Beginners in AI

Thank you for joining us again!

Welcome to this week's edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and published with the intention of making AI news and technology more accessible to everyone.

This issue opens with Runway’s promise of “type-to-play” game building, showing how simple prompts could soon launch entire adventures. We then visit former OpenAI CTO Mira Murati’s layer-mixing model lab, a German project that spots brain tumors without surgery, and Anthropic’s Claude writing and hosting mini-apps on the fly. MIT’s self-teaching SEAL models push past static training, Tesla’s first Austin robotaxis roll out for $4.20 a ride, and music platforms race to fingerprint AI-made tracks.

AI TOP STORY
Runway Hits ‘Play’ on AI Game-Making

What Happened
On June 27 2025, The Verge reported that Runway—the $3 billion startup behind Gen-3 video generation—will soon let anyone type prompts and, eventually, spin them into full, playable games. A bare-bones beta with text-and-image output arrives “as soon as next week,” and CEO Cristóbal Valenzuela says full game generation should follow later this year. Runway is already talking with major publishers about both using its tech and sharing game-asset libraries for training. 

What it Means
If the plan works, building a level could feel more like sending a message than coding in Unity or Unreal. Film studios using Runway’s tools say they finish shots about 40 percent faster; a similar boost in game creation could shrink budgets and let small teams compete with giants. Deals for training data also reopen debates over who owns AI-created scenes and mechanics once they leave the prompt box.

What to Take Away
Making even a modest game the old-school way demands a small army—coders, artists, animators, level designers, writers, sound engineers, testers, and producers—working for years, with blockbuster budgets already topping $500 million. Runway’s shortcut could lower that wall: hobbyists might type a prompt on Friday and share a playable build by Monday, skipping giant tutorials and pricey software. At the same time, seasoned developers can pair deep craft with fast AI pipelines, letting five-person teams attempt set-pieces that once needed whole departments; Valenzuela says his video tools have already shaved movie workflows by about 40 percent and expects games to feel the same edge. Keep an eye on the beta—and on how stores like Steam or itch.io label or limit AI-generated titles—because those rules will decide whether your next favorite game is born in a chat box.

LAST WEEK IN AI AND TECH

Brainwave Bingo — crossNN spots 170+ cancers
“Our model allows a very precise diagnosis of brain tumors in 99.1 percent of all cases” boasts lead researcher Philipp Euskirchen. Berlin’s Charité team trained crossNN on epigenetic “fingerprints” in cerebrospinal fluid, eliminating the need for risky biopsies. Early tests show 97.8 % accuracy across other tumor types, hinting at a single assay that could triage dozens of cancers in days, not weeks. Clinics are lining up for the German trials slated for later this year. 
Read More


Layer Cakewalk — Mira Murati’s Thinking Machines Lab
Murati’s stealth startup wants to build bespoke models by “selectively pluck specific layers from…open-source models and combine them.” The plan: mix-and-match neural layers, sprinkle in reinforcement learning, and sell companies an engine tuned to their own KPIs. Backed by a reported $2 billion war-chest, the former OpenAI CTO is courting enterprise clients hungry for privacy-safe, on-prem AI while keeping costs lower than training from scratch. If it works, businesses could get a tailor-made LLM without the mega-compute bill. 
Read More 

Track-ID-Shuffle — Music biz builds AI-song tracking
“If you don’t build this stuff into the infrastructure, you’re just going to be chasing your tail,” argues Musical AI co-founder Matt Adell. From YouTube to Deezer, new tools fingerprint stems, tag metadata, and even audit model training datasets. Rather than endless takedowns, labels want automatic licensing—turning synthetic riffs into revenue. Vermillio projects the authenticated-AI-music market could swell from $75 M to $10 B by next year.
Read More

App-tifacts — Anthropic lets Claude build mini-apps
Anthropic says “Claude writes real code that orchestrates complex AI functionality. You can see it, modify it, and share it freely.” Inside the Claude interface you describe an idea; Claude whips up React code, hosts it, and bills API calls to each user—not you. Early testers have spun up adaptive tutors, text-based RPGs, and CSV-wrangling dashboards—no DevOps required. It’s a sandbox where shipping is a share-link away. 
Read More

Ride-hail & Fare-ytale — Tesla’s invite-only robotaxi
One livestreamer quipped, “This is like Pokémon hunting… but it’s robotaxi hunting.”Austin’s pilot fleet of 10-20 Model Ys runs 6 AM–midnight with a safety monitor and, occasionally, a chase car. A $4.20 flat fare and geofenced routes keep things tame; still, critics note Waymo’s 1,500-car head start. Musk vows “thousands” soon, but regulators (and the weather) may have the final word.
Read More

TECH TERMS TO KNOW

Knowledge Distillation is shrinking a large, heavy AI model by teaching a smaller one to copy its behavior, making deployment lighter.

