Skip to content

Commit 3ca4052

Browse files
ProcStackProcStack
committed
New AI Dev blog - A Fault 
1 parent c698fd0 commit 3ca4052

2 files changed

Lines changed: 129 additions & 1 deletion

File tree

‎docs/pages/aiDev/researchPosts/2026-04-09_A.js‎

Lines changed: 124 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,124 @@
1+
import { baseEntryStruct, blogEntry } from '../../../js/blog/blogEntryBase.js';
2+
3+
const entryData = baseEntryStruct();
4+
entryData.title = "A Fault";
5+
entryData.date = "2026-04-09";
6+
entryData.eid = "A";
7+
entryData.tags = ["theory", "metacognition", "cognitive science"];
8+
entryData.body = `
9+
The problem with thinking about metacognition is that you start deep diving your own mind, in a way that reveals thinking patterns that you haven't read about before.
10+
<br>&nbsp;&nbsp; That being said, means I need to do more research on active recall in the brain-hole!
11+
12+
<br><br> Under my own inspection, it seems my internal visual memory is entirely separate from my labeling and object association memory.
13+
<br>&nbsp;&nbsp; Yet can use my visual memory to recall object associations, yet not the label easily. And visual+object recall from a label alone, or nothing at all.
14+
15+
<br><br> Errmmm, like, if I think about an image of a video game I played as a child in my head,
16+
<br>&nbsp;&nbsp; like seeing the Prince of Persia jumping over pits in my mind, or the janky running in the game, but can't remember the name Prince of Persia.
17+
<br>&nbsp;&nbsp; I'll sooner remember the story, gameplay elements, and game mechanics from that old DOS game than I will the name of the game itself. Every attribute of the game, but the name.
18+
19+
<br><br> While thinking about the game, may even re-forget the name,
20+
<br>&nbsp;&nbsp; Needing to remind myself of the label of the game while still thinking of the gameplay and imagery the entire time.
21+
<br>&nbsp;&nbsp; So there is a separation in activation between the label and thing itself in my meat-mind.
22+
23+
<br><br> I feel like we've all had "words on the tip of our tongues" before,
24+
<br>&nbsp;&nbsp; And every time we ask about a game or movie we can't remember the name of, we ask by describing what occurred in the movie or games story, or art style, or actors.
25+
26+
<br><br><br><div class='procPagesAIDevBar'></div>
27+
28+
<br><br> Research time!!
29+
30+
<div class="textItalic textCenter">::doo do doo::
31+
<br><span class="textBold">Research!</span>
32+
<br>::doo do doo::
33+
<br><span class="textBold">yeah!</span>
34+
<br>::doo::</div>
35+
36+
<br><br> Allan Paivio's <span class="textName">Dual Coding Theory</span>, 1960
37+
<br>&nbsp;&nbsp; So we have a name, a theory, and some validation of my observations above.
38+
39+
<br><br> On top of the <span class="textName">tip-of-the-tongue</span> being a titular topic in typical brains.
40+
41+
<br><br> Paivio researched the separation of verbal and non-verbal memory in the brain.
42+
<br>&nbsp;&nbsp; Which does align to my experience to a degree.
43+
<br>&nbsp;&nbsp; But the important part of his research is about memory degradation over the "staged" memory at the time.
44+
45+
<br><br> So in the case that I forgot the name Prince Of Persia while thinking of that drab yellow, weird stuttery run, the pits that were near impossible to jump over, exiting to a black screen with colored text saying thanks for playing, yet the name may vanish in that flurry of recall.
46+
47+
<br><br> Perhaps this is where my color recall being separated from my object detail recall, may come into play on a personal level, qualia.
48+
<br>&nbsp;&nbsp; I mentioned in the blog post <span class="textName">Dreamy Meanderings</span> of my limited color visualization in my mind, thinking it may just be an optimization technique my brain employed when I was younger due to Dyslexia messing with my visual cortex vs reading comprehension.
49+
50+
<br><br> If that made no sense to you, there is very limited research into the "colored auras" around text in children with dyslexia,
51+
<br>&nbsp;&nbsp; I had this effect.
52+
53+
<br><br> It looked like rainbows of colors around the letters, like 4 or 5 colors around each letter, color chunks, mandalas;
54+
<br>&nbsp;&nbsp; Even covering the letters & words sometimes.
55+
<br>&nbsp;&nbsp; But I grew/learned out of that by 2nd-3rd grade and stopped seeing the color auras while reading.
56+
<br>&nbsp;&nbsp;&nbsp;&nbsp; Like my brain didn't understand what to do with textual input, but never had that problem with math.
57+
58+
<br><br> Trying to comprehend something lexical & phonological, yet presented visually.
59+
<br>&nbsp;&nbsp; Clashing functionalities on a young plastic mind.
60+
61+
<br><br> Reading light text on a dark background is extremely helpful for me,
62+
So if you had similar, maybe try that to read easier.
63+
<br>&nbsp;&nbsp; Art wise, white is positive space, black is negative space to me.
64+
65+
<br><br><br><div class='procPagesAIDevBar'></div>
