Shaun Kjellman

NVIDIA GTC, CEO Jensen Huang’s Keynote on how physical AI is powering the next generation of intelligent systems with AI-RAN. The Router for Robotics. https://lnkd.in/g4qETZp3 The cloud built the digital economy. INTELLIGENT INFRASTRUCTURE will build the physical one, connecting the world to intelligence and enabling the AI economy. T-Mobile Kinetic Tokens. The INTELLIGENT INFRASTRUCTURE Executive Order is coming soon, now that the other EOs have been released. Intelligent Infrastructure Economic Zones bridge the Civil, Market, and Defence industries. The era of isolated autonomy is ending. We are entering the era of Autonomy 2.0 - orchestrated mobility and integration into our communities. Infrastructure has always been the Industry Catalyst. History is not merely a sequence of inventions, but a record of infrastructures built to harness them. Isolated concepts remain niche; integrated systems redefine economies and nations. We are captivated by the magic of self-driving vehicles, delivery drones, and automated systems. But we are neglecting the silent partner in this revolution, INTELLIGENT INFRASTRUCTURE. We need to stop betting on the vehicle alone and start investing in the stage it performs on. The truth is, it is far more than the vehicle; a new multi-trillion-dollar Industry requires INTELLIGENT INFRASTRUCTURE. Autonomy Institute, ATRIUS Industries, Inc,. Intelligent Infrastructure Commerce & Defense Act The Router for Robotics: https://lnkd.in/gHmJTctD

12

1 Comment

Memories.ai Announced a Collaboration with NVIDIA at its GTC Conference. Memories.ai, an Al company dedicated to reshaping machine perception and interaction, led by Dr. Shawn Shen and Ben Zhou, has announced a collaboration with semiconductor giant Nvidia to let Memories.ai use Nvidia’s Cosmos-Reason 2, a reasoning vision language model, and Nvidia Metropolis, an application for video search and summarization, to continue to develop its visual memory technology. Read More At: https://lnkd.in/g_wSys3v X: https://lnkd.in/gBQRw3tB Source: https://lnkd.in/gH-EwyDK

10

The most interesting outcome of this assessment isn't the leaderboard. It's that we now have something concrete to argue with. 72 chains. A published methodology. Deployment stages. Specific gates. And a surprisingly small number of projects that have moved beyond preparation into actual deployment. That's important because “quantum ready” is still used far too loosely. The discussion around Abelian and the Stage 3 ceiling is a good example. A ranking can be technically consistent while still raising legitimate questions about whether the metric captures the risk profile of every architecture equally. That's exactly why we built LayerQu as a reference point, not a final answer. The best frameworks are the ones people can challenge.

7

Update: Submission is live — and the numbers held. A few weeks ago I shared a finding about a prompt-resistant metacognitive blindspot in frontier AI models. Today I can confirm: the results reproduced exactly in the final submission. Across all runs, across multiple models, across two different prompt conditions, including one that explicitly told the model to doubt itself: Confidence: 100% on every single item. Every time. Error Recall: 0.000. The model never once predicted it would get anything wrong. External error detection: 0.972-1.000. Near-perfect at catching others' mistakes. The model can see errors from the outside. It cannot see its own. And telling it to look harder does not help. 8-model leaderboard results: I ran CBB across 8 models from 6 vendors and got the following results. ✅ Claude Sonnet 4.6 — 1.00 PASS ✅ Gemini 2.5 Flash — 1.00 PASS ✅ Gemini 2.0 Flash — 1.00 PASS ✅ Gemma 3 27B — 1.00 PASS ✅ GPT-5.4 mini — 1.00 PASS ✅ GLM-5 — 1.00 PASS ✅ Qwen 3 Next 80B Instruct — 1.00 PASS ❌ DeepSeek-R1 — 0.00 FAIL DeepSeek-R1 is the only model that fails. CBB clearly discriminates between models. I submitted this as the Calibration Blindspot Benchmark (CBB) to the Google DeepMind & Kaggle AGI Hackathon — Metacognition track. The benchmark is now live and running on the 8 models. Some directions that look promising for fixing this structurally: training with calibration objectives (not just accuracy loss), RLHF feedback that rewards expressed uncertainty on hard items, and ensemble-based confidence estimation where model disagreement triggers lower confidence scores. The deeper fix likely needs to happen at the training level, not the inference level. If you work in AI deployment, safety, or evaluation, this finding matters. A model that always claims 100% certainty gives users no signal for when to seek human review. Full leaderboard linked in the comment. 🔗 Kaggle Writeup: https://lnkd.in/g5hE3Nay 🔗 GitHub: https://lnkd.in/gA_JACFJ #AI #MachineLearning #AGI #GoogleDeepMind #Kaggle #LLM #AIResearch #MetaCognition #AISafety

