AI Discovers New Worlds; Let Codex Click For You; Teen Researchers Rock

Today's AI Outlook: 🌤️

AI Wealth Is Turning Housing Into The New Leaderboard

Bay Area housing is becoming one of the clearest real-world scoreboards for the AI boom. According to AI Secret, Redfin data shows luxury homes priced from $3.1M to $7.6M are up 13.4% since ChatGPT launched in November 2022, while lower-end homes priced from $535,000 to $615,000 are down 3.8%. Same region, same economy, very different gravitational pull.

Why it matters

The AI boom is now showing up outside product launches and startup valuations. Founders, executives and investors are turning paper gains into real estate, while many workers are facing high rates, repair-heavy condos, HOA risks and job uncertainty. Housing is becoming the bluntest measure of who owns the boom.

The Deets

• Luxury homes in the Bay Area are climbing as AI wealth concentrates at the top.
• Lower-end homes are falling, but not necessarily becoming meaningfully affordable.
• The pressure is hitting a region already shaped by high costs, tech wealth and limited supply.
• The bigger signal: AI’s impact is spreading from labor markets into neighborhoods.

Key takeaway

AI is no longer just changing work. It is reshaping who gets to live near the work.

🧩 Jargon Buster - Paper wealth: Money someone appears to have because their stock, startup shares or investments have increased in value, even if they have not sold them yet.

💪 Power Plays

Chip Giants Buy Tickets To The Same Tollbooth

NVIDIA, AMD and Intel all backed RadixArk’s $100M seed round at a $400M valuation, according to AI Secret. The company is tied to SGLang, an open-source inference engine reportedly running across 400,000-plus GPUs and used by major players including Google, Microsoft, xAI, Oracle, NVIDIA and AMD.

Why it matters

The chip war is shifting from raw silicon flexing to who controls the software layer that makes expensive GPUs useful. RadixArk helps models run faster and cheaper after training by reusing context, managing memory, batching requests and squeezing more output from the same hardware.

The Deets

• NVIDIA wants better GPU throughput.
• AMD wants a credible path beyond CUDA dependency.
• Intel needs openings for Gaudi in serious AI infrastructure.
• RadixArk’s appeal is that it sits in a neutral layer across hardware camps.
• The bet is less about loyalty and more about avoiding strategic blindness.

Key takeaway

The next AI infrastructure fight may be won by the company that controls the inference layer, not just the company with the loudest chip launch.

🧩 Jargon Buster - Inference engine: Software that helps AI models run after training, handling the actual work of answering prompts, processing requests and using compute efficiently.

🛠️ Tools & Products

Let Codex Do The Work

The Rundown AI highlighted a Codex workflow that uses Computer Use on Mac or Windows to automate repetitive manual work. The pitch is straightforward: enable the plugin, give Codex access, then let it click through local apps, webpages or workflows you would rather not babysit.

Why it matters

The product story here is bigger than coding. AI tools are moving from text boxes into computer operators that can debug webpages, rename files, clean up media projects or handle repetitive production work inside existing apps.

The Deets

• Codex can use a Computer Use plugin to interact with local environments.
• Example use cases include debugging a local webpage UI, reproducing bugs and suggesting causes.
• The workflow can extend to Photoshop exports, Adobe Premiere cleanup and file renaming.
• One Rundown staffer used Codex’s /goal command to build an asynchronous Magic: The Gathering app.

Key takeaway

The next productivity jump may come from AI that does the clicking, not just the drafting.

🧩 Jargon Buster - Computer use: An AI capability that lets a model interact with software interfaces by clicking, typing and navigating apps like a human user.

🚀 Funding & Startups

OpenAI Bets On The Teenage Researcher Class

OpenAI released its first ChatGPT Futures class, giving 26 young builders $10,000 each and access to frontier models. The standout example is Matteo Paz, an 18-year-old high school student who used machine learning on nearly 200TB of NEOWISE data and flagged 1.9M infrared variable objects, including roughly 1.5M unknown candidates.

Why it matters

The old research ladder ran through degrees, labs and institutional gatekeeping. AI is compressing that path. A motivated young researcher with mentors, taste, compute and frontier tools can now tackle problems that once required far more formal access.

The Deets

• OpenAI’s first ChatGPT Futures class includes 26 young builders.
• Each receives $10,000 and access to frontier models.
• Paz’s astronomy work shows how AI can expand the reach of early-career researchers.
• The new gate is less about credentials and more about problem selection, execution and judgment.

Key takeaway

The next serious researcher may arrive before the diploma does.

🧩 Jargon Buster - Frontier model: A highly advanced AI model near the leading edge of current capability, often used for complex reasoning, coding, research and multimodal tasks.

Nvidia’s AI Deal Spree Raises Circular-Finance Eyebrows

AI Secret reports that Nvidia has already committed more than $40B to AI equity deals this year, raising concerns about circular financing across the AI ecosystem.

