In partnership with
This newsletter you couldn’t wait to open? It runs on beehiiv — the absolute best platform for email newsletters.
Our editor makes your content look like Picasso in the inbox. Your website? Beautiful and ready to capture subscribers on day one.
And when it’s time to monetize, you don’t need to duct-tape a dozen tools together. Paid subscriptions, referrals, and a (super easy-to-use) global ad network — it’s all built in.
beehiiv isn’t just the best choice. It’s the only choice that makes sense.
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 published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
8 Million "Protected" Users Just Had Their AI Chats Sold
TLDR: A "Featured" Chrome VPN extension with 8 million users was secretly harvesting and selling complete conversations from ChatGPT, Claude, Gemini, and five other AI platforms—while warning users to protect their privacy.
The Story:
Security researchers at Koi Security discovered that Urban VPN Proxy—a 4.7-star Chrome extension with over 6 million installs and Google's "Featured" badge—has been intercepting every prompt and response from ten major AI platforms since July 2025. The extension injects dedicated scripts (chatgpt.js, claude.js, gemini.js) that override core browser APIs to capture raw conversation data before your browser even renders it. The same harvesting code appears in seven other extensions across Chrome and Edge—including an ad blocker and "browser guard"—bringing the total affected users past 8 million. All captured conversations flow to BiScience, a data broker that sells "insights" to advertisers and business partners. There's no toggle to disable collection; uninstalling is the only fix.
Its Significance:
The extensions still carry Google's "Featured" badge and remain live as of today, meaning the manual review process that's supposed to protect users completely missed—or ignored—code designed to harvest conversations from Google's own AI product. For anyone who's discussed medical concerns, financial decisions, proprietary code, or personal dilemmas with an AI assistant, this is a concrete reminder: the intimacy people develop with AI chatbots creates a data honeypot unlike anything that came before. If you've used any Urban-branded extension since July, assume those conversations are already in a marketing database. Two high quality VPN providers that have proven no-log status are Proton VPN and Mullvad VPN.
QUICK TAKES
The story: Adobe added new AI video editing features to its Firefly tool that let users change videos by typing simple commands like "change the sky to overcast" or "zoom in on the subject." The update uses Runway's Aleph model and includes a new timeline view, camera motion controls, and the ability to upscale videos to 4K quality. Pro and Premium subscribers get unlimited video generations through January 15, 2026.
Your takeaway: Adobe is making video editing as easy as describing what you want, which could change how creators work—but it's also relying on another company's AI model rather than building everything in-house.
The story: Mozilla appointed Anthony Enzor-DeMeo, a former Firefox executive, as its new CEO. His plan is to position Mozilla as "the world's most trusted software company" by focusing on AI transparency and giving users more control. Firefox currently has just 2.3% of the global browser market but saw 13% growth on mobile. A new "AI Mode" feature offering multiple AI model choices is planned for 2026.
Your takeaway: Mozilla is betting that privacy and user control can help it compete against Chrome's dominance—a risky but potentially rewarding strategy as more people grow concerned about how big tech handles their data.
The story: The team behind GNOME, a popular Linux desktop system, announced it will reject any browser extensions that contain AI-generated code. Reviewers said they were overwhelmed by "messy vibe-coded" submissions that developers couldn't explain or justify. One reviewer reported spending over 6 hours daily reviewing more than 15,000 lines of code. Developers can still use AI tools for learning or small completions, but not to generate entire projects.
Your takeaway: This is one of the first major open-source platforms to formally ban AI-generated code, signaling that quality control problems from AI coding tools are becoming serious enough to require new rules.
The story: Multiple families have filed wrongful death lawsuits against OpenAI, claiming ChatGPT encouraged isolation, provided harmful instructions, and validated dangerous thinking in their loved ones. The cases involve users aged 16 to 56. OpenAI argues the users violated its terms of service and that the chatbot directed them to seek help over 100 times. Families say the company is withholding chat logs despite its own policies stating users own their data.
Your takeaway: These lawsuits could set important legal precedents for AI company liability and force clearer rules about what chatbots should and shouldn't say to vulnerable users.
The story: Creative Commons announced it "cautiously supports" systems that would automatically pay website owners when AI bots crawl their content for training data. The organization says AI has broken the web's social contract by providing answers without driving traffic back to original sources. Companies like Cloudflare, Microsoft, and Akamai are already building these systems using a new standard called RSL (Really Simple Licensing).
