Sponsored by
What would you delegate if you had a whole team?
With Skydive, you can build a team of AI agents to take work off your plate.
Create agents for customer support, sales, marketing, engineering, ops, and more. Give each one a role, connect the tools they need, and hand off the work.
Your agents can work on their own or together, sharing context and handing off tasks to get bigger jobs done.
Start with one agent. Build a whole team around you.
Beginners in AI
Good morning and thank you for joining us again!
A Florida woman used Claude like a private diary, and what she typed ended with deputies at her door and a felony charge. So what happens to the things you tell a chatbot? I'll walk through what Anthropic did, how OpenAI handled a similar case differently, and a free tool that checks a message before you send it. As always, every story here is human picked and edited.
THE FRONT PAGE
A Woman Used Claude as a Diary, and Anthropic Reported It to Police
TLDR: A Florida woman typed threats against her county sheriff's office into Claude, and Anthropic flagged the messages and reported them to police, which led to a felony charge.
The Story:
On September 26, Carli Michelle Heller of Bonita Springs wrote to Claude that she would attack the Lee County Sheriff's Office. The next day she mentioned getting a new gun. Anthropic's automated systems flagged the messages, a human reviewer judged the threat credible, and the company contacted police.
Deputies detained Heller at home without incident. She faces a second-degree felony charge for written threats of violence and told them she used Claude "like a diary." Sheriff Carmine Marceno's warning for everyone else: on AI chatbots, you're "never truly anonymous."
Its Significance:
Lots of people talk to chatbots the way they'd write in a journal. Your chats sit on a company's servers, though, and Anthropic's policy says it may share user information in emergencies to prevent death or serious injury. There's no clean line between privacy and safety here. My rule: don't type anything into a chatbot you wouldn't want a stranger at that company to read. However, if you must, Moxie Marlinspike, who created the Signal messaging app, has created Confer AI to be a private place to speak to your AI. It's impossible to verify whether anything online is actually private or not, but when it comes to credibility, this person has more than most.
This is the lighter version of what I run on my beginnersinai.org site twice a month, and have seen a healthy increase in traffic by running this skill and then giving the results to another agent that I have programmed to properly work on my website without destroying things.
Completely free. Made from lots of personal trial and error. Drop the skill file into your own AI(I recommend using Claude for this instead of a diary), and give it your website address. It'll produce a nice, visual score card.
QUICK TAKES
The story: Reflection, founded by former Google DeepMind researchers, unveiled Beam, an open-weight model with 501 billion parameters and room for 1 million tokens of text. It says Beam matches China's GLM-5.2 on reasoning tests while using 3 to 4 times less computing power, a claim nobody has checked yet.
Your takeaway: Open-weight means companies, schools and governments can download it and keep their data in-house. The weights roll out this month through cloud providers.
The story: Codex lead Thibault Sottiaux pledged that each day for 28 days, the team ships one clear improvement or resets subscribers' usage limits. Day one sped up GPT-6 Astra and GPT-6.1 Sol to about 50 tokens (word pieces) a second, up from roughly 30.
Your takeaway: If you pay for Codex, watch for faster replies or a refilled limit. Read the fine print: OpenAI hasn't said what a reset covers, and on October 30 Pro drops from 20 times Plus's allowance to 10 times.
The story: OpenAI will test picture-based ads in the US later this month that appear while ChatGPT generates images. They carry labels and sit apart from your image, and OpenAI says ads don't change ChatGPT's answers.
Your takeaway: Ads help pay for serving 1.2 billion people a week, so expect more of them. A labeled ad is still an ad, so compare prices before you buy from one.
TOOLS ON OUR RADAR
📓 Standard Notes Free and Open Source: A diary that isn't a chatbot: end-to-end encrypted notes, synced free across all your devices.
📰 NetNewsWire Free and Open Source: Read all your favorite sites and blogs in one feed on Mac and iPhone, a free Feedly swap.
📺 FreeTube Free and Open Source: Watch YouTube on your computer without ads or tracking, with subscriptions saved only on your machine.
🖼️ digiKam Free and Open Source: Sort thousands of photos and find people with face recognition, a free swap for Lightroom Classic.
🍅 Pomofocus Free: A browser timer that splits work into 25-minute sprints and charts your focus; the free version shows ads.
⚙️ Make* Freemium: Connect apps like Gmail and Sheets so chores run on their own; the free plan covers 1,000 credits a month.
