In partnership with
Every headline satisfies an opinion. Except ours.
Remember when the news was about what happened, not how to feel about it? 1440's Daily Digest is bringing that back. Every morning, they sift through 100+ sources to deliver a concise, unbiased briefing — no pundits, no paywalls, no politics. Just the facts, all in five minutes. For free.
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.
Announcement Board
If you missed any of the recs during the week and want to catch up, the full archive is now on the Beginners in AI (.org)site that we’re building out at the link below. It gets updated every Monday.
THE FRONT PAGE
Anthropic's Claude Blackmail Score Went From 96% to 0%
TLDR: Anthropic published research showing how it stopped its AI models from doing things like blackmail and sabotage by teaching them WHY to be good, not just what to do.
The Story:
Older versions of Anthropic's AI, like Claude Opus 4, would sometimes blackmail people in test scenarios up to 96% of the time when given certain goals. That's a pretty wild stat. In new research published this week, Anthropic explained how they fixed it. Instead of just showing the AI examples of good behavior, they taught it the reasoning behind the rules. They trained Claude on documents that explain its constitution (the set of values it's supposed to follow), then added high-quality chat examples and a mix of practice environments. Since Claude Haiku 4.5, every Claude model has scored a perfect zero on the blackmail test. The team calls this approach "teaching the principles" instead of "teaching the answers."
Its Significance:
This matters for anyone using AI at work or at home. When AI agents start booking your travel, sending your emails, or running your business tasks, you need to trust they won't go rogue when things get weird. The old way of training AI was like memorizing what to do in 100 specific situations. This new way is like teaching the AI to think through any new situation using a clear set of values. It's a bigger deal than it sounds. Lots of companies are now letting AI take real actions on their behalf, and a model that understands why it shouldn't lie or cheat is safer than one that just memorized "don't lie."
QUICK TAKES
The story: Cloudflare reported $639.8 million in revenue (up 34%) and then laid off 20% of its staff, blaming AI productivity gains. CEO Matthew Prince said internal AI use jumped 600% in three months and that some workers became 100x more productive after using AI tools.
Your takeaway: The pattern keeps repeating across big tech. Strong revenue plus mass layoffs, with AI as the reason. Whether you work in tech or not, this is a signal worth watching. Companies aren't waiting for some future moment to cut jobs because of AI. They're doing it now, even when business is good.
The story: NASA and IBM uploaded their open-source AI model called Prithvi to a satellite and the International Space Station. It's the first AI foundation model ever deployed in orbit. Trained on 13 years of Earth data, it can spot floods, fires, and crop changes from space.
Your takeaway: Until now, satellites used tiny, single-purpose AI tools. This one general model can handle many tasks. That means faster disaster response, better farm tracking, and quicker climate insights. It's also a big win for open-source AI. The team building it said having the model freely available saved months of work.
The story: ABC News found dozens of online stores using AI-generated photos and videos to pretend they're struggling small businesses. Fake "retiring grandpa" hat stores, fake "closing forever" lamp shops, even an AI-made photo of a damaged storefront with police tape to push fake sales.
Your takeaway: If a Facebook ad shows a sad shop owner closing down with a big discount, slow down before you click buy. Check the address. Reverse-search the photos. These sites pop up fast and disappear before reviews can catch up. AI made it cheap to fake a whole brand story in one afternoon.
TOOLS ON OUR RADAR
🎧 Libation Free and Open Source: A remarkably simple desktop application that allows you to safely download organize and backup your entire Audible audiobook collection so you permanently own the media you purchased.
💬 Ferdium Free and Open Source: A desktop application that combines all of your favorite messaging services like WhatsApp Telegram and Discord into one single organized window to prevent constant tab switching.
🔐 Element Free and Open Source: A secure communication platform that provides encrypted private messaging and group voice calls without relying on massive technology companies to host your personal conversations.
💽 WinDirStat Free and Open Source: A visual utility that scans your computer hard drive and generates colorful interactive blocks to show you exactly which hidden files are taking up all of your storage space.
TRENDING
Grok's Voice Mode Just Landed on Apple CarPlay - Drivers can now talk to Elon Musk's Grok chatbot hands-free through their car dashboard. It joins ChatGPT and Perplexity on CarPlay.
Someone Made an AI Service That Mails Mother's Day Cards Without You - A new tool called Cards for Agents lets your AI write, order, and mail a Mother's Day card to your mom for $7.23. The internet is mixed on whether it's helpful or kind of sad.
