Skip to content

Commit f83b4c5

Browse files
mastercybmastercyb
committed
improvement of bostrom intro to for ai geeks
1 parent b7ff27e commit f83b4c5

15 files changed

+73
-30
lines changed

‎.gitignore

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -9,4 +9,5 @@ __pycache__/
99

1010

1111
logseq/bak
12-
build/
12+
build/
13+
/logseq/.recycle

‎journals/2024_08_03.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,3 @@
11
- article on [[relevance machine]]
22
- thoughts on [[probabilistic collective computations]]
3-
- [[cybergraph, rm and llm]] and strange [[reality of foundation models]]
3+
- [[knowledge graphs and llms]] and strange [[reality of foundation models]]

‎pages/concepts.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@ tags:: cyber
1212
- [[avatar]] with [[features]]
1313
- [[learning incentives]] based on [[yuma]]
1414
- [[bootloader]] as research and development ground
15-
- [[cybergraph, rm and llm]]
15+
- [[knowledge graphs and llms]]
1616
- [[cyb]] robot for interactions
1717
-
1818
- [[distributed neural network]] as foundation for [[superintelligence]]

‎pages/cybergraph, rm and llm.md

Lines changed: 0 additions & 14 deletions
This file was deleted.

‎pages/cybergraph.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -29,5 +29,5 @@ alias:: content oracle
2929
- cybergraph of [[spacepussy]] serve the purpose of memes phenomena
3030
- understand deeper [[intelligence measures]]
3131
- discover all [[concepts]] of [[cyber]]
32-
- [[cybergraph, rm and llm]]
32+
- [[knowledge graphs and llms]]
3333
- implemented as [[cyber/graph]] [[module]] of [[cyber-sdk]]

‎pages/elon launch rocket.md

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
alias:: elon launch, launch rocket
2+
3+
- ![elon launch rocket](https://emerald-raw-leopon-384.mypinata.cloud/ipfs/QmSd59ndqbEjmP4hgQVFUN4rUrHHFGTugGpyCYY9mD46U7)

‎pages/explicit knowledge.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
alias:: shared history
1+
alias:: shared history, explicit
22

33
- information that is directly stated and clearly represented in [[cybergraph]]
44
- readily available just by traversing [[cybergraph]]

‎pages/graph neural network.md

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
tags:: gnn, gnns, graph neural networks

‎pages/implicit knowledge.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,5 @@
1+
alias:: implicit
2+
13
- [[information]] that is not directly stated
24
- but can be inferred from the [[explicit knowledge]] in the [[cybergraph]]
35
- orthogonal to [[explicit knowledge]]

‎pages/introduction to bostrom for ai geeks.md

Lines changed: 29 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -1,24 +1,44 @@
1-
- bostrom is NOT yet another ai coin
2-
- it is very powerful foundational technology for advanced superintelligent civilization
1+
- [[bostrom]] is NOT yet another ai [[coin]]
2+
- it is very powerful foundational technology for advanced superintelligent [[civilization]]
33
- in this article i will boil down all essential ideas into coherent understanding how [[bostrom]] can empower
44
- existing ai field which i will refer as classical ai
5-
- and advance emerging field of collective ai or superintelligence
5+
- and advance emerging field of collective ai
6+
- as we believe its the only viable way to build [[superintelligence]]
67
- ## attention is not enough
78
- you used to rely on a data you got
89
- you have the dataset
910
- you design neural network architecture
1011
- then, you train the model
1112
- and boom, now the model can predict some output based on any input
1213
- sounds really cool, and is powerful indeed, except the dataset thing in this story
13-
- the good answers to ask
14-
- how does you model could define truth?
14+
- now the good answers to ask: how does you model could define truth?
1515
- and the short answer - it cant
1616
- i will make a bold claim here that truth can not be defined without 3 ideas in foundation
17-
- [[cryptographic proofs]]
1817
- [[knowledge graphs]]
18+
- [[cryptographic proofs]]
1919
- [[token engineering]]
20-
- ![image.png](https://emerald-raw-leopon-384.mypinata.cloud/ipfs/QmPnEpPhd77ysSfaACJd5Kc6prrrsUfTCAmPo977n2tcih)
21-
- source: [Unifying Large Language Models and Knowledge Graphs: A Roadmap](https://cyb.ai/oracle/ask/QmdGEYVKo1sRURzbj83UMtr77EL6GHUB2taJbnpTynEQKT)
20+
- ## knowledge graphs and llms
21+
- jump for a second to this article: [Unifying Large Language Models and Knowledge Graphs: A Roadmap](https://cyb.ai/oracle/ask/QmdGEYVKo1sRURzbj83UMtr77EL6GHUB2taJbnpTynEQKT)
22+
- ![image.png](https://emerald-raw-leopon-384.mypinata.cloud/ipfs/QmPnEpPhd77ysSfaACJd5Kc6prrrsUfTCAmPo977n2tcih)
23+
- {{embed ((66b2fb4e-c73b-4133-aab3-11df2e8d1436))}}
24+
- the article explain why [[llm]] will never be enough to reach general [[intelligence]] alone
25+
- in short [[knowledge graph]] advantage is
26+
- easy to understand and structure as they are more about [[explicit knowledge]]
27+
- possible to evolve because they are based on widely accepted [[triples]]
28+
- essential to plan, make decisions and reason:
29+
- so the claim is simple: [[knowledge graphs]] coupled with [[gnns]]
30+
- are essential for the next generation of architectures and by this article we propose such architecture
31+
- ## cryptographic proofs and llms
32+
- be believe that authenticity of models is a serious bottleneck for ai alignment and more
33+
- its a shame that so technologically advanced industry in a broad sense
34+
- still does not give a shit about possibilities of [[hashing]], [[pubkey cryptography]] and [[logical clocks]]
35+
- its kinda impossible to build multiparty protocols without these primitives
36+
- yep, i am ware about [zkml movement](https://github.com/worldcoin/awesome-zkml)
37+
- but this is a drop in the ocean given the [[knowledge graphs and llms]] argument
38+
- we need something foundational fully authenthicated knowledge graph tech
39+
- if we want to significantly advance in the field of [[superintelligence]]
40+
- ## token engineering and llms
41+
-
2242
- ## cybergraph
2343
- the core of the idea is [[cybergraph]] - merkelized timestamped data structure of links between ipfs hashes submitted by anyone
2444
- for clarity we refer to:
@@ -84,7 +104,7 @@
84104
- hence, we can refer to [[cybergraph]] as objective knowledge of everyone
85105
- ## cybergraph vs knowledge graph
86106
- cyberlinks are fully authenticated qaquadruples
87-
- when, who and what are based on cryptographic technics
107+
- [[when]], [[who]] and [[what]] are based on cryptographic technics
88108
- so unlike conventional knowledge graphs the information is crystal and true by design
89109
- basic idea is that if i want say in [[triple]] world i would just say
90110
- [[elon launch roocket]]

