In partnership with
Exploring AI Voice With SuperBloom
Scaling a campaign globally means finding a voice that resonates everywhere—without losing the overall message. For Deel's "Feeling of Deeling" campaign, agency SuperBloom needed exactly that: a consistent brand voice across markets, deployed fast, without sacrificing quality or consent. They built it with Branded AI Voice, powered by real, professional talent.
SuperBloom and Voices break down how the campaign came to life in this on-demand video session—from strategic talent selection through seamless production workflows and global scale, to the governance that future-proofs an audio strategy built to last. You'll hear directly from the team on how they made this happen, plus their advice if you're looking to explore AI voice for your brand.
If you're a marketing executive, agency creative, or brand leader mapping campaigns in international markets, watch this on-demand session to get a real playbook, not a hypothetical—lessons any creative team can apply to their own global rollout.
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
The Man Who Helped Build ChatGPT Made an AI That Can't Write a Word
TLDR: A former OpenAI researcher built an AI that won't write you a single sentence, and that's the whole point of it.
The Story:
Diogo Almeida worked at OpenAI and helped write the InstructGPT research that ChatGPT was built on. On September 15 his startup, TypeSafe AI, released a model called Jev. It can't write an email or explain your code. You hand it some text and a question with a fixed list of possible answers, and it picks one, then tells you how sure it is.
Here's what that looks like. A shopper emails a store saying they got charged twice. Jev reads it and returns a label, "payment issue," plus a number showing how confident it is that the person wants a refund. The store's software takes it from there. TypeSafe says answers come back in 70 to 500 milliseconds, and input costs $0.042 per million tokens with output free. The company raised $40 million before anyone outside had used it.
Its Significance:
Most AI tools today write words that a person reads. Sorting emails. Flagging refunds. Deciding when a real human needs to step in. TypeSafe says Jev can't make things up, but that only covers the format. It will always pick from your list. It can still pick wrong, and its own test results compare Jev's answers to other AI models rather than to known correct ones. We'll likely see a combination of this used in conjunction with current AI models to get the best of both worlds. Access is behind a waitlist for now here typesafe.ai
QUICK TAKES
The story: OpenAI told the New York Times it has made "substantial progress" on a second Millennium Prize Problem, one of seven famous math puzzles that carry a $1 million reward each. The company won't say which one, though a lot of people online are guessing the Hodge Conjecture, a 1950 puzzle about counting holes in shapes too complicated to picture.
Your takeaway: The first claim, on September 8, is still being reviewed by the Clay Mathematics Institute, and NYU mathematician Tristan Buckmaster says he and an Anthropic researcher finished the same proof first. Checking work like this usually takes years. Treat "progress" as a press release until somebody outside OpenAI confirms it.
The story: Ethereum's co-founder pushed back on the idea that AI-powered hacking makes defense hopeless. He argues the same models good enough to crack hard math problems can also prove that a piece of software does exactly what it claims and nothing else.
Your takeaway: Buterin says about 90% of his money is in crypto, so he's betting his own wallet on this. He also admits the hard part isn't the proving. It's writing down what "secure" even means before you start.
The story: OpenAI released a version of its GPT-6 Astra model built for lawyers, with a search index covering US case law, statutes, regulations and court rules across more than 230 million web addresses. It launched with 26 partner plugins from companies like Thomson Reuters, Harvey and iManage, and ChatGPT for Word is now open to everyone.
Your takeaway: Notice the pattern. Healthcare in one month, financial services the next, now law. If your field has expensive research software and people who bill by the hour, a version of this is coming for it.
TOOLS ON OUR RADAR
🔁 Habo Free and Open Source: Build habits with a fast, privacy-respecting tracker that works on both Android and iOS, unlike most open-source habit apps that are Android-only.
🎤 Slidebean Freemium: Build unlimited presentations for free, exporting or sharing them is what costs money.
📖 GoodLinks Paid: Save articles to read later with a clean, ad-free reader, one-time purchase, no subscription, iOS and Mac only.
🗂️ Baserow Free and Open Source: Build a spreadsheet-database hybrid like Airtable, self-host for free or use their hosted free tier.
