AI HW: AMD Ryzen AI Max+ 395 vs Nvidia DGX Spark (GB10)

it does not seem to be really available for $1500, but for $3400...

AMD’s CEO Destroyed NVIDIA's Most Expensive Supercomputers With a $1,500 Lunch-Box PC! - YouTube

• AMD Ryzen AI Max+ 395 (Strix Halo): The dedicated XDNA 2 NPU on this chip is rated for approximately 50 TOPS (INT8). When accounting for the CPU and the integrated Radeon 8060S GPU, the total platform capability is often cited as reaching up to 126 TOPS.
• Nvidia DGX Spark (GB10): The "1000" figure you mentioned refers to the device's marketing performance at FP4 precision (often cited as up to 1 PetaFLOP of compute). Comparing the two is complex because they operate on different architectures and precision formats.

Amazon.com: GMKtec AI Mini PC Ryzen Al Max+ 395 (up to 5.1GHz) 2TB PCIe 4.0 SSD 64GB LPDDR5X 8000MHz (8GB*8) Quad Screen 8K Display/WiFi 7/ USB4/ SD Card Reader 4.0 EVO-X2 : Electronics

 AMD Destroyed Nvidia’s $4,000 AI Box! 😱 - YouTube

The video is explaining the Ryzen AI Max+ 395 (codenamed "Strix Halo") APU from AMD, showcased inside a compact, lunchbox-sized mini PC (such as the GMKtec EVO-X2).
The Technology
• The Processor: The system is powered by AMD's flagship Ryzen AI Max+ 395 processor, built with 16 Zen 5 CPU cores and a massive Radeon 8060S integrated graphics chip (RDNA 3.5).   

• The Unified Memory Trick: Unlike traditional PC setups where the graphics card (VRAM) and system memory (RAM) are completely separate, this APU uses up to 128GB of high-speed unified memory. On Linux, it can allocate up to 110GB of that memory directly to the GPU.   
The Nvidia "Killer" Context
• The VRAM Problem: Large Language Models (like the 235-billion parameter DeepSeek-R1) require massive amounts of video memory (VRAM) to run locally. To do this on Nvidia hardware, you would typically need multiple expensive graphics cards or dedicated AI enterprise hardware (like the Nvidia DGX Spark) costing $4,000+.   

• The Price Advantage: Because AMD's "lunchbox" mini PC can use its unified system memory as VRAM, it can hold these massive AI models entirely local for a starting price of around $1,499. AMD claims this allows the mini PC to outperform a single Nvidia GeForce RTX 5080 by up to 3x in specific DeepSeek-R1 inference workloads due to Nvidia's strict VRAM limits.
Ryzen AI Max+ 395: Run a 235B Model Locally on a Mini PC 
XDNA 2, over 50 TOPS (about 126 TOPS combining CPU, GPU and NPU)

house design: Single Slope Cottage

a very basic house... no insulation... simple efficient design... but materials are suboptimal.

12x20 Single Slope Cottage | Eagle Ridge Barn Builders (TX)

$12,879
SQFT: 240 => $53/sqft, almost the same as shipping container price... and similar quality :)

18x28 Single Slope Cottage | Eagle Ridge Barn Builders

SQFT: 392.  $20,590 => $52.5/sqft

"cabin" in shipping container: $7.5K

interesting "housing solution" in areas exposed to forest fires.

dual door 20ft shipping container is apparently about $4K,
and with added basic insulation and glass doors comes to about $7.5K;
he did have wood planks for walls, and labor is "free"... not quite realistic price... 

the width of container is 8ft outside, likely about 7ft inside, so total about 140sqft
comes down to about $50/sqft;

there are ways to make much better house for less money per sqft...

still, this is clever, since does not require cutting metal and doors can protect as needed.

Shipping Container Tiny Home Build - YouTube by Wilson Forest Lands - YouTube

Cheap Way to Transform A Shipping Container Into A Cozy Home - YouTube

Cheap Way to Transform A Shipping Container Into A Cozy Home - YouTube

Everyone Overcomplicates Container Homes… Try This Instead - YouTube

AI and Real Estate prices?