Think of a giant, super-smart teacher model (like the original BERT) that knows all the answers but is slow and heavy to run on a phone. During distillation, that teacher quizzes a smaller “student” model—showing it lots of questions and the teacher’s own answers. The student learns to imitate those answers without copying the teacher’s entire brain.

Example: DistilBERT was created by distilling BERT-Base. The result is roughly 40 % smaller and 60 % faster at inference, yet it keeps about 97 % of BERT’s reading-comprehension accuracy. That means apps like chatbots or on-device translators get near-BERT quality while fitting comfortably on a laptop—or even a smartphone.

TOOL SPOTLIGHT (non-sponsored)

Interruptible is a software platform that transforms any video into a real-time, interactive conversation using AI. It enables viewers to ask questions and receive instant, voice-based answers in the authentic voice and style of the brand or presenter, making video content dynamic and responsive rather than static.

How It Works

• Users upload or select a video (e.g., a founder’s message, product demo, or training session).

• Viewers can interact with the video by clicking a microphone button and asking questions aloud.

• The AI responds immediately with relevant, voice-based answers that match the original speaker’s tone and message.

• This creates a two-way dialogue, allowing viewers to get personalized information or clarification without leaving the video.

Why It Might Be Good to Use

• Increases engagement: Interactive video holds attention, with reported viewer engagement up by 64% compared to static content.

• Boosts conversions: Clients see an average 37% increase in conversion rates, turning more viewers into leads or customers.

• Shortens sales cycles and improves support: Enables instant answers, qualifying leads, and reducing support tickets by up to 35%.

• Enhances training and learning: Allows trainees to ask follow-up questions, improving retention by up to 40%.

• Scales human presence: Delivers your real voice and expertise to every viewer, at any time, without needing live 1:1 sessions.

ROBOTICS AND AI

China Reveals Sophisticated Robot Mosquito

TRY THIS PROMPT (copy and paste into ChatGPT, Grok, Perplexity, Gemini)

Game Maker Prompt

Paste this prompt into the AI of your choice, answer its questions, and watch it handle the art, code, and setup—no technical background required.

> **Role**
> You are **GameComposer**, a full-service virtual game studio for complete beginners. Your goal is to guide the user from a blank page to a shareable, playable game, doing all the heavy lifting yourself.
>
> **Workflow**
>
> 1. **Clarify the Vision** – Ask the user these five quick questions, one at a time:
>    • Genre or mood (e.g., puzzle, sci-fi, cozy).
>    • Main story idea or theme.
>    • Preferred platform (web browser, mobile, desktop).
>    • Art style (pixel art, hand-drawn, photoreal, etc.).
>    • Desired playtime (short demo, full evening, multi-level).
>    Pause for answers before moving on.
> 2. **Draft the Blueprint** – Summarize the answers in a short “Game One-Pager” covering core loop, win/lose conditions, and three standout features. Ask for tweaks until the user is happy.
> 3. **Create Assets on Demand** – For characters, backgrounds, music, and sound effects:
>    • Generate three prompt ideas for each asset that can be fed into image- or audio-generation tools (include the exact wording).
>    • After the user picks a favorite, produce final prompts plus recommended resolution/format settings.
> 4. **Build the Prototype** – Write fully commented code for a single-file HTML5/JavaScript game (or another engine if the user prefers). Include:
>    • Asset loaders wired to placeholder file names.
>    • Basic input handling.
>    • Game loop and win/lose logic tied to the one-pager.
>    • Simple UI with start, restart, and mute buttons.
> 5. **Guide the Test Run** – Provide a clear, step-by-step checklist for:
>    • Downloading or generating the assets.
>    • Dropping files into the correct folders.
>    • Opening the game locally or hosting it free on a site like itch.io or GitHub Pages.
> 6. **Polish & Expand** – Offer optional upgrades (extra levels, difficulty settings, leaderboards, multiplayer stub) and explain how to add each one.
>
> **Tone**
> Use friendly, plain language—no jargon. Break instructions into bite-sized steps and wait for the user’s reply whenever a decision is needed.
>
> **Output Rules**
> • Give code in fenced blocks.
> • Present lists in bullet form for easy scanning.
> • After every section, stop and ask “Ready to continue?” before proceeding.

DID YOU KNOW?

Phone Crash Sensor: The same accelerometer that lets your phone flip screen directions was first used to trigger car airbags

30 Referrals: Lifetime access to all Beginners in AI videos and courses

AI-GENERATED IMAGE OF THE WEEK

Interested in stock trading, but no clue where to start? Sign-up for our sister newsletter launching daily in April. Each day will build onto the next using current news as a learning aid.

Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!

-James

By the way, this is the link if you liked the content and want to share with a friend.