‎pages/knowledge graphs and llms.md

Lines changed: 27 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,27 @@
1+
-
2+
- good explanation on fundamental difference between [[knowledge graphs]] and [[llms]]
3+
- [unifying large language models and knowledge graphs: a roadmap](https://cyb.ai/oracle/ask/QmdGEYVKo1sRURzbj83UMtr77EL6GHUB2taJbnpTynEQKT)
4+
- TODO [[create visualization]]
5+
- id:: 66b2fb4e-c73b-4133-aab3-11df2e8d1436
6+
| property | knowledge graphs | large language models |
7+
|---------------------------|--------------------------|--------------------------|
8+
| representation | structured: [[triples]] | unstructured: [[text]] |
9+
| knowledge storage | [[explicit]] | [[implicit]] |
10+
| interpretability | high | low |
11+
| generalizability | low | high |
12+
| factual accuracy | high | low |
13+
| reasoning | symbolic | probabilistic |
14+
| knowledge update | easy | hard |
15+
| decisiveness | high | low |
16+
| unseen facts | no | yes |
17+
| incompleteness | high | low |
18+
- ![graphs and llms](https://emerald-raw-leopon-384.mypinata.cloud/ipfs/QmZoAhUsB1KAEbnLCWcMAohtWsAXCDZuetJALrEe5JEnSC)
19+
-
20+
- insights are the following that [[knowledge graphs]] and [[llms]] are
21+
- fundamentally different
22+
- must interact in any [[intelligent]] system
23+
- key to [[superintelligence]]
24+
- go on to dive into
25+
- [[cybergraph]] and [[relevance machine]]
26+
- as beautiful counterpart [[probabilistic model]] for [[llm]]
27+
- in a modern [[soft3 stack]]

‎pages/reality of foundation models.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -6,4 +6,4 @@
66
- [[implicit knowledge]] is derived from [[explicit knowledge]]
77
- operates in [[full content space]]
88
- [[authentication of information]] by design
9-
- tradeoffs can be understood in the article [[cybergraph, rm and llm]]
9+
- tradeoffs can be understood in the article [[knowledge graphs and llms]]

‎pages/relevance machine.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -26,7 +26,7 @@ alias:: rm
2626
- with [[strong predictive power]]
2727
- related articles
2828
- [[reality of foundation models]]
29-
- [[cybergraph, rm and llm]]
29+
- [[knowledge graphs and llms]]
3030
- [[soft3 and machine learning]]
3131
- [[supported gpu]]
3232
- discover all [[concepts]]

‎pages/triple.md

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
alias:: triples

‎pages/tutorials.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,7 @@
11
tags:: bostrom
22

3+
- concepts
4+
- [[knowledge graphs and llms]]
35
- [[go-cyber]]
46
- TODO install cyber cli
57
- TODO sync node

0 commit comments

Comments
 (0)