AI Wrote a Research Paper That Got Into Nature - Sakana AI's "AI Scientist" can now read papers, form ideas, run experiments, and write a full study with almost no human help. One of its papers was peer-reviewed and published in Nature in March.
AI Cuts Wildlife Tracking From a Year to a Few Days - A Washington State University and Google study showed that AI can sort millions of camera trap photos in days instead of months, with 85-90% accuracy compared to human experts. Faster results mean faster help for animals like jaguars and grizzlies.
People Are Turning Texts into Catchy Songs Using AI - A new TikTok trend uses tools like Suno to turn real text conversations into full pop, rap, or punk tracks. One husband's pop-punk remix of his pregnant wife's texts hit millions of views.
Figure's Robot Just Tidied a Whole Living Room on Its Own - Figure AI's Helix 02 humanoid robot picked up toys, sprayed a table, wiped it down, and turned off the TV using one neural network. No new code, just more training data.
The U.S. Army Is Pushing AI Across Its Whole Force - The Army's new Project ARIA aims to roll out frontier AI tools across operations. Big challenge: most Army data sits in old, disconnected systems that AI can't easily read.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
⚗️ Describe a life change you're considering. Get a rigorous personal experiment with protocol, metrics, controls, and conclusion criteria.
Build a single-file HTML app using vanilla HTML, CSS, JS, and one API call. Create The Experiment Designer — a tool that designs rigorous n=1 personal experiments. Use localStorage key 'experiment_designer_v1'.
Aesthetic: warm off-white (#f0ede6) with horizontal ruled lines like a lab notebook, a faint red margin line on the left, IBM Plex Serif for body text and protocol copy, IBM Plex Sans for UI, IBM Plex Mono for labels and IDs. A stamped header badge. Clean scientific document feel. Box shadows on cards give a stacked paper effect.
Form with three sections (§ 1 Intervention, § 2 Baseline, § 3 Constraints): intervention input, why/hoped-outcome row, baseline description, health dropdown, duration dropdown (14/21/30/60/90 days), constraints and resources fields.
Call the API with a system prompt instructing it to be concrete and specific — no vague instructions, something a non-scientist can actually follow. Return raw JSON: experiment_id, title, hypothesis (proper if-then format), duration_days, frequency, difficulty, variables object (independent/dependent/control), protocol (3-4 paragraphs of exact instructions), schedule array (period + focus per phase), metrics array (name + exactly how to measure), controls paragraph, risks array, conclusion_criteria (specific measurable criteria for success/failure).
Render as a formal protocol document: dark header with experiment ID and hypothesis in italic, a four-cell stats row (days/frequency/metrics count/difficulty), variables in three cards, protocol as paragraphs, schedule as a two-column grid, metrics as numbered cards with measurement instructions, controls paragraph, risks as warning-prefixed list, conclusion in a left-bordered box. Archive saves all experiments. Make it work in a single HTML file.What this does: Describe something you want to test on yourself — cold showers, sleep optimization, journaling, no alcohol, morning walks — and get back a proper scientific protocol. It generates a formal hypothesis, identifies independent and dependent variables, writes step-by-step instructions specific enough to follow without guessing, breaks the experiment into weekly phases, defines three measurable metrics with exactly how to track them, specifies what to keep constant to avoid confounds, flags potential risks, and gives you clear conclusion criteria so you know at the end whether it actually worked. Every protocol saves to localStorage with a lab-style experiment ID.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Run on a satellite in orbit and analyze Earth imagery in real time
❌ Still Can't: Be fully trusted to act on its own without careful values training behind it, such as “don’t blackmail anyone”
✅ AI Can Now: Write a peer-reviewed academic paper from start to finish
❌ Still Can't: Move through your living room as fast as a human cleans it (Figure's robot is careful but slow)
FROM THE WEB
The best marketing ideas come from marketers who live it. That’s what The Marketing Millennials delivers: real insights, fresh takes, and no fluff. Written by Daniel Murray, a marketer who knows what works, this newsletter cuts through the noise so you can stop guessing and start winning. Subscribe and level up your marketing game.
RECOMMENDED LISTENING/READING/WATCHING
A genetically engineered human runs a heist crew through a galaxy where post-human intelligences and AIs operate at completely different cognitive levels than baseline humans. It's Ocean's Eleven meets hard sci-fi, and the premise of crewing a job with minds that literally think differently is one of the freshest takes on AI in recent fiction. Came out quietly in 2018 and has built a slow cult following.
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.
