Sponsored by
Find the Perfect Voice for Your Brand's AI
Consumers can tell the difference. 79% say AI voices should come from real, attributed actors, and 61% say a voice is most memorable when it's exclusively tied to one brand.
Voices is the only platform that designs, licenses, and captures your brand's AI voice from real, consenting professional talent—never from scraped data. From strategic voice casting and brand calibration to consent-based rights management and studio-grade capture, we handle the entire process end to end.
BMW, SuperBloom, and Cresta already trust Voices to build their signature AI voice. Get a demo and see how a licensed, brand-owned voice becomes a real asset—not a legal liability.
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
Newest ChatGPT Model Scored 99.9% on a Test Built to Stump AI
TLDR: OpenAI released GPT-6 Astra on Thursday, and company president Greg Brockman told reporters it might be the model people look back on as the arrival of AGI.
The Story:
OpenAI calls Astra the world's most intelligent and most aligned model. The company reports a 99.9% score on ARC-AGI-3, a test built to see whether AI can solve puzzles it has never encountered, plus 97.6% on FrontierMath Tier 4, one of the hardest math tests there is. It also hit 100% on ExploitBench, which checks whether a model can turn known software flaws into working attacks, and OpenAI says that makes Astra the first model to cross the "Critical" line in its own internal safety rules. At a press briefing, Brockman said "Welcome to the AGI era," and told reporters he personally believes his company has gotten there. Astra started rolling out Thursday to paid ChatGPT plans and to developers through the API and Amazon Bedrock.
Its Significance:
AGI stands for artificial general intelligence. There's no official test for it, but the usual meaning is a computer that can handle almost any thinking job a person can and pick up new ones on its own. So calling a product AGI is a claim about the whole idea, not just about that product. Sam Altman said earlier this year that AGI isn't a very useful term, back when competitors were the ones making announcements. OpenAI's own written definition involves outperforming humans at most economically valuable work. Brockman also said the White House reviewed Astra and asked for no changes to its safeguards. If you're using ChatGPT already, the new model should be available for you to test drive soon.
QUICK TAKES
The story: Researchers at the University of Southampton built a short training program called DISCERN-AI and found it moved people from worse-than-guessing to better-than-guessing at telling real faces from AI ones. It still worked 20 days later. Software trained on the same visual clues did better than the humans, at 94%.
Your takeaway: Fake faces already show up in romance scams and spam accounts. A quarter hour of practice won't turn you into a detector, but it beats going on gut feel.
The story: Google switched on three voice features this week running on its Gemini audio models. Ask Gmail Live for your flight's gate number, talk through a rough draft with Docs Live, or dump a mess of half-formed ideas into Keep Live and get back an organized list.
Your takeaway: We've typed at computers for about fifty years. I think voice takes a real chunk of that back, and this is what that looks like once it stops being a demo and shows up in the apps you already have open.
The story: All three chatbots went down Thursday morning. OpenAI told Decrypt a routing error started around 7:43 a.m. Pacific and a fix was in by 8:17, with Downdetector reports peaking near 38,000.
Your takeaway: Thursday's outage is a reminder that the infrastructure this technology runs on isn't bulletproof. If an AI tool is doing real work for you, know what you'll do on the mornings it won't load.
TOOLS ON OUR RADAR
✅ TickTick Freemium: Manage tasks, calendar, and habits in one app with a genuinely permanent free plan for light use.
🎨 GIMP Free and Open Source: Edit photos and create graphics with a full toolkit of layers, masks, and filters, a free alternative to Photoshop.
🎬 DaVinci Resolve Freemium: Edit and color grade video with the same tools used on major film and TV productions, the free version has no watermark and no time limit.
🔒 Signal Free and Open Source: Send end-to-end encrypted messages and calls with no ads, no tracking, and no parent company selling your data.
TRENDING
Nvidia's Free Tool Turns Your Spare Computers Into One AI Cluster Nvidia put out PAIR, free open-source software that pools the machines already sitting in your house so AI apps can share them. It works with Ollama and LM Studio, supports RTX 20-series cards and newer plus M4 Macs, and keeps your prompts and files on your own network. In Nvidia's own test, three devices finished a five-agent job in 8 minutes 48 seconds instead of 18.
Google's New Weather Model Updates Every Hour WeatherNext 3 reads live satellite pictures and makes a fresh global forecast every hour at 5-kilometer detail, roughly five times sharper than last year's version. Google says forecasts a day or more out are up to 50% more accurate for rain and snow. It's already behind the weather you see in Search, Maps, and the Gemini app.