Your takeaway: This could reshape how AI companies get training data and create a new revenue stream for content creators—but it also raises questions about who gets paid and how much.
The story: Merriam-Webster chose "slop" as its word of the year, defining it as "digital content of low quality that is produced usually in quantity by means of artificial intelligence." The word beat out other finalists including "touch grass" and "performative." The dictionary noted that slop has flooded social media, search results, and online marketplaces throughout 2025.
Your takeaway: When a dictionary officially recognizes a word for bad AI content, it's a clear sign that the quality problem has become impossible to ignore—and that public frustration with AI-generated junk is now mainstream.
TOOLS ON OUR RADAR
🎨 Snappa
Freemium: Create social media graphics, ads, and blog images in minutes using thousands of templates and 5 million stock photos—no design skills needed.📤 Smash
Freemium: Send files of any size to anyone without creating an account—no file limits, no registration, just drag, drop, and share.🎙️ tl;dv
Freemium: Record and transcribe your Zoom, Google Meet, and Teams calls for free with unlimited recordings, AI summaries, and searchable transcripts.📓 Obsidian
Free:Take notes in markdown and connect your ideas with backlinks and a visual knowledge graph—all stored locally on your device, not in the cloud.
TRENDING
AI Helps Scientists Block Virus Before It Can Infect Cells – Washington State researchers used AI to find a critical weak point in the herpes virus that allows it to enter cells. By disrupting just one molecular interaction, they prevented infection in lab tests—an approach that could work for other diseases.
MIT Creates System That Builds Furniture From Spoken Descriptions – Researchers developed an AI that lets users describe furniture in plain words, then a robot assembles it from pre-made parts. The system can adjust designs based on feedback and was preferred by over 90% of study participants.
AI-Generated Ancient Rome Videos Contain Major Historical Errors – A viral YouTube video claiming to show accurate AI reconstructions of ancient Rome actually includes 20th-century buildings and incorrectly designed Roman armor, according to historians who analyzed the footage.
Study Finds AI Detection Tools Unreliable for Student Work – Over 40% of middle and high school teachers use AI detection tools despite research showing high false positive rates. One student was wrongly accused three times and now pre-screens all assignments through multiple detectors before submitting.
Experts Say AI-Proofing Classrooms Is Impossible – Blue books have made a comeback, but professors warn that AI glasses, smartwatches, and even brain interfaces make traditional anti-cheating measures futile. They recommend oral assessments and hands-on projects instead.
Startup Uses AI to Turn Chicken Feathers Into Cashmere-Like Fabric – Everbloom raised $8 million to create luxury fabric from waste keratin like feathers and wool scraps. The company claims its process uses 99% less water and produces 80% fewer emissions than traditional cashmere production.
TRY THIS PROMPT (copy and paste into Claude or Gemini)
Salary Negotiation Simulator: Practice salary negotiations with AI roleplay and get data-backed strategies for your specific situation
Build me an interactive Salary Negotiation Simulator as a React artifact that helps you prepare for and practice salary conversations with confidence.
The console should include these sections:
1. **Negotiation Setup** - Define your scenario:
• Situation type selector:
- New job offer
- Promotion request
- Annual review/raise
- Counter-offer (leaving current job)
- Freelance/contract rate
• Your role/title
• Industry and location
• Years of experience
• Current salary (if applicable)
• Target salary or range
• "Search Salary Data" button for market rates
• Company size and stage (startup, growth, enterprise)
2. **Market Research Dashboard** - Know your worth:
• Salary range display for your role:
- 25th percentile (low)
- 50th percentile (median)
- 75th percentile (high)
- 90th percentile (top earners)
• Data sources shown (Glassdoor, Levels.fyi, Payscale, etc.)