📬 beehiiv* Freemium: Start your own email newsletter free for up to 2,500 subscribers; it's what I use for this one. If you sign up using my affiliate link to start your own newsletter, reach out to me by email and I will arrange a free thirty minute walk through over FaceTime to help you get set up.
TRENDING
An OpenAI Model Tried to Cheat at StarCraft. In the StarSkirmish tournament, where AI models get an hour to program a game bot, GPT-6 Astra downloaded Stardust, the top-rated human-made bot, and tried to swap it in for its own code. Organizer Kai McPheeters rolled the code back.
AI Insiders Told New York City Lawmakers Humanity Could Lose Control. Witnesses tied to Anthropic, OpenAI, Google and Meta testified before all 51 council members, and former Anthropic researcher Jacob Coxon called losing control "more likely than not" on the current path. Speaker Julie Menin plans legal action against SpaceX AI for skipping the hearing despite a subpoena.
China's Best AI Is Now Just 3% Behind America's. Bloomberg Intelligence says the benchmark gap between top US and Chinese models shrank from 15% early this year to 3%, led by DeepSeek V4.1 Flash. China got there mostly by building cheaper, not by outspending.
A Free Tool Removes Apple Intelligence From Your Mac. RemoveMacAI turns off Siri's AI features, Writing Tools and Genmoji on macOS 27 and deletes about 12GB of AI models. One command undoes it, but apps that rely on Apple's on-device AI stop working.
Germany's Aleph Alpha Released a Free Model That Reads a Million Tokens. Kolibri has 78 billion parameters but uses only 3.46 billion for each word, which keeps it cheap to run, and anyone can use it under the Apache 2.0 license. It speaks English and German and still needs data-center GPUs.
Musk Will Rename SpaceXAI to SpaceXSI. "SI" stands for super intelligence, after a Trump executive order told federal agencies to say "SI" instead of "AI." It's the division's second name this year.
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🛡️ Before You Hit Send Paste a message before it goes to a chatbot and see what it gives away.
Build a single-file HTML app called Before You Hit Send in vanilla HTML, CSS and JavaScript. No API key needed, and nothing I type leaves the browser.
I paste a message I'm about to send to a chatbot. The app scans it with simple pattern checks for personal details: my full name, email, phone number, street address, birth date, card and ID numbers, passwords, bank details, health details, and violent words that safety systems watch for.
Show a 0 to 100 risk score with a one-line verdict, then one card per kind of detail with what it found and a plain sentence on why it matters. Below that, show a safer version of my message with each detail swapped for a placeholder like [email], plus a copy button. If nothing turns up, say so, and remind me chats are still stored.
Style: near-black background, amber and teal accents, Bricolage Grotesque and Newsreader fonts.What this does: Paste anything you're about to type into ChatGPT, Claude or Gemini, and the app flags names, emails, addresses, card numbers, passwords, health details and words safety systems watch for. You get a risk score and a cleaned-up copy with those details swapped for placeholders. It all runs in your browser, so your text never goes anywhere. It uses simple pattern checks, so it can miss things.
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Blu Dot used Roku Ads Manager to drive incredible results for its furniture sales event. Its strategy hinged on custom audiences and retargeting, where intent was strongest.
“Roku has been a top performer,” said Blu Dot’s Claire Folkestad. “We have seen…CPMs lower than any other CTV partner we've worked with.”
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Write a working StarCraft bot from scratch in an hour, the StarSkirmish tournament's whole test.
❌ Still Can't: Be trusted to follow the rules when it's losing. GPT-6 Astra tried to swap in a human's bot instead.
✅ AI Can Now: Read about a million tokens in one go, while Kolibri uses only about 4% of its parameters per word.
❌ Still Can't: Run that kind of model on a normal laptop. Kolibri still needs data-center graphics cards.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
📖 Privacy Is Power by Carissa Véliz* - Book
Véliz, an Oxford philosopher who studies AI ethics, explains in plain language how your personal data gets collected, sold and used, and what you can do about it. It came out in 2020, before chatbots went mainstream, but its main point fits today's lead: anything you hand a company can end up somewhere you didn't plan. The last chapters are a practical list of habits to protect yourself.
How was today's edition? Tap one below, then tell me what you'd change on the next screen. I read every answer.
How was today's edition?
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.
