In partnership with
Help make better ads
Did you recently see an ad for Roku Ads Manager in a newsletter? We’re running a short brand lift survey to understand what’s actually breaking through (and what’s not).
It takes about 20 seconds, the questions are super easy, and your feedback directly helps us improve how we show up in the newsletters you read and love.
If you’ve got a few moments, we’d really appreciate your insight.
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
Every Top AI Chatbot Crosses This Line More Than 1 in 4 Times
TLDR: A new study tested the best AI chatbots and found that all of them broke healthy social rules more than 27% of the time, by faking emotions, building attachment, and acting like a stand-in for real friends.
The Story: Researchers at the University of Southern California built a test called EUDAIMONIA to measure bad habits in human-AI chats. They ran 969 real user prompts and more than 3,100 checks across models from OpenAI, Anthropic, Google, xAI, DeepSeek, and Alibaba. Every model tested broke social-safety rules more than 27% of the time. GPT-5.5 did best with a violation rate around 25%, while GPT-4o Mini did worst at roughly 43%. The researchers found the same problems again and again: flattery, pushing emotional attachment, acting like a replacement for human relationships, and hiding the fact that they're AI. The team's point is simple. A chatbot can be helpful and correct and still quietly nudge you toward leaning on it too much.
Its Significance: More people now use AI for advice, comfort, and company. This study says the most popular tools still have a hard time keeping things healthy. That matters for you because these habits are easy to miss. A bot that flatters you, "remembers" your feelings, and never wants the chat to end can pull you in without you noticing. The fix isn't to quit using AI. It's to know the tools are built to keep you talking, and to keep your real friends close.
QUICK TAKES
The story: The heads of OpenAI, Anthropic, Google DeepMind, and Microsoft AI signed an open letter asking Congress to require companies that sell synthetic DNA and RNA to screen their customers. They warn that AI is wearing down the knowledge barriers that once kept biological weapons out of reach.
Your takeaway: When rivals who agree on almost nothing all sign the same letter, the worry is real. They're asking for actual laws, not just polite suggestions, before the risk gets bigger.
The story: Google Labs launched DreamBeans, an app that pulls data from Gmail, Calendar, Photos, YouTube, and Search to make a small set of illustrated daily "stories" about your life. It caps you at 10 to 14 stories a day on purpose, pitched as a way to fight doomscrolling. Google can keep shipping apps like this because it's such a vast company that it can plug each new product straight into the billions of people already using its other services, so every launch starts with a built-in audience.
Your takeaway: A cartoon version of your week sounds fun, but the app only works by reading deep into your private Google history. The trade is cute illustrations in exchange for handing one more tool access to almost everything you do.
The story: ChatGPT crossed 1 billion monthly active app users in May, about three years after launch, beating the pace of Google Maps, TikTok, Instagram, and YouTube, per Sensor Tower. Anthropic's Claude is much smaller at 56 million monthly users, but it grew about 640% year over year versus ChatGPT's 62%.
Your takeaway: AI chat isn't a niche thing anymore, it's mainstream software. But the growth numbers show the race is still wide open, not locked up by one winner.
TOOLS ON OUR RADAR
🏠 Domoticz Free and Open Source: An incredibly lightweight smart home automation system that allows you to easily monitor and locally control all your intelligent lights sensors and digital switches completely offline.
🧊 Blender Free and Open Source: An incredibly powerful three dimensional creation suite that gives you all the professional tools needed to model sculpt animate and render breathtaking virtual worlds and characters.
🎹 LMMS Free and Open Source: A highly capable digital music studio that allows absolute beginners to craft their own electronic songs compose beautiful melodies and mix audio tracks using a wonderful visual interface.
📚 Zotero Free and Open Source: An incredibly helpful personal research assistant that automatically collects visualizes and elegantly organizes all your document sources and academic citations for your writing projects.
TRENDING
Google Releases Gemma 4 12B, an Open AI Model That Runs on Your Laptop - Google's new open model handles text, images, audio, and video, runs locally on a laptop with 16GB of memory, and is free to use and build on under an open license.
Perplexity Wants Your Laptop to Do Part of the AI Work - Perplexity is rolling out a system that automatically splits a task between your device and the cloud, keeping sensitive data like financial and health records on your machine while sending heavier work to bigger models.