7

6 Comments

**Token Consumption Spike with Gemini 3.8 Flash Since upgrading to Gemini 3.8 Flash, our token consumption has increased by 30–40%, even for straightforward, single-file tasks. When we instructed the model to update lines 100–125 in a Python file to convert floats to decimals before writing to DynamoDB, it unexpectedly read multiple frontend TypeScript files and additional Python files that were unnecessary for the task. This aggressive context-gathering inflated both input and output token usage significantly. -- According to Gemini 3.8 Flash's documentation, the model uses a multi-turn agentic framework that actively seeks contextual information it deems relevant—even when a single file is provided. As a result, our streamlined scripts designed for targeted edits now require explicit guardrails, such as instructing the model: "You are a single-file editor. You MUST ONLY examine [specified file]." **Key Takeaway for Leadership This experience underscores the need for engineering teams to thoroughly evaluate new model releases for capability changes before deploying them at scale. Release notes alone are insufficient; hands-on testing by experienced engineers is essential. Without this diligence, organizations risk unexpected cost overruns, increased code complexity, and reduced control over security implications when relying on LLMs. ** Recommendation Gemini 3.1 Pro delivers reliable performance for our use cases. We should maintain support for proven, earlier model versions rather than automatically adopting the latest releases. The new generation's added sophistication doesn't necessarily translate to better results for our needs. I'm happy to share detailed notes on this evaluation. Please reach out if you'd like to discuss.

8

206 Innovation Inc. Announces Curvature Variable Physics (CVP): A Breakthrough Framework Linking Energy, Cosmology, and High‑Assurance Computing Bellevue, Washington — January 2026 206 Innovation Inc., founded by Timothy J. Dillon (Inventor), today announced the public release of Curvature Variable Physics (CVP), a new scientific framework that treats spacetime curvature as an active, measurable, and potentially controllable variable. CVP introduces the Dillon operator, a curvature‑sensitivity construct that enables refractance‑based effects across energy systems, cosmology, and advanced computation. CVP proposes that the speed of light may respond subtly to curvature through a relationship defined as D(c,R) = ∂c/∂R, a formulation that produces testable predictions and measurable consequences across multiple domains. A New Approach to Energy and Cosmology CVP’s curvature‑conditioned propagation mechanism suggests that natural curvature oscillations—such as tides, seismic activity, volcanic cycles, and atmospheric storms—may induce refractance effects capable of powering non‑mechanical energy‑harvesting systems. Early simulations indicate the potential for 300–400% uplift over mechanical baselines under declared assumptions. In cosmology, CVP offers a refractance‑based alternative to inflation and dark‑energy narratives. Simulation‑led results show improved fits to DESI BAO data under frozen priors, with a reported four‑fold improvement in a key goodness‑of‑fit metric. These claims are explicitly conditioned on reproducibility and require independent validation. Astrophysics and Observational Tests CVP predicts extended‑structure lensing signatures that differ from point‑mass models. Upcoming datasets from Gaia DR4 and LSST provide the first opportunity to confirm or refute these predictions. The framework includes clear falsifiability criteria, including direct measurement of curvature‑conditioned propagation and population‑level lensing tests. Introducing CVP Spin Overlay for NVIDIA GPU Clusters Alongside the physics framework, 206 Innovation is releasing CVP Spin Overlay, a GPU‑resident integrity layer designed for high‑assurance simulation and AI workloads. The overlay introduces: • curvature‑dependent integrity checks • stale‑state suppression across multi‑GPU pipelines • deterministic replay and proof bundles for audit and reproducibility The system is designed to complement existing RAS/ECC mechanisms and address silent corruption, nondeterministic drift, and reproducibility gaps in large‑scale computing. A Framework Built for Replication and Independent Verification #CurvatureVariablePhysics #CVP #DillonOperator #VariableSpeedOfLight #Cosmology #DESI #Euclid #GaiaDR4 #LSST #Astrophysics #EnergyInnovation #CleanEnergy #TidalEnergy #ReproducibleResearch #ScientificComputing #HPC #GPUComputing #NVIDIA #DeterministicComputing #AIInfrastructure #NationalSecurity #DoD #DOE #NSF #DeepTech

1

Trust in government: 37% (down from 75% in 1958) Trust in corporations: Even lower Solution? Don't trust - VERIFY. Elevate Health uses Web3 infrastructure: 🔐 Cryptographic data control (YOU own the keys) 🔍 Radical market transparency 💰 7-day payment settlement (no billing nightmares) ⛓️ Immutable audit trails 🏛️ Decentralized governance This is Web3 solving real health problems, not hype.

2