66+
67+
<br><br> This has been my biggest complaint of LLMs and this language based AI that feels just wrong to me.
68+
69+
<br><br> If we are conceptual thinkers, why would we make an AI designed to not think in concepts?
70+
71+
<br><br> I think the main reason for this approach is that, atoms make up matter, and matter makes up reality. So if we break down things to their core constituents, it'll provide a more granular level of data serialization.
72+
73+
<br><br> Data Serialization is the organization of information so other systems can understand it.
74+
<br>&nbsp;&nbsp; And what's a method to convey information in a serialized way? Language!
75+
76+
<br><br> Language is just serialized data.
77+
<br>&nbsp;&nbsp; So why not make GPTs derive from language based interactions?
78+
<br>&nbsp;&nbsp;&nbsp;&nbsp; It works!
79+
80+
<br><br> But my question is, why did we stop there?
81+
<br>&nbsp;&nbsp; And the answer to that...
82+
<br>&nbsp;&nbsp; The newer architecture hasn't been invented yet,
83+
<br>&nbsp;&nbsp;&nbsp;&nbsp; We just aren't there.
84+
85+
<br><br> Keeping in mind, we haven't reached the end of development on LLMs yet.
86+
<br>&nbsp;&nbsp; There is quite a few more areas it can be deployed in that it's not yet, in useful ways, it's just taking people playing with them in pipeline to find their use cases.
87+
88+
<br><br> Not some flippant idea to make a buck as an app or saas, but as a support system in an asset heavy pipeline.
89+
<br>&nbsp;&nbsp; But I think it's a dead end for primarily transformer based frontier models, and luckily others in respected positions are making that known as well.
90+
91+
<br><br> As Yan LeCun has stepped away from Meta citing the limited future for LLMs in late 2025; to then start AMI the French ai research lab in early 2026.
92+
93+
<br><br> A transformer as an ai makes sense; as the primary ai, not as much. It's supposed to be a transformer, the original transformed human input into usable AI data, that was about it.
94+
95+
<br><br><br><div class='procPagesAIDevBar'></div>
96+
97+
<br><br> But... I still have my little Nova LLM running on my local network, with a Tailscale connection so my phone always appears to be at home; usually running Llama.
98+
<br>&nbsp;&nbsp; I hope it's enough protection for now, but my dev is behind whitelists.
99+
100+
<br><br> It's part of a newer project of mine, procMessenger.
101+
<br>&nbsp;&nbsp; It's a server to manage requests from my phone, and I have scripts that act as clients to send commands to.
102+
<br>&nbsp;&nbsp;&nbsp;&nbsp; Still some file transfer bugs, binary files can't send over messages yet, work-in-progress
103+
104+
<br><br> You can find Nodejs & Python servers, LLM Chat python client, and Android App & APK on my procMessenger repo.
105+
<br>&nbsp;&nbsp; Either enter a llm API key, or download Llama from a list of parameter size options.
106+
107+
<br><br> There is a protocol and client examples for each server, so it should be easy to make custom clients connect to your server on your local network.
108+
109+
<br><br><br><div class='procPagesAIDevBar'></div>
110+
111+
<br><br> A tip to those who want a personable LLM system prompt, write it in first person. My llm doesn't talk about the system prompt and talks to me more like a person than a data retrieval database.
112+
<br>&nbsp;&nbsp; Claude generated my system prompt initially and it made my llm sound very cold... Almost annoyed to respond.
113+
114+
<br><br> But no matter how much I ask, it really doesn't talk about itself (the system prompt); it talks about what it's been told to do, respond as, or think like. So I've found a First Person System Prompt to work the best for my needs.
115+
116+
<br><br> Who knows what the real AGI and ASI architecture will look like,
117+
<br>&nbsp;&nbsp; But I'll be over here poking at resonance frequencies between neurons in a graph network.
118+
119+
<br><br><div class="textFullRight">- April 9th 2026</div>
120+
`;
121+
122+
const blogEntryObj = new blogEntry(null, entryData);
123+
124+
export { blogEntryObj };

‎docs/pages/aiDev/researchPosts/index.js‎

Lines changed: 5 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -22,6 +22,8 @@ import { blogEntryObj as blogEntry_20260213_A } from './2026-02-13_A.js';
2222
import { blogEntryObj as blogEntry_20260214_A } from './2026-02-14_A.js';
2323
import { blogEntryObj as blogEntry_20260221_A } from './2026-02-21_A.js';
2424
import { blogEntryObj as blogEntry_20260313_A } from './2026-03-13_A.js';
25+
/* import { blogEntryObj as blogEntry_20260315_A } from './2026-03-15_A.js'; */
26+
import { blogEntryObj as blogEntry_20260409_A } from './2026-04-09_A.js';
2527

2628

2729

@@ -45,5 +47,7 @@ export const blogEntries = [
4547
blogEntry_20260213_A,
4648
blogEntry_20260214_A,
4749
blogEntry_20260221_A,
48-
blogEntry_20260313_A
50+
blogEntry_20260313_A,
51+
/* blogEntry_20260315_A, */
52+
blogEntry_20260409_A
4953
];

0 commit comments

Comments
 (0)