TRENDING
Security Experts Want the US and China to Promise Never to Let AI Control Nukes, a new report from Brookings and Shanghai's Fudan University asks Trump and Xi to agree that only people, not AI, can order a cyberattack on the other side's nuclear command systems. Those systems spot threats and carry launch orders, so a bad read there could start something nobody meant to start. The researchers called it a race against time.
Grok Bot Gets a Voice on Desktop and Mobile, xAI began rolling out spoken conversation for its Grok Bot assistant on September 17, so you can talk to it instead of typing. It runs on Grok Voice Think Fast 2.0, which xAI says is 1.4 times faster than the version before it. The rollout is going out account by account, so check your app before assuming it's broken.
Claude Projects Get Rebuilt Around Parallel Work, Anthropic redesigned Projects in Claude Code so one request can be split across several coding sessions running at the same time, with a coordinator handing out the pieces and putting the results back together. You can steer it from your phone and it keeps going after you close the laptop. Beta started September 17 for some Pro and Max subscribers.
Grandmother Jailed Six Months After a Face Match Sues for $10 Million, Angela Lipps, 50, was arrested in Tennessee in July 2025 while babysitting her neighbor's kids, then sent to North Dakota on bank theft charges after facial recognition software matched her social media photo to a fake ID. Her complaint says she didn't resemble the suspect in build, features or tattoos, and she was let go on Christmas Eve once bank records proved she was a thousand miles away. She lost her home, her car and her dog. Fargo says it doesn't comment on pending lawsuits.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
⏱️ Name the skill. See real published milestones for how long it takes, not a vague "it depends."
Build a single-file HTML app with vanilla HTML/CSS/JS. The Fastest Way to Learn It Timeline — name a skill, get real published time-to-competency milestones via live web search, rendered as a visual timeline. No API key needed.
Aesthetic: dark indigo (#12111c), violet (#8c6ce0) primary with a glow top-left, progress-green (#5cc28a) secondary with a glow bottom-right. Inter throughout, JetBrains Mono for time-estimate labels.
Form: single skill text input.
Technical: call the Anthropic Messages API WITH tools:[{type:'web_search_20250305', name:'web_search'}] (works automatically), system prompt as a learning-science researcher who finds real published time estimates from reputable sources (deliberate practice research, FSI-style proficiency scales, documented skill-acquisition studies), building 3-4 milestones from earliest to most advanced, each with a real time estimate and concrete description of capability at that level, grounding every number in real findings, honest that individual pace varies. Return raw JSON: skill, milestones (3-4: level, time_estimate, what_it_looks_like), source_note, acceleration_tips (3-4), honest_caveat.
Render: a vertical timeline with a gradient connecting line (violet to green) and color-shifting milestone dots, each showing level name, time estimate in mono, and description. An italic source-note card. A green tips card. A persistent caveat line.What this does: Type a skill, touch typing, conversational Spanish, chess, and it searches the live web for real time-to-competency data from reputable sources: deliberate practice research, established proficiency scales like the FSI language categories, documented skill-acquisition studies. You get a visual timeline of three or four real milestones, each with a time estimate and a concrete description of what you can do at that stage, plus practical acceleration tips grounded in real learning science. It's honest that individual pace varies rather than presenting these numbers as guarantees.
What this looks like:
One idea shouldn't take six rewrites to post.
Posting everywhere means rewriting one idea six times, so you post to one, or none. SureThing turns one idea into native posts for every platform.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Return a decision with a confidence score in as little as 70 milliseconds, fast enough to sit inside software instead of in front of a person.
❌ Still Can't: Tell you whether that decision was right. Sticking to your list of options isn't the same as choosing the correct one.
✅ AI Can Now: Search more than 230 million pages of US case law, statutes and court rules to pull up the authorities behind a legal question.
❌ Still Can't: Take responsibility for what it finds. A lawyer still has to read the passages and sign the filing.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Manna by Marshall Brain - Book
A short, free-to-read 2003 novel by the founder of HowStuffWorks, following one worker's life as a piece of retail management software called Manna spreads from fast food into nearly every job in the economy, sorting humans into two very different outcomes depending on which country they happen to live in. Written well before large language models existed, and still one of the clearest, most widely circulated thought experiments on how workplace AI could reshape a society, entirely for free on the author's own website.
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.
