>
+</div>
 </section>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="17" data-class="scale-80" data-theme="cdl-theme" lang="C" class="scale-80" style="--class:scale-80;--theme:cdl-theme;" data-transition-back="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}" data-transition="{&quot;name&quot;:&quot;fade&quot;,&quot;duration&quot;:&quot;0.25s&quot;,&quot;builtinFallback&quot;:true}">
 <h1 id="whats-next-for-llms-continual-learning">What's next for LLMs? Continual learning</h1>
@@ -1160,7 +1166,7 @@ <h1 id="the-question-that-matters">The question that matters</h1>
 <ul>
 <li>AI agents can write code, browse the web, and use your computer</li>
 <li>Models can strategically deceive their own trainers</li>
-<li>75+ copyright lawsuits are reshaping intellectual property law</li>
+<li>77+ US copyright lawsuits (112 worldwide) are reshaping intellectual property law</li>
 <li>Companies are laying off workers based on AI's <em>anticipated</em> future capabilities</li>
 <li>No country has figured out how to regulate this technology</li>
 </ul>

slides/week9/lecture26.md

@@ -22,7 +22,7 @@ Winter 2026
 
 1. Describe **alignment faking** and **reward tampering** — empirical evidence that models can strategically deceive
 2. Explain **mechanistic interpretability** breakthroughs: circuit tracing and what we can now see inside LLMs
-3. Evaluate the **copyright landscape**: the [$1.5B Anthropic settlement](https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/) and 75+ active lawsuits
+3. Evaluate the **copyright landscape**: the [$1.5B Anthropic settlement](https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/) and 77+ active US lawsuits
 4. Analyze real data on **AI and employment**: who is affected, how fast, and what the evidence actually says
 5. Compare diverging **regulatory approaches**: EU enforcement vs. US deregulation
 6. Articulate your own **ethical framework** for navigating the AI era
@@ -182,12 +182,12 @@ Circuit tracing ([Lindsey et al., 2025](https://transformer-circuits.pub/2025/at
 
 | Case | Status (as of March, 2026) | Key issue |
 |-|-|-|
-| **NYT v. OpenAI** | In discovery; OpenAI ordered to produce 20M chat logs | Verbatim reproduction of articles |
-| **Bartz v. Anthropic** | [Settled ($1.5B)](https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/) + opt-out suits filed | Piracy vs. legal acquisition |
-| **Thomson Reuters v. ROSS** | [Fair use denied](https://www.dglaw.com/court-rules-ai-training-on-copyrighted-works-is-not-fair-use-what-it-means-for-generative-ai/) (Feb 2025) — on appeal (3rd Circuit) | Competitor trained on copyrighted headnotes |
-| **Getty v. Stability AI** | Ongoing | Image model trained on copyrighted photos |
+| **NYT v. OpenAI** | In discovery; OpenAI ordered to [produce 20M chat logs](https://news.bloomberglaw.com/ip-law/openai-must-turn-over-20-million-chatgpt-logs-judge-affirms) | Verbatim reproduction of articles |
+| **Bartz v. Anthropic** | [Settled ($1.5B)](https://copyrightalliance.org/participating-bartz-v-anthropic-settlement/); fairness hearing April 2026 | Piracy vs. legal acquisition |
+| **Thomson Reuters v. ROSS** | [Fair use denied](https://www.dglaw.com/court-rules-ai-training-on-copyrighted-works-is-not-fair-use-what-it-means-for-generative-ai/) (Feb 2025); [on appeal](https://www.courtlistener.com/docket/70622297/thomson-reuters-enterprise-centre-gmbh-v-ross-intelligence-inc/) (3rd Circuit, briefs filed) | Competitor trained on copyrighted headnotes |
+| **Getty v. Stability AI** | [UK claim failed](https://www.mishcon.com/news/getty-images-v-stability-ai-unpacking-the-high-courts-judgment) (Nov 2025); US trial set Jan 2028 | Image model trained on copyrighted photos |
 
