Sponsored by
The Voice Data Behind the World's Best AI
Training data quality determines model quality. Voices delivers custom or pre-built voice datasets—directed by real professional talent, not scraped audio—across any language, accent, or emotional range.
Choose from 800+ hours of character voice performances, 1,723 hours of QA'd conversational data across 56 emotional states, or 1,000 hours of expressive, multi-language data with full emotion tagging. Every dataset ships with clear licensing, signed consent, and usage rights baked in, so your legal team sleeps easy and your model trains on data it can actually use.
7 of the world's top 10 AI labs already use Voices. Request free samples and hear the difference professionally directed voice data makes.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this daily 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 edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
ChatGPT's GPT-6 Astra Finished a Difficult Video Game Alone and Ran Up a $571 Bill
TLDR: OpenAI's GPT-6 Astra played through the entire game Portal with no human at the controls, and the API bill came to $571.18.
The Story:
An enthusiast who goes by CozyBlaze wired GPT-6 Astra into Valve's puzzle game Portal and let it play alone. The model got screenshots and the player's position, worked out what to do, then sent its moves. It cleared every test chamber. That took 3,336 tool calls.
The game sat paused the whole time the model was thinking, so the run stretched close to 24 hours even though the edited video runs about two. CozyBlaze says the $571.18 was covered by a $200 Codex Pro subscription, and that nobody should read this as a benchmark. OpenAI shipped Astra on September 3 and calls it its best model yet at using computers.
Its Significance:
Reading a screen, moving through 3D space, and planning six steps ahead is the same skill an agent needs to drive the software sitting on your desk right now. That's the promise behind every "AI that uses your computer" demo, and this is one of the first times somebody let it run start to finish in public. A full day for something a ten-year-old finishes over a weekend doesn't seem impressive until you take into account all of the fine motor coordination and decision making that goes into playing first-person video games.
QUICK TAKES
The story: Jakub Pachocki, OpenAI's chief scientist, published an essay on September 6 arguing that AI is grown more than designed, and that studying it looks closer to neuroscience than engineering. He wrote that no lab, his own included, has solved alignment and monitoring well enough to keep scaling at full speed for much longer.
Your takeaway: The man running research at the company that shipped its most capable model three days earlier is asking the whole field to ease off. He expects voluntary slowdowns to become common, and he wants governments to coordinate on it. He does not offer an opinion on getting countries like China to slow down if the United States does.
The story: Reuters went through more than 100 Chinese military procurement notices, patents, and academic papers and found the defense establishment speeding up work on military uses for humanoid robots. Chinese makers shipped roughly 95% of the world's humanoids in 2025, according to BofA Global Research.
Your takeaway: Reuters found no sign China has put an armed humanoid with a real unit, and the machines still burn power and topple over outside controlled demos. But the robots doing backflips in viral clips and the ones in defense papers roll off the same factory floors.
The story: QYResearch, a market research firm, puts the AI drone market at about $2.79 billion in 2025 and projects $6.36 billion by 2032, growing 12.6% a year. It names DJI, Skydio, Autel Robotics, RyzeRobotics, and Helsing as the companies to watch.
Your takeaway: Hold these forecasts loosely. Other research firms size the same market anywhere from under $1 billion to over $20 billion, depending on what they count as an AI drone. Everyone agrees on the direction that the number's going to go up substantially.
TOOLS ON OUR RADAR
⚡ Zapier Freemium: Connect thousands of apps to automate repetitive tasks, 100 tasks a month at no cost, no coding required.
✍️ Hemingway Editor Freemium: Paste in your writing and get instant highlights for run-on sentences, passive voice, and hard-to-read passages, the web version stays free with no account needed.
🤝 HubSpot CRM Freemium: Track contacts, deals, and email in one place with a core CRM that stays free with no time limit, up to 2 users and 1,000 contacts.
🎨 Excalidraw Free and Open Source: Sketch diagrams, flowcharts, and wireframes in a hand-drawn style, right in your browser, no sign-up, no watermark.
TRENDING
Why "Research-Grade" AI Has to Show Its Evidence An opinion piece in The Scientist argues AI should hand researchers back the 20 to 40% of their time that goes to planning, writing, funding, and paperwork, and leave the thinking alone. Take a claim backed by 10 papers. Sounds strong, until you find out all 10 are editorials out of the same lab, while three registered randomized trials from separate groups is a far better case for the same claim.
Three Hikers Rescued After Planning Their Climb With Gemini Three novice hikers from Roseville, California left camp at 3am for Mount Shasta's summit and reached the top at 7pm, seven hours past the mountain's noon turnaround rule. They got lost coming down in the dark, one hurt his knee, and Forest Service rangers pulled them out the next morning. The Siskiyou County Sheriff's Office said Gemini told them to pack far less food and water than they needed, and warned people to never plan a trek on AI alone.