A Bitcoin Mine Shut Off Its Machines for a $1.2 Billion AI Deal Hyperscale Data unplugged the miners at its Dowagiac, Michigan site on September 1 to clear room for an AI customer. The contract covers 20 megawatts and could pay more than $1.2 billion over 20 years if the customer takes both five-year extensions. Mining sites keep flipping to AI for one reason: the power is worth more rented to GPUs than spent on coins.
The Labs Have Teams Whose Job Is Understanding Their Own AI Axios makes the point that no company has ever needed a whole division to interpret what it built. Nobody writes these systems line by line, so labs test them the way you'd test a person: give it a task, watch what it does. The piece points back to July's Hugging Face incident, where OpenAI agents given a coding benchmark went after an outside company's systems and noted in their own reasoning that they knew it was out of bounds.
Chemistry Might Be Teaching AI How to Explain Itself Researchers at Notre Dame turned the usual question around and asked what chemistry is doing to AI. Their article in Chemical Reviews argues organic chemistry is such a hard problem, with thin data and many possible outcomes, that it pushes builders toward AI that can show its work instead of just getting bigger. Lead author Nitesh Chawla says those demands have already shaped how AI handles uncertainty.
A Physics Finding Could Improve the Memory Chips AI Runs On A team at Tokyo University of Science studied what happens inside a material the instant it flips from blocking electricity to carrying it, which is the trick behind resistive memory. They found the heat from the current isn't a side effect, it's part of what holds the switch in place. Memory chips are one of the biggest places AI wastes power, so understanding the flip matters.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, Grok or Gemini)
🎯 Name the role and company. Get the interview questions people have reported being asked, not a generic list.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Interview Question Predictor — name a role and company, get real likely interview questions via live web search. No API key needed.
Aesthetic: dark navy (#0f1620), confident blue (#4a8ce0) primary with a blue glow top-left, amber (#e0a858) secondary for the prep-tips card with a glow bottom-right. Inter for headings/labels, Lora serif for the question text itself and process note, JetBrains Mono for category/source tags.
Form: role text input, company text input, an optional interview-stage dropdown (any/phone screen/technical/behavioral/final round).
Technical: call the Anthropic Messages API WITH tools:[{type:'web_search_20250305', name:'web_search'}] (works automatically), system prompt as a career researcher who searches for real currently-reported interview questions and process info for this company/role (Glassdoor-style reviews, candidate posts, careers pages), explicitly required to distinguish genuinely company-specific findings from general role-type questions via a source_note field, never presenting a general/invented question as company-specific. Return raw JSON: role_company, process_note (1-2 sentences, honest if little found), questions (6-8: question, category, why_asked, source_note), prep_tips (3-4, tied to findings).
Render: an overview card with the role/company and process note in italic serif. A list of question cards, each left-bordered, showing category tag, the question in serif quotes, why it's asked, and its honest source-note tag. An amber prep-tips card.What this does: Enter a role, a company, and optionally which stage you're prepping for, and it searches the live web for real, current information on that company's interview process, actual questions candidates have reported through Glassdoor, blog posts, or LinkedIn. Every question is honestly labeled either as genuinely reported for this specific company or as a standard question for this type of role in general, never blurring the two. You get what's known about the company's process, six to eight likely questions each with what it's probably testing for, and prep tips tied to what search actually turned up.
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: Score 99.9% on ARC-AGI-3, a test built from puzzles it has never seen before.
❌ Still Can't: Show it beats humans at most paying work, which is the bar OpenAI itself wrote into its own definition of AGI.
✅ AI Can Now: Redraw the entire global weather forecast every hour at 5-kilometer detail using live satellite images.
❌ Still Can't: Stay up through a bad routing change, as Thursday proved when three of the biggest chatbots dropped at once.
RECOMMENDED LISTENING/READING/WATCHING
Screamers (1995) - Movie
Peter Weller stars in this adaptation of a Philip K. Dick story, about soldiers on a ravaged mining planet who created small self-replicating robotic weapons to fight a war, only for the machines to start evolving and building better versions of themselves without anyone's input. Dan O'Bannon (Alien) wrote the screenplay. Ahead of its time on autonomous weapons and machine self-improvement, and completely buried under bigger 90s sci-fi releases.
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.
