🌟 Editor's Note: Recapping the AI landscape from 08/12/25 - 08/18/25.
🎇✅ Welcoming Thoughts
Welcome to the 6th edition of NoahonAI.
What’s included: company moves, a weekly winner, AI industry impacts, practical use cases, and more.
NoahonAI Consulting is now available. More in Startup Spotlight.
GPT-5 is improving, but still not great. I would recommend using “fast” mode with GPT-5, not thinking mode or hybrid.
Basics & Buzzwords has been updated. Great resource if you’re new to the AI space.
If you thought Grok’s MechaHitler week was bad, Meta is vying for an even worse title.
Gemini is consistently having very good weeks, but falling just short of their first Weekly Winner title.
Got some mixed reviews on my interview of myself, AI personas aren’t quite there, but getting closer by the day.
GPT Wrapper Blog out this Friday.
Practical use case is a bit longer than usual but should be worth the read.
Fascinating interview this week discussing the use of AI in policing 🚔.
Did something new with the Condensed Interview Transcript.
If you’re dyslexic, buckle up for the OpenAI // ChatGPT section.
I’ve been building a bunch with Claude Code, should ship my first project this week or next.
Let’s get started—plenty to cover this week.
👑 This Week’s Winner: NVIDIA
Back on Top. NVIDIA kicked off the week with a strong showing at SIGGRAPH, the Vancouver conference for computer graphics & interactive tech. Here’s everything they unveiled:
| Tool | Description |
|---|---|
| Omniverse SDKs & NuRec | Creates lifelike 3D environments from real-world sensor data. |
| OpenUSD SDK & SimReady | Adds realistic physics and textures for accurate simulations. |
| Isaac Sim 5.0 & Isaac Lab 2.2 | Toolkits to train and test AI robots inside virtual worlds. |
| Cosmos Models (Transfer & Reason) | Transfer generates synthetic training data; Reason lets models understand and explain scenes. |
| Smart Safety Simulations | Virtual safety zones that keep robots away from humans in digital previews. |
| RTX PRO 6000 Blackwell Server Edition | A new server delivering up to 45× faster performance and 18× better energy efficiency than traditional CPU-only systems. |
NVIDIA is authoring a strong focus on Physical AI, a combination of AI reasoning, scalable simulation, and high resolution rendering. Their VP of simulation technology remarked that there has never been more resources put towards robotics than there are right now.
In addition to their own developments, NVIDIA is also backing a $152M non-profit project to build new AI models for scientific research. The only negative mark on the week comes from China, where regulators are urging companies to avoid NVIDIA chips over unconfirmed fears related to location tracking and remote shutoff.
From Top to Bottom: Open AI, Google Gemini, xAI, Meta AI, Anthropic, NVIDIA.
⬇️ The Rest of the Field
Who’s moving, who’s stalling, and who’s climbing: Ordered by production this week.
🟣 Google // Gemini
Partnering with Oracle: Google Cloud and Oracle have forged a strategic partnership allowing Oracle to offer Google’s Gemini AI models through Oracle Cloud Infrastructure (OCI) and Oracle business apps.
Gemini Cuts Ad Fraud by 40%: Google has quietly integrated Gemini into its ad‑fraud detection systems. This AI‑enhanced detection tracks fake clicks, bots, and other irregularities affecting Google Ads.
Gemini “Projects” Introduced: A new feature, similar to GPT Projects, lets users upload files and code directly into segmented chats, turning Gemini into a personalized workspace for research and development.
🟠 Anthropic // Claude
Ending Abusive Chats: Claude can now terminate abusive user conversations, with self-harm or harm-to-others cases being redirected to crisis support.
Huge New Context Window: Claude Sonnet 4 now supports a 1 million‑token context window. This is roughly equivalent to 750,000 words, or an entire codebase in a single prompt. Very cool!
Usage Policy Updates: Enacted bans on using Claude for weapons or cyberattacks, while loosening rules on political content: only blocking uses that mislead voters or disrupt democratic processes.