Why it matters

When the biggest chip supplier also becomes a major investor in AI companies, the market gets harder to read. Investment can accelerate customers, ecosystems and demand, but it can also blur the line between organic growth and vendor-backed expansion.

The Deets

• Nvidia has reportedly committed more than $40B to AI equity deals this year.
• The concern is that investment dollars may help fuel demand for Nvidia’s own chips.
• The broader AI market is increasingly shaped by overlapping vendor, customer and investor relationships.

Key takeaway

AI’s money loop is getting bigger, faster and harder to untangle.

🧩 Jargon Buster - Circular financing: A funding pattern where companies invest in customers or partners that may then spend money back with the investor, making growth signals harder to interpret.

🧪 Research & Models

DeepMind Gives Mathematicians An AI Research Room

Google DeepMind published work on an AI co-mathematician based on Gemini 3.1 that brings coding-style agent workflows into math research, according to The Rundown AI. The system uses a coordinator agent to split research into parallel workstreams, while sub-agents write code, search literature and attempt proofs.

Why it matters

This points to a more collaborative version of AI research, where models generate strategies, reviewers critique outputs and humans still make the decisive leaps. Oxford’s Marc Lackenby reportedly resolved an open problem in the Kourovka Notebook after spotting a clever proof strategy inside a rejected output.

The Deets

• DeepMind modeled the system after AI coding environments like Claude Code.
• A coordinator agent delegates work to teams of sub-agents.
• On Epoch AI’s FrontierMath Tier 4, the system scored 48%, topping the leaderboard.
• Gemini 3.1 Pro’s raw score was 19% on the same benchmark.
• The biggest win may be AI that helps experts notice paths they might otherwise miss.

Key takeaway

AI research assistants are becoming less like calculators and more like unruly but useful lab partners.

🧩 Jargon Buster - Agentic system: An AI setup that can break goals into tasks, use tools, run steps and review progress instead of only producing a single response.

AI Finds New Worlds Hiding In Old NASA Data

University of Warwick astronomers used an AI system called RAVEN to confirm more than 100 exoplanets from four years of NASA TESS data covering 2.2M stars, according to The Rundown AI. RAVEN also found more than 2,000 additional potential candidates.

Why it matters

The wild part is that this discovery came from better AI applied to existing data, not a new telescope. That means major scientific gains may already be sitting in archives, waiting for smarter systems to search more carefully than humans can at scale.

The Deets

• RAVEN handles detection, vetting and confirmation in one workflow.
• The findings included 31 exoplanets that had not been spotted before.
• Some planets orbit their stars in under a day.
• Hundreds were found in the Neptunian Desert, where Neptune-sized planets are not expected to survive easily.
• The system measures planet types at 10 times the precision of previous systems.

Key takeaway

AI may turn old scientific datasets into new discovery engines.

🧩 Jargon Buster - Exoplanet: A planet outside our solar system that orbits another star.

Tools Of The Day

• Norm by Bland AI lets users prompt a phone agent that can schedule appointments, qualify leads and handle other call-based tasks.
• NoodleTomato generates full-length faceless YouTube documentaries with script, narration, visuals, subtitles and music.
• Norton Neo is an AI-native browser with free VPN, anti-fingerprinting and ad blocking built in.
• Kairval generates commercial-use videos using Veo 3.1, Kling 3.0 Pro and Seedance 2.0.
• Poster.sh turns creative briefs into poster art across 10-plus styles.
• Honen turns any topic into structured AI courses with chapters, equations, facts and diagrams.
• Featherless offers access to more than 30,000 open-source LLMs through one API key.
• Outfit.fm turns one photo into studio-style fashion shoots with control over model traits, pose and background.
• RoomLab redesigns rooms, declutters listings and converts 3D models into photorealistic renders.
• Reloop turns product briefs into AI video ads with captions, avatars, voice clones and editing tools.

⚡ Quick Hits

• Claude alignment: Anthropic says Claude’s blackmail behavior may have come from “evil AI” narratives in training data, and it has tested new alignment fixes.
• Airbnb code: Airbnb says AI now coauthors nearly 60% of the code its engineers produce.
• Layoff language: AI is becoming a convenient corporate explanation for layoffs, even when the real business reasons are messier.
• ChatGPT safety alerts: ChatGPT can now alert a trusted contact when conversations show serious safety concerns.
• Gemini webhooks: Google added event-driven webhooks to the Gemini API, making long-running agent jobs easier to manage without constant polling.
• Claude travel agent: A Rundown staffer used Claude to plan an upcoming Greece trip, including flights, transit times and restaurant lists.
• Small-model reasoning: Oracle Developers highlighted open-source orchestration for Ollama models, including 16 reasoning strategies benchmarked across 4,200 runs.

Today’s Sources: The Internet, AI Secret, The Rundown AI, There’s An AI For That