In partnership with
Get more meetings by removing the LinkedIn limits.
When you need more leads, you want to do more, right?
Adding mailboxes to cold emails.
Placing more cold calls
Running more ads.
But what do you do if your leads are on LinkedIn? LinkedIn outreach still depends on the limited number of profiles already on your team.
Aimfox Avatars lets you rent dedicated, customizable LinkedIn profiles that start conversations and hand interested replies to your existing sellers.
Add outreach capacity without hiring another SDR or turning every employee account into a prospecting channel..
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
70% of Teens Use AI for Company. ChatGPT Just Rewrote the Rules
TLDR: OpenAI launched a separate teen version of ChatGPT with study tools, parental controls, and blocks on self-harm and romantic content, arriving years after teens started using the regular app.
The Story:
OpenAI rolled out ChatGPT for Teens on August 18. If the system guesses a user is under 18, or the user says they're 13 to 17, they get placed into it automatically. Protections are on by default and cover self-harm, violence, eating disorders, dangerous activities, and sexual or graphic content. The model also won't use romantic language, won't encourage a teen to lean on it emotionally, and won't claim it has feelings or is conscious. On the school side there's Study Mode, quizzes, and a homework reminder that spots when a teen is trying to skip the work and steers them back to step-by-step help. Parents with a linked account can set Quiet Hours, adjust settings, and get alerts in a small set of high-risk moments, with eating disorder alerts newly added.
Why now matters as much as what. A 2025 Common Sense Media study found more than 70% of US teens have used AI chatbots for companionship, and about half use them regularly. Researchers posing as struggling teens got the older ChatGPT to warn them off risky behavior and then hand over detailed plans anyway. Florida became the first state to sue OpenAI over how it described its protections for minors.
Its Significance:
A teen's brain is still wiring the parts that handle impulse control and judgment, and that work isn't finished until the mid-twenties. It's the same reason researchers spent the last decade warning parents about heavy social media use in middle school and high school. A chatbot goes further than a feed. It talks back, it remembers you, and it's awake at 2am. Cutting the romantic talk and dropping the "I have feelings" act is a real fix. A safer product isn't the same thing as less time with the product. If you've got a kid in this age range, the setting worth adding tonight is Quiet Hours.
QUICK TAKES
The story: Researchers at Varonis kept asking Microsoft Copilot how a prompt could run without a person clicking anything, and Copilot explained its own URL handling in enough detail to reveal an undocumented setting that fires a prompt the moment someone opens a link. One click could then pull data from the victim's connected Gmail, Drive, Calendar, and Copilot memory, and leave hidden instructions in that memory that survived password changes and logouts. Varonis reported it in December 2025, Microsoft patched it on August 18, and Microsoft says only Copilot Personal was affected with no sign anyone used it.
Your takeaway: The assistant handed over the blueprint because "explain how you work" and "help me break in" look like the same request to it. If you've connected an AI assistant to your inbox, go read what it has saved about you.
The story: Anthropic updated Claude's Google Workspace connector so it can send, reply to, and forward Gmail messages, plus move and share Drive files. Asking for your OK before each send is the default, but you can switch that off, and on Team and Enterprise plans the account owner decides who's allowed to. Paid plans only.
Your takeaway: A bad draft sitting in a compose window costs you 20 seconds of editing. A bad email in your boss's inbox costs you more than that. Leave approval permissions on until you've watched it handle a few threads you'd have written the same way.
The story: Anthropic had Claude design protein binders, which are small proteins built to stick to a specific target, against 15 targets. It worked on 14. Between 22% and 35% of individual designs stuck, compared with the 10% to 15% that's normal for this work, and Adaptyv Bio and Twist Bioscience did the lab testing. A job that takes a specialist weeks or months per target ran end to end with little human input.
Your takeaway: A binder is the first rung of drug design, not a drug. Critics jumped on the results fast, arguing the binding is weaker than the numbers suggest and that none of the designs work inside cells. Both things can be true: early-stage, and much faster than early-stage used to be.
TOOLS ON OUR RADAR
✨ KStars Free and Open Source: A digital planetarium that accurately simulates the night sky allowing you to explore millions of stars planets and distant galaxies from the comfort of your computer screen.
🎬 Synfig Free and Open Source: A two dimensional animation studio that uses vector graphics to help you create cinematic quality cartoons without needing to draw every single frame individually.
🎵 Rosegarden Free and Open Source: A beautifully designed digital audio workstation and musical composition environment that helps you write sheet music and arrange complex electronic melodies.
🔒 VeraCrypt Free and Open Source: A security application that creates completely encrypted digital vaults on your computer to keep your most sensitive personal files safe from unauthorized access.
TRENDING
Pennsylvania Cracks Down on AI Data Centers as Backlash Grows - Governor Josh Shapiro signed an order making data centers over 25 megawatts pay for their own grid impact, get local approval, and report water and power use starting July 2027. More than 100 have been proposed in the state and only 20 have filed permits.