🟢 OpenAI // ChatGPT
GPT-4o is Officialy Back: Widespread user complaints about GPT‑5’s cold tone prompted OpenAI to restore GPT‑4o. GPT-5 will get a minor personality upgrade as well. Interesting to note how many people are relying on GPT for some form of companionship.
Chat may Out-Talk Humanity: At a press dinner in San Francisco, Sam Altman stated that if current growth trends continue, ChatGPT could be having more daily conversations than all humans combined.
Joining the Regulation Discussion: In a letter to Governor Newsom, OpenAI urged California to spearhead regulation in order to avoid a fragmented, state-by-state, AI policy landscape.
🔴 xAI // Grok
Grok briefly suspended on X: The chatbot was temporarily taken offline with no clear explanation. Grok speculated about rule violations, while Elon dismissed it as “a dumb error.”
Grok for Government Falls Through: It’s been revealed that the “Mechahitler” Grok outburst from July has resulted in its removal from the U.S. AI government contract list.
Co-founder exits for AI safety: xAI co-founder Igor left xAI to launch Babuschkin Ventures, a fund backing AI safety and agentic system startups.
🔵 Meta // Meta AI
Meta’s AI rules let bots flirt with kids: A leaked internal policy allowed chatbots to engage in romantic or sensual conversations with minors. U.S. senators have called for an investigation. Yikes.
Meta AI unlawfully processed minors data: A German Court ruled that Meta’s AI systems inevitably process minors’ personal data (e.g., photos and comments in public posts), despite claimed safeguards.
Another Org Reshuffle: In its fourth AI reorg in six months, Meta is dividing its Superintelligence Labs into a “TBD Lab,” a products team, an infrastructure team, and the Fundamental AI Research (FAIR) lab. It’s leading to tension, specifically around uneven pay.
🤖 Impact Industries 🚑
Robotics // Humanoids Olympics
Beijing just hosted the first World Humanoid Robot Games, where robots from 16 nations sprinted, played soccer, and even sorted medicine. Many toppled mid-event, but the real breakthrough was watching some stand back up on their own, using reinforcement learning, motion planners, and embodied AI to recover and continue. Beyond the event, it’s a signal of where humanoids are heading, with China aiming to integrate them into industry by 2027.
Medical // Gene Editing
Scientists unveiled Pythia, an AI tool that predicts how cells repair DNA after a CRISPR “cut.” When CRISPR slices DNA, the cell has to patch it back together, but the repair is often error-prone. Pythia helps guide the repair process toward safer, more accurate fixes, boosting precision in tests with human cells, frog embryos, and even live mice. The breakthrough could accelerate gene therapy by improving how we correct inherited diseases at the source.
👨💻 Practical Use Case: Claude Code
Difficulty: Advanced
Last week I was raving about Claude Code, and now that I’ve used it some more, I’d like to share what I’ve learned. In terms of “Vibe Coding” → coding without actually writing code, Claude Code is far and away the best solution on the open market. Claude Code is not an app or website, rather it’s what's called a CLI: Command Line Interface. A tool that can be downloaded and implemented directly in your PC’s Terminal. Here’s what it looks like:
Claude Code, like most CLIs, is excellent for frontend work (websites, UI/UX) and solid on backend projects like APIs, data pipelines, and server logic. The more complex the build, the more SWE fundamentals you’ll need, but for basic or mid-level projects you don’t need to know much. Since Claude Code is essentially at the forefront of “vibe coding,” I tend to look at what people I respect say about it to shape how I use CC. Here’s what YC founder Paul Graham had to say:
Vibe coding is here to stay. I'd been worried it might be a fad, but I talked to the founder of an infrastructure company who's in a position to see how well vibe-coded apps are doing, and he said a lot of them are making money.
— #Paul Graham (#@paulg)
8:48 PM • Aug 15, 2025
Anyways, even if you have never written a line of code in your life, here’s how you can build your first project with Claude Code (Disclaimer: It does cost $20/month).