Zillow reports shocking housing U-TURN. 40% losses emerge. - YouTube

This video provides an update on the 2026 US housing market, highlighting signs of a cooling market and localized price corrections. Here are the key points discussed:

Market Indicators Show a Downturn:

• Declining Demand: Data from ShowingTime indicates a significant drop in home tours, down by roughly 35% to 40% compared to previous periods (0:37). Additionally, Google Trends data shows reduced consumer searches for homes for sale (0:49).
• Rising Costs of Ownership: Existing homeowners are facing higher monthly payments due to increases in insurance, taxes, and mortgage rates, with costs rising approximately 40% over the last 6 years (3:34 - 3:58).

Significant Price Corrections:

• The video highlights specific instances of six-figure losses for sellers, including examples of homes in Atlanta and Texas selling or listed at 40% to 48% below their purchase prices from a few years ago (1:13 - 2:37).
• Regional Variations: While many markets are experiencing price declines—particularly on the West Coast, parts of the South, Denver, and Seattle—some areas like downtown San Francisco are seeing gains, which the creator attributes to a concentrated boom in AI companies (5:10 - 7:05).

Strategic Advice for Buyers:

• Affordability Focus: The creator notes that buyers are increasingly budget-conscious, with many targeting monthly payments around $1,500 (10:29). Certain states, such as Alabama, Arkansas, and West Virginia, offer more affordable payment options (11:15).
• Due Diligence: The video encourages prospective buyers to conduct thorough research, use listing analysis tools, and offer below-list prices on properties that have been sitting on the market, as sellers may become more desperate through the late summer and fall (8:15 - 13:07).

 

architecture: round house redesign

making round house design "work" is not easy... this is clever and simple, good solution

 I Turned this Awkward Old Cabin into a Calm, Minimalist Retreat - YouTube

Tesla Model Y L: $62K

 Tesla launches Model Y L in US — 6 seats, 325 miles, $61,990 | Electrek

Tesla Semi for price of Model Y?

 SpaceX Is About to Take Over The Internet - YouTube

The video explains that while the estimated price of a Tesla Semi is approximately $290,000, qualifying businesses in California can significantly reduce this cost through two stackable incentive programs (11:10 - 11:56).

• Clean Truck Voucher Program (HVIP): This program provides up to $120,000 per truck (11:30 - 11:35).
• California Clean Fuel Reward: This newer incentive offers an additional $120,000 (11:39 - 11:43).

By stacking these incentives, eligible small fleets can reduce the purchase price of a Semi by as much as $240,000, bringing the cost down to roughly $50,000 (11:45 - 11:58). The speaker emphasizes that this helps remove a major barrier to adoption—the upfront purchase price—while offering lower operating costs over the life of the vehicle compared to diesel trucks (12:34 - 13:17).

free training: Agentic AI Foundations by Oracle

 Oracle Agentic AI Foundations: Get skilled for the Agentic AI Era | oracleuniversity

blogs.oracle.com/oracleuniversity/wp-content/uploads/sites/118/2026/06/Oracle_Agentic_AI_Foundations_Full_Tour.mp4

By the end of the course, you will be able to:

• Understand core AI agent concepts.
• Design AI agents using LangChain and the OpenAI Agent stack.
• Implement Model Context Protocol (MCP) concepts.
• Build agents using the OCI Enterprise AI platform.
• Apply Oracle AI Database capabilities for agentic AI.

The course is organized into six modules that build on each other from first principles to enterprise deployment.

Module 1: Introduction to AI Agents

The mental model for everything that follows: an LLM-based agent is an LLM plus tools plus a loop. We cover what makes an agent goal-directed, autonomous, tool-using, and iterative; the core reasoning patterns (Chain-of-Thought and ReAct); a walkthrough of your first agent; and a layered, defense-in-depth approach to safety and guardrails.

Module 2: LangChain for AI Agents

This module introduces LangChain and the LangChain Expression Language (LCEL). You’ll build your first agent, then go under the hood to see what a single agent.invoke() call is really doing: building tool schemas, parsing tool calls, executing functions, and deciding whether another model call is needed. That’s the understanding you need to debug agents and move them to production.

