It struck me while walking the dog.
AI -- and especially Agentic AI -- is a multiplier.
And the reason some orgs are struggling with it... they're (wait for it)
MULITPLYING BY ZERO
If you try to agent a thing you're not able to do currently, you WILL fail 100%.
Been playing with grok imagine to restore old family photos. Amazing to see what it can do with 1950/1960's slide scans. I have an n8n workflow that can automate this against the api
I trained this @ltx_model LTX 2.3 LoRA of George Costanza at home on my 5090 in about a day with AI Toolkit. I generated this 30 second video with @ComfyUI on my 5090 in 6 minutes. Open source is, always has been, and always will be, the future of generative AI. (SOUND ON)
I joined the claw family at last tonight. Starting with some data collection tasks prepping for custom ai generation in the future. Also fun to a/b openclaw with nemoclaw.
Accomplishment for today was getting the refactored backup server sorted. This is offsite storage to ensure that anything sensitive I do is preserved. Still 100% sovereign, just a basic intel atom box with a few hard drives. Did find out that I couldn't have all the drives I wanted at the same time as the 10Gbe NIC due to platform limits...oh well.
Also have the minimax 2.5 model running on the DGX spark with 140k context capability. This is an impressive model for local agents and coding, though oddly it had issues with the tool calling mode in n8n.
Got the network switched over today. Old Unifi gear heading to ebay soon, new 25Gbps hardware installed.
It was a bit of a rough time getting all the akash nodes updated. Had a lesson that I need to stop the provider whenever a node is going offline or there's a chance it will hit a limit and close leases. Sorry to anyone who was on https://t.co/HJEqo7axKw today! I ensured there was enough free capacity, but alas. The cluster now has faster networking and is ready for future ISP / power upgrades. Good learning day for kubernetes / akash management which will help as I scale.
Also got my new offsite backup server ready. Critical data will remain sovereign and redundant.
The ceph one was overdue. My cluster has 4x1TB drives with a replication factor of 3. Basic math would expect 1200-1300 GB usable, yet the provider only showed a max of ~900. Turns out rook bakes some additonal room into the schema to allow for node failure/rebalancing so 900 is right.
Accomplishments in sovereign compute for today:
* n8n/qdrant loading and retrieval experience
* reracked upgraded gigabyte EPYC server with 64 core CPU
* dug into rook-ceph and confirmed everything is correct on akash
* cost-performance analysis for next hardware options
* progress on the church sermon training data ingest
The good news is that these bent pins had a happy ending. I was able to remove the damaged bumper from the side and reset the cpu which got the server to boot. I then noted that two of the memory sticks were not detected and the BMC reported no cpu temp. Pulled it a third time and found another damaged pin on the other side, fixed that and now everything works. Amazing that the cpu worked at all with the still bent pin.
I was prompted to sign into mspaint today.
@Microsoft when you look back and wonder what went wrong for Windows...
It was this. This is what went wrong.