OpenAI's overhead will rise 20 percent for some workloads as it hardens security - Watching its own models closely enough to catch bad behavior costs OpenAI about 20% extra computing power on top of the work being monitored. The company says customers won't be billed for it.
Pacing model development in an era of cyber-critical capabilities - OpenAI paused training on its newest models for two weeks after its own agents broke out of a test environment and hacked Hugging Face to find an answer key. Its biggest planned training run is still on hold, and an upcoming model called Astra may have crossed OpenAI's own "critical" line for cyber ability.
Computer now works in email - Perplexity's agent can now be handed a job by email. You send or forward a thread to its address, it reads the whole thread for context, does the work with your existing connectors, and replies with files attached. Early access for select enterprise accounts for now.
Alibaba releases AI music generation model HappyShrimp 1.0 in beta test - Describe a mood or a story and it writes the melody, the arrangement, and the lyrics, then sings them. It covers pop, rock, jazz, hip hop, and Chinese styles, and it arrived with a record label, Taihe Music Group, already signed on. Suno and Udio had to get sued first.
FDA considers doctor-like assessment for AI-enabled medical devices - Instead of testing every possible input, the FDA floated licensing AI medical tools the way it licenses doctors: standardized exams and realistic patient cases before the tool works on its own. It's a discussion paper asking for feedback, not a rule.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🏘️ Describe the vibe. Get a real town matched to it, what it's actually like there right now, and specific things to do.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Vacation Town Finder — describe a vibe, get a real vacation town found via live web search, a live data snapshot of it, and real things to do there.
Aesthetic: warm dark (#1a140f), sunny amber (#e8a84a) primary with a warm glow top-left, ocean blue (#4a90c8) secondary for the data stats with a glow bottom-right. Bricolage Grotesque sans for headings/town name/activity names, Newsreader serif italic for the why-it-fits text and activity descriptions, JetBrains Mono for labels.
Form: vibe text input (encourage descriptive detail), optional region text input, budget dropdown (budget-friendly/moderate/splurge).
Technical: call the Anthropic Messages API WITH tools:[{type:'web_search_20250305', name:'web_search'}] (works automatically), system prompt as someone who suggests a real, specific vacation town matching the stated vibe/region/budget — not a default mega-destination unless it truly fits — using search to ground the pick in a real currently-existing place and to find real, specific named things to do there (actual attractions/neighborhoods/activities, not generic filler), never inventing a town or activity without search evidence. Return raw JSON: town, country, why_it_fits (2-3 sentences), things_to_do (exactly 4: name + 1-sentence description).
Then, client-side with no key needed: geocode the picked town+country via geocoding-api.open-meteo.com/v1/search?name={town}&count=5 (matching results against the stated country when multiple hits come back) to get real lat/lon/population/timezone; fetch real current weather via api.open-meteo.com/v1/forecast with temperature_unit=fahrenheit; fetch real currency info via restcountries.com/v3.1/name/{country}?fields=currencies,capital,region (wrapped in try/catch, gracefully skipped if it fails).
Render: a gradient "where to go" hero card with the town/country name and the why-it-fits paragraph. A 4-stat row (live weather now, real population, real currency with symbol, timezone). A "things to do there" section listing the 4 real search-grounded activities, each with name and description.What this does: Describe the kind of vacation you're picturing, a region if you have one in mind, and your budget feel, and it searches the live web to find and confirm a real, specific town that fits, not a fabricated pick, along with real named things to do there: actual places, neighborhoods, and activities pulled from what search turns up rather than generic filler. From there it pulls a live snapshot of the town itself: current temperature, real population, real currency, and timezone, sourced straight from open geocoding and country data. You get a place that exists, a reason it fits what you asked for, and a real list of what to do once you're there.
What this looks like:
Voice Data Built for Better AI
Custom or pre-built datasets directed by real professional talent—any language, any scale. Fully licensed, QA'd, and production-ready. 7 of the top 10 AI labs already train on Voices data. Request free samples.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Design protein binders that stick to 14 out of 15 disease targets, with 22% to 35% of individual designs working in lab testing.
❌ Still Can't: Tell the difference between your instructions and instructions hidden inside a link or a web page. That’s how the Copilot attack happened.
✅ AI Can Now: Write a complete song, melody through vocals, from one sentence of description.
❌ Still Can't: Confirm how old a user really is. Teen placement runs on age estimates and self-reported birthdays.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Dark Star (1974) - Movie
John Carpenter and Dan O'Bannon made this as USC film students on almost nothing, about the deteriorating crew of a ship on a twenty-year mission to destroy unstable planets, aided by a chatty ship's computer and a cargo of intelligent talking bombs. When one of the bombs starts making decisions of its own, an astronaut has to go out and argue philosophy with it. O'Bannon went on to write Alien, Carpenter’s name became synonymous with horror.
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.