Module 3: Introduction to MCP

The Model Context Protocol (MCP) gives agents a standard way to connect to tools, data, and prompts. We cover the MCP architecture and core components, then add an MCP server to your agent starting with a simple local math server and moving to a real-world OCI Usage MCP server. The takeaway: MCP decouples agents from tool implementations, enabling interoperability, discovery, and reuse at scale.

Module 4: OpenAI Responses API and Agents SDK

This module covers the OpenAI Agent stack and how to choose between its pieces – the Responses API for simpler, single-call use, and the Agents SDK for multi-step logic, multiple agents, guardrails, and tracing. We cover tools and function calling, multi-agent systems and handoffs, and safety, then put it together in a multi-agent customer-support system that routes requests to specialized agents.

Module 5: Agentic AI for Enterprises

Building an agent is one thing; running it reliably is another. This module introduces the OCI Enterprise AI platform and OCI Enterprise AI Agents, and the division of labor: you focus on the agent’s instructions, tools, knowledge bases, and outcome, while OCI handles hosted endpoints, scaling, memory and sessions, sandboxed tools, logging, and integrations. We finish with a discussion on how to build agents using OCI Enterprise AI Platform.

Module 6: Agentic AI for Oracle AI Database

This module focuses on bringing agents to your data. We cover Oracle AI Vector Search and its workflow, the Oracle AI Database Private Agent Factory, the Select AI Agent for building agents that live inside the database, and the Oracle Autonomous AI Database MCP Server – showing how agentic capabilities can run close to your data, governed by the security you already rely on.

AI in space: Planetary Intelligence

 Planetary Intelligence: AI Leaves the Internet, Moves Into Space, Earth Gets a Nervous System | #266 - YouTube

This episode of Moonshots explores the convergence of AI, space infrastructure, and planetary intelligence. Here are the key points discussed:

• Large Earth Models (LEM): Will Marshall (CEO of Planet) explains how Planet is transitioning from a satellite imagery company to a provider of "Large Earth Models." By leveraging 150 petabytes of historical data, they aim to make the Earth searchable like the internet (05:30 - 07:50).
• The Power of Real-World Data: Unlike LLMs trained on theory and text, LEMs are being trained on actual sensor data to provide predictive analysis for agriculture, disaster response, and national security (16:06 - 18:09).
• Compute in Space: A major theme is the potential to move compute power into orbit. By processing data at the edge (on the satellites), companies can reduce response times for critical events like wildfires from hours to minutes (36:00 - 36:42).
• Efficiency and Infrastructure: The discussion highlights that while launch costs are dropping, the true revolution is in satellite performance density and energy-efficient chips (like Google’s TPUs). The ability to perform inference efficiently is identified as a critical competitive advantage in the AI-space race (50:24 - 52:32, 118:51 - 120:30).
• Openweight vs. Closed AI: The panel discusses the rise of Chinese models like GLM 5.2, which demonstrates near-competitive performance to Western frontier models at half the cost, signaling that frontier intelligence cannot be easily monopolized (201:50 - 208:35).
• The Great Filter & Existential Risk: The guests touch on the Fermi Paradox, suggesting that the "Great Filter" could be the stage where a civilization builds technology faster than its social systems can manage, highlighting the need for thoughtful AI alignment (211:02 - 212:13).

house construction: MgO boards

 Magnesium oxide wallboard - Wikipedia

Magnesium oxide, more commonly called magnesia, is a mineral that when used as part of a cement mixture and cast into thin cement panels under proper curing procedures and practices can be used in residential and commercial building construction. Some versions are suitable for general building uses and for applications that require fire resistance, mold and mildew control, as well as sound-control applications. Magnesia board has strength and resistance due to very strong bonds between magnesium and oxygen atoms that form magnesium oxide crystals (with the chemical formula MgO).

Magnesia boards are used in place of traditional gypsum drywall as wall and ceiling covering material and sheathing.

