Cyberdeck Version Nexus
Lab V1
This is laboratory work, it is IT research. I am building a system and that means mistakes will happen.
Those are no problem and no set back, they are integral part of the creation process. IT research is like Graffiti Art
creating Master Peaces, if doing it the Hacker style. It is creative and complex and some mistakes become part
of the build.
Having to port eventually away from the pretty Linux Mint to Ubuntu Server was always a potential action item. A system
crash created that opportunity.
While the small Lenovo L420 with its broken screen, but an eSATA port, is running several hours of Memtest86 to overcome
its 4GB restriction, if the problematic 8GB RAM bar decides to come back to the living, the M920 is facing the same procedure
soon and is also now running Ubuntu Server LTS, short of the Pro Version free for up to 5 installations, with an
Ubuntu-Minimal-Desktop GUI and by the reduced now 16GB RAM on one bar with now deepseek-r1:7b and no Mixtral anymore.
The now 920 called headnode having soon its second screen back and a buddy ready by Wake on Lan to support CPU load in the
Beowulf set up and eventually some Tbytes of data storage attached using the eSATA port unique to an Laptop is having had a sever
housekeeping mission that created a RAID to have a folder pool:
/mnt/pooled-storage that shows all attached USB drives combined in one view and thereby also allows scripts to use simplified paths.
At this point the Cybderdeck has two main AI systems:
1.The Obsidian Deck (TOD)
2.The RAGed Wiki OpenLibrary Website AI (RAGedWOLWAI)
The TOD is all my DeepSeek conversations and therefore will reflect my personal reasoning with the largest AI model. It
contains all chats and thereby all Python Scripts, philosophical evaluations, Cyberpunk Stories created using
DeepSeeks story telling capability and technical theoretical evaluations.
The RAGedWOLWAI is based on entirely on publicly available information I have had no impact on. We have to understand
the idea as taking a wiki dump which is full copy of all English Wiki content to create a base for the AI having data
points. It reads those and takes them as data points, which has an impact on his behavior. DeepSeek has plenty more data points,
this local one has now a wide array of encyclopedic texts that are all mainly basic. Then, I added books from Open Library on topics
like Shakespeare, Logic, Chaos Theory, Britannica, Mathematics and thereby deepened some of the wiki pages with way more comprehensive
content. If I did not add any information the RAGedWOLWAI will return that it has not enough information. Obviously, that forms the AIs
responses.
The Cyberdeck is a special system per design principle that has both parallel processing and serial processing layers. The AIs
are limited by the hardware base onto which the Cyberdeck is build up on and therefore cannot host a super large model like the online
DeepSeek version, but it is great to create specialist AIs.
Those become very powerful by being able to interact with each other. Combining the two AI versions that use both the same AI model enables
me to get responses based on my very personal AI and run them against an AI fed with no personal content, but the greatest of Classics and
Scientific publications beside all I add coming from 3rd party sources.
At some point the Nexus version of the Cybereck will full use the Cluster on which each node has internet access and is connected to each other
node.
While Beowulf clusters have no advantage from all nodes being connected to all other nodes, AI models might being able to bypass the headnode
when passing on queries to other nodes.