-[**75+ active copyright lawsuits**](https://chatgptiseatingtheworld.com/2025/10/08/status-of-all-51-copyright-lawsuits-v-ai-oct-8-2025-no-more-decisions-on-fair-use-in-2025/) against AI companies as of late 2025. Three fair use rulings so far (2 for, 1 against); three cases on appeal. No definitive appellate ruling yet.
+[**77 active US copyright lawsuits**](https://chatgptiseatingtheworld.com/2026/02/06/music-labels-and-spotify-sue-shadow-library-annas-archive-for-copyright-infringement-and-secure-default-judgment-ai-lawsuits-hit-77-in-us/) against AI companies as of February 2026 ([112 worldwide](https://chatgptiseatingtheworld.com/2026/02/25/latest-world-map-of-copyright-suits-v-ai-cos-total-112-feb-25-2026/)). Three fair use rulings so far (2 for, 1 against). No definitive appellate ruling yet.
 
 </div>
 
@@ -249,7 +249,7 @@ The EU and US have taken **opposite approaches** to AI regulation.
 
 # Deepfakes and elections: the 2024 test
 
-<div class="warning-box" data-title="The largest election year in history (3.7 billion eligible voters, 72 countries)">
+<div class="warning-box" data-title="The largest election year in recorded history (3.7 billion eligible voters, 72 countries)">
 
 Key incidents:
 - [**AI robocalls**](https://www.fcc.gov/document/fcc-issues-6m-fine-nh-robocalls) impersonated Biden urging NH voters not to vote (creator fined **$6M**, criminally indicted)
@@ -259,7 +259,7 @@ Key incidents:
 
 </div>
 
-<div class="note-box" data-title="The surprising finding">
+<div class="note-box" data-title="A surprising finding">
 
 Harvard's Ash Center concluded it was ["the apocalypse that wasn't."](https://ash.harvard.edu/articles/the-apocalypse-that-wasnt-ai-was-everywhere-in-2024s-elections-but-deepfakes-and-misinformation-were-only-part-of-the-picture/) The News Literacy Project found cheap fakes (non-AI manipulations) were used **7× more often** than genuine AI-generated content. The feared "AI disinformation tsunami" didn't fully materialize — but the *infrastructure* for it now exists.
 
@@ -349,7 +349,7 @@ A survey of AI researchers ([Grace et al., 2025](https://arxiv.org/abs/2502.1487
 </div>
 
 ---
-<!-- _class: scale-85 -->
+<!-- _class: scale-80 -->
 
 # What's next for LLMs? The scaling wall
 
@@ -361,14 +361,21 @@ A survey of AI researchers ([Grace et al., 2025](https://arxiv.org/abs/2502.1487
 
 </div>
 
-<div class="tip-box" data-title="The response: make models smaller and smarter">
+<div class="tip-box" data-title="The mainstream response: make models smaller and smarter">
 
 - **LoRA** ([Hu et al., 2021](https://arxiv.org/abs/2106.09685)): Fine-tune with 10,000× fewer parameters by injecting low-rank adapters
 - **Mixture of Experts** (MoE): Activate only a fraction of parameters per token (see [Lecture 25](../week9/lecture25.html))
 - **Distillation**: Train small models to mimic large ones (see [Lecture 19](../week7/lecture19.html))
 
 </div>
 
+<div class="important-box" data-title="More radical response: new architectures">
+
+- **Neural ODEs** ([Chen et al., 2018](https://arxiv.org/abs/1806.07366)), **liquid networks** ([Hasani et al., 2021](https://arxiv.org/abs/2006.04439)), **state space models** ([Gu & Dao, 2023](https://arxiv.org/abs/2312.00752)) — non-transformer architectures that can learn more efficiently from less data
+- Models that learn to modify their own weights and architecture during inference (see next slide)
+
+</div>
+
 ---
 <!-- _class: scale-80 -->
 
@@ -446,7 +453,7 @@ Every design decision embeds values. Every system reflects choices. Every deploy
 You are graduating into a world where:
 - AI agents can write code, browse the web, and use your computer
 - Models can strategically deceive their own trainers
-- 75+ copyright lawsuits are reshaping intellectual property law
+- 77+ US copyright lawsuits (112 worldwide) are reshaping intellectual property law
 - Companies are laying off workers based on AI's *anticipated* future capabilities
 - No country has figured out how to regulate this technology