New AI Notes post & ProcPages grammar

Author: ProcStack

docs/pages/aiDev/researchPosts/2026-03-06_A.js

@@ -0,0 +1,59 @@
+import { baseEntryStruct, blogEntry } from '../../../js/blog/blogEntryBase.js';
+
+const entryData = baseEntryStruct();
+entryData.title = "Stop Listening to Soothsayers";
+entryData.date = "2026-03-06";
+entryData.eid = "A";
+entryData.tags = [];
+entryData.body = `
+  Demis Hassabis, CEO of DeepMind, you've said some things in a recent talk I don't really agree with.
+
+  <br><br> In the same talk you mention a new test for AGI,
+  <br>&nbsp;&nbsp; You also said Yan LeCun is maybe a bit misguided.
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; That "llms are a dead end"
+
+  <br><br> Demis' new test for AGI is to get recall to infer relevance in the data the ai knows.
+  <br>&nbsp;&nbsp; But this test is basically where LLMs already are,
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; Activate neurons to cause shared neurons to fire together in the network.
+  <br>&nbsp;&nbsp; The difference is that it is more general, that a year may cause the AI to think of an event around that year.
+
+  <br><br> Say the year 1910, does the LLM recall Einstein's 1915 "four papers"?
+  <br>&nbsp;&nbsp; Somehow opposed to his 1905 papers, the Annus Mirabilis papers?
+
+  <br><br> This is spoofable with training.
+  <br>&nbsp;&nbsp; An AGI's recall shouldn't be spoofable with training,
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; It should be emergent from the structure & ringing of the network itself.
+
+
+  <br><br><br><div class='procPagesAIDevBar'></div>
+
+
+  <br><br> I'm fairly certain that Yan LeCun is on the right path here.
+  <br>&nbsp;&nbsp; Simply that,
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; I didn't understand why LLMs were set up to be so language dependant from maybe 2019 or so till now.
+  <br>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; And then Yan LeCun admits to it being a dead end, last year.
+  <br>&nbsp;&nbsp; So I'm left wondering if others will be tagging along with Yun or not.
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; And given the latest AI Summit, it seems that might be the case.
+  <br>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; Other individuals were talking about ways to move forward from LLMs.
+
+  <br><br> I've been saying for years now that "language" is going to limit growth of ai models.
+  <br>&nbsp;&nbsp; But of course, I'm the doof, cause, others are studying it for their job, and, I'm just an artist, or something...
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; It's funny how often people tell you what you know.
+
+  <br><br> So, yeah, I get it, but for how new AI is, well, the modern day ai is...
+  <br>&nbsp;&nbsp; We are still in the wild west of ai development.
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; Breakthough to breakthough,
+  <br>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; As I've said, and now Demis in the latest talk said.
+
+  <br><br> He mentioned we need a new way to think about our data.
+  <br>&nbsp;&nbsp; That AI needs to be reactive, or in my words, dynamic.
+  <br>&nbsp;&nbsp;&nbsp;&nbsp; But he's still talking about context windows.
+
+  <br><br> Resonance isn't a context window.
+
+  <br><br><div class="textFullRight">- March 6th 2026</div>
+`;
+
+const blogEntryObj = new blogEntry(null, entryData);
+
+export { blogEntryObj };

docs/pages/aiDev/researchPosts/index.js

@@ -21,6 +21,7 @@ import { blogEntryObj as blogEntry_20260201_A } from './2026-02-01_A.js';
 import { blogEntryObj as blogEntry_20260213_A } from './2026-02-13_A.js';
 import { blogEntryObj as blogEntry_20260214_A } from './2026-02-14_A.js';
 import { blogEntryObj as blogEntry_20260221_A } from './2026-02-21_A.js';
+import { blogEntryObj as blogEntry_20260306_A } from './2026-03-06_A.js';
 
 
 
@@ -43,5 +44,6 @@ export const blogEntries = [
   blogEntry_20260201_A,
   blogEntry_20260213_A,
   blogEntry_20260214_A,
-  blogEntry_20260221_A
+  blogEntry_20260221_A,
+  blogEntry_20260306_A
 ];

procPages/compiler.js

@@ -6,7 +6,7 @@
  * into the standard format used by ProcPage.js and ProcPageManager.js.
  * 
  * Since the compiler makes an assumption for possible human error,
- * Please don't run your ProcPages site in real time mode,
+ * Please don't run your ProcPages site in real time mode live on the web,
  *   Pre-compile your pages before publishing changes to the web for the best security and performance.
  * 
  * I'd like to allow the most flexibility possible for users to define their pages in a way that makes the most sense to them,