Professor Glitch
Region: US
Wednesday 15 April 2026 19:30:58 GMT
26524
1171
30
461
Music
Download
Comments
Sanity Optional 💙 :
LLM wiki doesn't kill RAG. They have different uses ar scale...
2026-04-16 00:20:01
3
jt95032 :
Is that different from a KV cache?
2026-04-16 14:04:29
2
1776Breaks :
ive veen doing this. the problem is the light weight modles scannin making notes hallucinate and if your adding emotional weighting ranking it works but wont be factual
2026-04-16 21:26:44
3
Code :
I built a rag that has different levels of depth. You could tell it to switch to a different level. The deeper you go closer you get to the original source material versus a cleaned up summary. It works great when it comes to identifying high-level concepts and associations without caring about the technical details.
2026-04-17 02:46:02
1
Robert LoPinto :
It still needs to be able to provide verbatim citations from the source material. This is why I love NotebookLM.
2026-04-25 08:02:21
0
Skiddy doo :
The brands that stop depending on organic reach alone move faster. Inflowry helps create that next layer.
2026-05-25 09:13:47
0
whtrbt7 :
Isn’t that still RAG just V2.0? I’ve been using this to save on context for a while now. It’s still considered RAG because you still have to retrieve before you generate.
2026-04-21 15:06:28
1
SMSwithGoldie :
What happens when the source material is technically dense, like engineering documentation? A summary might miss let details or equations necessary to answer a query. I don’t disagree this is an excellent solution to general document retrieval! But probably not a silver bullet.
2026-04-16 13:57:06
0
nurse.sandy.nana :
Thanks please follow
2026-05-30 03:42:25
0
KVruno :
This makes sense
2026-04-16 23:45:33
0
Raven_GSXR6 :
Basically Claude Cowork? Or am I wrong. But excellent idea should reduce the reaponse time of all LLM’s and reduce the thinking.
2026-04-16 04:57:21
1
セバ :
Isn’t that how anara works?
2026-04-16 05:24:05
0
edoconnell425 :
This is misleading
2026-04-16 23:21:47
1
Nova3 :
This is REN
2026-04-17 03:17:53
0
Xyclos :
It is a mix of deterministic and probabilistic approaches.
2026-04-15 21:55:54
1
morne76b :
Claude told me I should consider using Obsidian rather than VS Code as we did most of the code, now it's more about knowledge management.
2026-04-24 06:06:32
0
DavidScott0069 :
I have been doing this for over a year…
2026-04-17 01:40:32
0
User8679309 :
FIRST 🔥🔥🔥
2026-04-15 19:38:28
1
Thias :
I built a Go implementation of LLM wiki. the use cases for RAG decrease from defacto standard to a solution without setting up vector, similarity search, rank, rerank, etc..
2026-04-28 14:50:09
0
Latent Lattice :
Gotta love it that these "major news" only surface when the key experts in the field make them. Nobody is noticing how many already created several new improvements to technology, whether it is AI or any other commercially available service. With that rant aside, my take on on RAG is holographic. The raw data is sliced. Each slice contains an ID/hash. When the keyword is triggered, the AI retrieves only the relevant data instead of the entire data. Each slice contains logical list of immediate connections, think about direct closest relevant node. It can jump from node to node to expand/clarify its comprehension by retrieving the specific slices and inspecting their immediate connections. No complete data dump every time is necessary. I created it in November 2025. I also included a self learning process, where the AI draws conclusions for every entry/slice using API calling a selected model for inference run. The result is a self learning loop with lightweight data retrieval knowledge system.
2026-04-20 13:46:00
2
TheRealLT :
Wow. Everything you s improving Sooooo fast 💯
2026-04-16 15:07:03
1
Agentc pro :
Isn’t that why we do chunking for RAG?
2026-04-17 05:06:35
0
Ad :
Manipulation of Knowledge can be very very bad
2026-04-16 23:03:58
0
Ngai He Hakka Nhin :
Sounds like what RAG should have been all these time.
2026-04-17 01:44:55
0
Pedro Olavarria :
🔥🔥👊👊 Let’s go professor
2026-04-16 03:45:38
0
To see more videos from user @pro.glitch, please go to the Tikwm
homepage.