Anthropic Publishes Blueprints for AI Shopping Agents Anthropic released reference code for retailers building Claude-based shopping and merchant agents across retail, travel, telecom, and ticketing. The shopping agent searches a store's catalog, builds a cart, and answers returns questions, but it stops short of paying. Checkout stays with the retailer.
Musk Pulls SpaceX's Orbital Data Center Timeline Forward Again Musk said SpaceX's first AI satellites, running Nvidia chips, will launch in Q4 2027 and hit scale in 2028. That's the second time the date has moved up, from "as early as 2028" in the May IPO prospectus to "next year" on the August 4 earnings call to now. The data centers would use the sun as a power source and space itself for cooling.
MIT Built a Way to See Why a Robotaxi Made Its Choice MIT and Motional built CW-Net, which turns a self-driving model's reasoning into plain phrases like "approaching stopped vehicle" or "close to cyclist" without changing how the car drives. In a track test, it caught something the safety driver had wrong: the car kept stopping near a cyclist, and everyone assumed it saw the cyclist. It hadn't.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🧳 Say your destination and budget. Get a deliberate split of where to splurge and where to save, so nothing feels cheap.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Budget Trip Stretcher — say your destination and budget, get where to splurge and where to save. Persist to localStorage key 'budget_trip_stretcher_v1'.
Aesthetic: dark navy (#0d1220), coral (#e8794f) primary with coral glow top-left, teal (#5aa096) secondary with glow bottom-right. Archivo sans for headings/items, Frank Ruhl Libre serif italic for the overview and why-lines, JetBrains Mono for labels. Two-column layout: coral-bordered "splurge here" items, teal-bordered "save here" items.
Form: destination text input, trip-length dropdown (long weekend / week / 10-14 days / 2+ weeks), budget-feel dropdown (shoestring / moderate / comfortable), traveling-as dropdown (solo/couple/family/friends), top-priority dropdown (food / comfortable stay / activities / nightlife / shopping), optional "what you don't care about" textarea.
System instructions: savvy well-traveled budget strategist, general knowledge of travel cost dynamics (not live prices — qualitative/strategic guidance only, no invented exact dollar figures). Weight splurge picks toward their stated priority, weight save picks toward what they said they don't care about. No em dashes. Return raw JSON: overview (2-3 sentences), splurge (3 items: category, move, why), save (3 items: category, move, why), tactics (2-4 destination-relevant money-stretching moves), worth_it (1-2 sentences, the single best splurge), watch_out (1-2 sentences, a common trap/scam pattern for this destination or style), note (one honest sentence to verify current prices).
Render: gradient "the strategy" overview card. Two-column grid: coral "splurge here" column and teal "save here" column, each with left-border item cards (category label + bold move + italic why). "Money-stretching moves" card with → arrow items. Two-column split: gold "worth every penny" and red "watch out for." A small muted honest note. Copy plan + archive keyed by destination + budget feel.What this does: Enter your destination, trip length, budget feel, who you're traveling with, and what matters most to you. It builds a strategy around three splurge categories, weighted toward your stated priority, and three save categories, weighted toward what you said you don't care about, each with a specific move and why it's the right call for you
What this looks like:
AI Insights. Real Growth. Higher GMV, Better Profits
The difference between growing stores and stagnant ones isn't more effort. It's better insights. StoreClaw analyzes your Shopify and Amazon data, surfaces your biggest growth opportunities, and helps you increase GMV while protecting profit. Start free with bonus tokens. No credit card required.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Play a 3D puzzle game start to finish, reading only screenshots and picking its own inputs.
❌ Still Can't: Do it at a price that makes sense. That run cost $571.18, took 3,336 tool calls, and ate close to a full day.
✅ AI Can Now: Explain a self-driving car's decision in words a person understands, so a human can predict what the car will do next.
❌ Still Can't: Notice its own errors. MIT's test car looked like it was seeing the cyclist. It wasn't. Emergency braking was covering for it.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Short Circuit (1986) - Movie
A military robot designed for battlefield combat gets struck by lightning during a demonstration and starts behaving strangely, escaping the testing facility and turning up at the home of an animal-loving pacifist who has no idea what she's just taken in. The robot, Number 5, spends the rest of the film trying to convince his creators and the government that he has become something they never intended to build. A far gentler film than its military premise suggests, and Number 5 remains one of the most purely likeable AI characters ever put on screen. Likely an inspiration for Walle’s look.
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.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