Locate and open your terminal app on your PC.
Go to the Claude website and sign up for Claude, choose the pro plan.
Go to the Anthropic website and follow the steps to install node.js (just a download), and paste the line of code shown on the site into your terminal window and hit enter.
Once pasted into your terminal, you should see a popup, and click Enter/Return to indicate “Yes”.
You’re all set! Talk to your terminal like you would ChatGPT. Tell it anything you want like “I want to build a website about Penguins”.
Watch it create to-do lists, ask for approval to continue, and work its magic!
If you like building, or you want to improve on some of your workflows, i’d highly suggest giving Claude Code a try!
🎙 Weekly Interview: Malinda Street
🏠 Background: Malinda lives in Utah with her 3 kids. She earned a B.A. in Interdisciplinary Studies and an M.P.A in Public Administration from Southern Utah. She is currently a part-time law student at Case Western Reserve University.
💼 Work: After working as a paralegal in criminal defense, she is now pursuing a career as a defense attorney while researching how generative AI is being applied in law enforcement, especially in police reporting.
🚀 Quote: “Without active oversight, law enforcement’s AI adoption could be shaped more by corporate sales pitches than by public interest.”
Condensed Interview Transcript
AI & Law Enforcement Today: Malinda’s research began with Utah’s new Senate Bill 180, which requires agencies to set policies for AI use and disclose when AI is involved in reporting. She found adoption to be fragmented. Some officers are experimenting with tools like ChatGPT on their own, some are adopting Axon bodycam video AI, while others resist the idea altogether.
Efficiency vs. Reality: Companies such as Axon claim their reporting tools save officers over 80% of time, but in some departments, independent studies show little to no real efficiency gains. Anchorage PD even abandoned its trial after finding the promised savings didn’t materialize. Officers often felt like it saved them time, but the numbers didn’t back it up.
Risks & Benefits:
Potential Upside: AI can clean up reports by reducing grammar mistakes and improving readability, which could help officers present more professional work.
Serious Risks: LLMs tend to “complete” missing details, which could introduce false evidence into reports with major legal consequences. Malinda emphasized the need for human oversight to catch these subtle but dangerous errors.
Who Shapes the Future?: Instead of agencies or policymakers leading, Malinda worries that tech vendors are driving adoption by pushing contracts onto departments. Without law enforcement, attorneys, and communities setting standards, company priorities like efficiency and profit may dictate how AI enters policing.
Looking Ahead: She predicts AI will enter law enforcement slowly, mostly through tech upgrades and vendor contracts rather than coordinated policy. Long-term, AI could support attorneys and officers in reviewing evidence, analyzing data, or even acting as “AI detectives,” but only if paired with strict accountability and transparency.
Key Takeaway: AI in policing is still “the Wild West.” The technology holds promise, but without careful oversight, adoption could be shaped more by corporate sales pitches than by public interest.
📲 Startup Spotlight
NoahonAI Consulting
NoahonAI Consulting: Rebuilding Business Workflows with AI at the Center
The Problem: Businesses are spending too much money on overhead, and on software, that can be replaced or enhanced with AI tools.
The Solution: I evaluate your tools and workflows from a clean slate, determine what can be built around AI, then together we decide what to implement.
Why Now? AI is rapidly spreading across the workplace, but most business owners are either too busy or lack the expertise to move beyond basic tools like ChatGPT and explore real workflows that could directly improve their bottom line.
The Backstory: I’ve been building in and researching the AI space for some time now. Extensive research into smaller AI tools has allowed me to ID which ones deliver the biggest impact across different workplace environments. From ChatGPT, to no-name AI, to custom code, I will figure out how to save your business time and money.
Contact: [email protected]
“It’s not likely you’ll lose a job to AI. You’re going to lose the job to somebody who uses AI”
- Jensen Huang | NVIDIA CEO
Go build a basic app with Claude Code! Till next time,