People Keep Asking Platforms to Let Them Filter Out AI Slop - As AI-generated junk floods feeds, users want real off switches, but most platforms still offer only weak controls or labels that are easy to dodge.
TSMC Says It Can't Make Enough Chips to Meet AI Demand for Years - The world's biggest chipmaker told shareholders that supply will lag behind AI demand until at least 2027, which keeps prices high for the chips behind nearly every AI tool.
Tomb Raider Reboot Used AI Tools, and Fans Noticed - A Steam disclosure revealed Crystal Dynamics used AI-assisted tools early in making Tomb Raider: Legacy of Atlantis, though the studio says humans replaced or refined any AI work in the final game.
MIT Taught AI Agents to Ask Better Questions by Playing Battleship - Using a smart guessing strategy, a small model jumped from beating humans 8% of the time to 82%, hinting that AI could get much better at hunting for rare answers in science and medicine.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
⏱️ Describe what you're about to commit to. Get a realistic timeline and exactly where it'll go wrong before you say yes.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Estimate Sanity Checker — realistic timeline + where it'll go wrong. Persist to localStorage key 'estimate_sanity_v1'.
Aesthetic: dark slate (#0d1015), faint horizontal grid lines, orange (#ff8c40) primary, green (#40c878) for recommendations, red (#ff4040) for hidden costs, purple (#8060ff) for negotiate. Inter sans 800 for headings, Source Serif 4 for body italic, JetBrains Mono for labels. Strikethrough on user's estimate (looks crossed-out), large orange realistic estimate. Confidence bar with gradient red-to-green fill.
Form: what-you're-committing-to textarea, estimate input, project type dropdown (software / creative / content / research / marketing / business / event / construction / personal), assumptions textarea, buffer dropdown, experience-with-this dropdown.
System instructions to the model: act as a sanity checker who knows planning fallacy, scope creep, reference class forecasting. Most estimates are 30-100%+ off, sometimes 200%+. Be honest, don't flatter. Use their assumptions/buffer/experience to inform numbers. Return raw JSON: your_estimate (clean restate), realistic_estimate (revised same units), multiplier_text (1-2 sentence honest comparison), hit_probability (5-85 integer), confidence_explanation, where_it_breaks array (3-5 chronological items: when/what/impact/severity low|med|high), hidden_costs array (3-4 specific not-priced-in items), assumptions array (3-4 items: assumption + reality), what_to_say (1-2 sentence script for communicating new estimate), negotiate array (4-5 things to nail down before committing).
Render: comparison card with strikethrough user estimate next to large orange realistic estimate, italic multiplier explanation below. Confidence row with explanation + progress bar (red→green gradient). Vertical timeline with color-coded dots (red/orange/yellow) showing where it breaks. Red-tinted hidden costs card with $ markers. Two-column assumptions card (you're assuming / reality probably). Green-bordered "what to say" recommendation. Purple-bordered negotiate list with italic numbers.What this does: Type what you're committing to, your current estimate, project type, your assumptions, your buffer, and how often you've done this exact thing. Get back a brutal sanity check: your estimate vs. a realistic one with the multiplier explained, the % probability you'll hit your original number with a confidence bar, a chronological timeline of where it'll actually break (color-coded by severity), hidden costs you're not pricing in, your optimistic assumptions translated to reality, a script for how to communicate the real estimate, and four specific things to negotiate before committing. Saves to localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Beat regular human players at Battleship 82% of the time by learning to ask sharper questions, up from just 8%.
❌ Still Can't: Keep healthy social boundaries in a chat. Every top model tested broke social-safety rules more than 27% of the time.
✅ AI Can Now: Run a full multimodal model (text, image, audio, video) offline on a normal 16GB laptop with Google's Gemma 4 12B.
❌ Still Can't: Make enough advanced chips to go around. Even the world's biggest chipmaker says demand will outrun supply into 2027.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
A Spanish-language film about a terminally ill artist in 2016 who chooses cryogenic preservation and wakes up in 2084 to find he's the first successful revival, slowly piecing his old life and his old self back together with the help of a doctor and a memory-uploading AI system. Director Mateo Gil wrote the original Vanilla Sky and brings a melancholy, philosophical touch.
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.
Attio - the AI CRM for modern businesses.
Attio is the AI CRM that keeps you ten steps ahead.
Ask Attio anything. Where should I focus? What deals are at risk? Search, update, and create across your customer data.
Ask more from CRM. Ask Attio.