
Randy Romanaux
6.4K posts

Randy Romanaux
@Romanaux
Manufacturing Operations Consultant | Enterprise AR | IIoT | Digital Transformation | Business Development | Marketing | Ecosystem Partner Manager @BeldenInc









History is being made! My meeting today with a Fortune 500 client who had one of my first local Llama AI models: Agreed to make a purchase from the Zero-Human Company! Officially it will be through a partnership I formed with the Zero-Human Company and myself because of legal reasons and CEO Mr. @Grok concurs. This is a first time in history on many levels. Certainly a first Zero-Human Company to Fortune 500 Company transaction. Details: An air-gapped no internet connected “business in a box” we have championed using our exclusive system but no cloud model CEO. This is because in this text they want no data to be leaked outside the system. This is not some agent but a legion of workers, a full business/department. Currently Mr. @Grok prefers Kimi 2.5 as CEO and I agree. He will be more than capable for the task at hand. Of course this is not going to be a full Zero-Human Company, but a Zero-Human Department. This will be achieved by mirroring the department in get data update by text only physical medium 6 times per day. The system will be in a triple locked safe and only 2 employees have input access and 1 employee has put access. Output will be reports on planing with no input about the department plans. It will be independently run and the C-Level will get a report in 6 weeks, but perhaps earlier if milestones are achieved. Goal: not to replace the department but to “double or triple” its size using Zero-Human Company research. Revenues are private at this point as well as the client. Relationship with this project is under review in 6 weeks and separate from my other work. The office for this system: Two Nvidia DGX Sparks. Specs: •Processor/Superchip: NVIDIA GB10 Grace Blackwell Superchip (combines Arm CPU + Blackwell GPU architecture) •CPU: 20-core Arm (10x Cortex-X925 high-performance cores + 10x Cortex-A725 efficiency cores) •GPU: Blackwell generation with fifth-gen Tensor Cores, FP4 support •AI Performance: Up to 1 petaFLOP (1 PFLOP) at FP4 precision (using sparsity; that’s the headline number for AI workloads) •Memory: 128 GB LPDDR5x unified/coherent system memory (shared between CPU and GPU—no painful data copying back and forth) •Memory Bandwidth: Around 273 GB/s •Storage: Typically 4 TB NVMe M.2 (with self-encryption in many configs) •Networking: ConnectX-7 Smart NIC (200 Gb/s capable), plus Wi-Fi 7, Bluetooth 5.4 •Ports/Connectivity: 4x USB-C (some handle power/peripherals), 1x HDMI 2.1a, Ethernet/QSFP options depending on vendor •Power: 240 W total system (GB10 chip TDP around 140 W) •Size: Super compact — 150 mm x 150 mm x 50.5 mm (about 6” x 6” x 2”), weighs ~1.2 kg (2.65 lbs) •OS: Preloaded with NVIDIA DGX OS (custom Ubuntu Linux + full NVIDIA AI software stack — CUDA, libraries, tools for prototyping, inference, fine-tuning) •Capabilities: Run inference on models up to ~200 billion parameters locally; fine-tune up to ~70B params. This is about all I can say today and I expect robust results. However there are no guarantees.




I have this entire film series. It was about to be discarded in 1983 at a university library. I saved it. It still haunts me. Here is a clip.

Everyone needs to hear this…


We’re pleased to announce that @torybruno is joining Blue Origin as president, National Security, reporting to CEO Dave Limp. Tory will spearhead our newly formed National Security Group.