What Materials Are Used in Boxabl Homes

The Ultimate Guide to MgO Structural Insulated Panels: Building Smarter with Innovation

Magnesium Oxide Boards | MGO Panels | Magpanel

The Future of Building with MGO SIPS - YouTube

This video introduces MGO (Magnesium Oxide) SIP panels from SRD Supreme as a high-performance, sustainable solution for modern construction. Key benefits highlighted include:

• Durability & Health: The panels are structurally strong, mold-resistant, rot-resistant, and free of off-gassing (0:05-0:15).
• Safety: They offer superior fire resistance compared to traditional building materials (0:17-0:24).
• Environmental Impact: Designed for sustainability, the panels help achieve a lower carbon footprint, generate less on-site waste, and last for decades, which minimizes the need for future rebuilds (0:26-0:39).

Overall, SRD Supreme promotes a shift toward stronger, safer, and greener building practices (0:40-0:46).

AI model: Claude Sonnet 5

 Introducing Claude Sonnet 5 \ Anthropic

Claude Sonnet 5 is built to be the most agentic Sonnet model yet. It can make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models.

What's new in Claude Sonnet 5 - Claude Platform Docs

Anthropic launches Claude Sonnet 5 as a cheaper way to run agents | TechCrunch

Why Claude Fable 5 Is Back: U.S. Lifts Controls on Anthropic's AI Models | HowStuffWorks

AI: Fine Tuning LLMs with InstructLab

Fine Tuning Large Language Models with InstructLab - YouTube

• The Purpose: Fine-tuning allows developers to customize and specialize general LLMs for specific tasks, automate repetitive work, and handle complex, domain-specific problems.

• The Tool: InstructLab provides an open-source community-driven approach to model alignment, making it easier to add new knowledge and skills to a base model.

• The Workflow: The video demonstrates the step-by-step process of setting up InstructLab, generating synthetic data, and training the model to improve its performance on targeted queries.

You can learn more about the technology or explore the guide mentioned in the video by visiting the IBM InstructLab Overview or downloading the IBM AI Model Guide.

A new way to collaboratively customize LLMs - IBM Research

InstructLab Summary

InstructLab is an open-source project launched by IBM and Red Hat designed to lower the cost and barrier to entry for customizing large language models (LLMs). Instead of retraining a model from scratch, it allows a community-driven, collaborative approach to adding new knowledge and skills to base models.

Key Features

• Synthetic Data Generation: Uses the LAB (Large-Scale Alignment for ChatBots) method to amplify small amounts of human-curated "seed" data into high-quality training data.

• No Overwriting: Its phased-training regimen allows models to assimilate new skills without losing or overwriting previously learned information.

• Open-Source Workflow: Users can test out quantized models locally on a laptop using a command-line interface (CLI) and submit new skills or knowledge via standard GitHub pull requests.

---

Project Repository

You can contribute to the community and view the project taxonomy directly on the InstructLab GitHub Organization.

py + ts

Securing AI Business Models

Based on the video How to Secure AI Business Models by IBM Technology, here is a quick summary and the key points of the presentation:

Summary

The video focuses on how organizations can safely adopt and secure generative AI technologies within their business models. The presenter introduces a Security for Generative AI Framework designed to balance technological advancement with risk mitigation, focusing on core pillars like trust, privacy, and accuracy.

---

Key Points

• The Dual Relationship (AI for CS vs. CS for AI): The presentation highlights the intersection of using AI to augment cybersecurity defenses while simultaneously needing specialized cybersecurity measures to protect AI models from unique vulnerabilities.

• Core Pillars of AI Security: To secure business models utilizing AI, organizations must actively protect four main areas:

  • Trust: Ensuring the outputs are reliable and the system operates as intended.

  • Privacy: Safeguarding sensitive training data and user inputs from leaking.

  • Accuracy: Defending against data poisoning or manipulation that could skew AI decisions.

  • Cybersecurity Posture: Implementing standard defenses to protect the underlying AI infrastructure.

• Introduction to MLDR: The video introduces concepts like Machine Learning Detection and Response (MLDR) to actively monitor AI pipelines for anomalies, adversarial attacks, and prompt injection attempts.

• Securing the AI Lifecycle: Protection must be integrated across the entire pipeline—from securing the initial training datasets and the model architecture to monitoring live application outputs in real time.

Artificial Intelligence (AI) Cybersecurity Solutions | IBM

The power of AI: Security | IBM