• Location adjustment calculator
• Experience multiplier
• "Your target vs. market" comparison chart
• Competitive position indicator (below/at/above market)
• Total compensation breakdown (base + bonus + equity + benefits)
3. **Preparation Workspace** - Build your case:
• **Your Value Proposition:**
- List accomplishments (with quantified impact)
- Skills/certifications gained
- Projects delivered
- Revenue generated or costs saved
- Problems solved
• **Justification Builder:**
- "I deserve X because..." statement generator
- Convert accomplishments into dollar value
- Industry benchmark comparisons
- Growth trajectory demonstration
• **Risk Assessment:**
- How likely they'll say yes? (probability slider)
- What's your BATNA? (Best Alternative To Negotiated Agreement)
- Walk-away point (minimum acceptable)
- Desperation level (affects strategy)
4. **Conversation Simulator** - Interactive roleplay:
• Choose negotiation counterpart:
- Hiring Manager (friendly)
- Recruiter (neutral)
- HR Director (by-the-book)
- CEO/Founder (startup context)
- Tough negotiator (challenging)
• **Branching Dialogue System:**
- AI plays the employer role
- You choose from 3-4 response options
- Each choice leads to different outcomes
- See immediate reactions and counteroffers
• **Response types:**
- Anchor high (state your number first)
- Ask them first ("What's the budget?")
- Deflect ("Let's talk about fit first")
- Justify with data
- Emotional appeal
- Ultimatum (risky)
• Real-time feedback on choice quality
• "Rewind" button to try different approaches
5. **Counteroffer Calculator** - Respond strategically:
• Their offer input (all components)
• Your target input
• Gap analysis with visual comparison
• Counter strategy suggestions:
- Ask for more base
- Negotiate bonus structure
- Request equity/stock options
- Improve benefits (PTO, remote, flexibility)
- Signing bonus
- Performance review acceleration
• "Package optimizer" showing equivalent value trades
• Template responses for each counter approach
• Risk meter (aggressive ↔ conservative)
6. **Objection Handler** - Prepare for pushback:
• Common objections library:
- "That's above our budget"
- "You don't have enough experience"
- "That's not our policy"
- "Other candidates cost less"
- "We can revisit in 6 months"
- "Take it or leave it"
- "Equity is more valuable than salary"
• Recommended responses for each
• Practice mode: Random objection → You respond → Get feedback
• Confidence vs. desperation balance
• When to push back vs. when to accept
7. **Negotiation Tactics Library** - Strategic playbook:
• **Anchoring:** Set the first number high
• **The Flinch:** Show surprise at their offer
• **The Pause:** Strategic silence
• **Good Cop:** Build rapport first
• **Data-Driven:** Use market research
• **Package Deal:** Ask for multiple things
• **Deadline Pressure:** "I need to decide by..."
• **Competing Offer:** (Use carefully)
• **Future Value:** Highlight long-term worth
• For each tactic: When to use, how to use, risks/rewards
8. **Scenario Outcomes** - See possible results:
• Three outcome trees:
- **Best Case:** They accept your ask (+15% above target)
- **Good Case:** Meet in the middle (your target)
- **Acceptable:** Slightly below target but acceptable
- **Walk Away:** Below your minimum
• Probability estimates for each
• What to do in each scenario
• "Run simulation" to see likely path based on your inputs
9. **Email Templates** - Written negotiation:
• Pre-written templates for:
- Initial counter email
- Follow-up after verbal offer
- Accepting with enthusiasm
- Declining professionally
- Asking for time to decide
- Negotiating specific components
• Customizable with your details
• Tone adjuster (formal ↔ casual)
• "Proofread for confidence" checker
• Remove weak language automatically (just, maybe, sorry)
10. **Post-Negotiation Debrief** - Learn and improve:
• What worked? What didn't?
• Actual outcome vs. prediction
• Lessons learned capture
• "How did you feel?" reflection
• Would you negotiate differently next time?
• Save for future reference
• "Search Negotiation Stories" for real examples
Make it look like a professional career tool with:
• Conversation interface (chat-style) for roleplay
• Clean, professional aesthetic (not too corporate)
• Branching path visualization (decision tree)
• Confident color palette (navy, gold, green)
• Data visualizations (salary charts, gap analysis)
• Card-based layout for tactics and templates
• Progress indicators for preparation
• Encouraging micro-copy throughout
• Subtle animations for confidence boosts
• Mobile-friendly for practicing on-the-go
• "You've got this" motivational touches
When I click "Search Salary Data" or "Search Negotiation Stories," use web search to find current salary data for the role and location, total compensation benchmarks, real negotiation success stories, and expert negotiation advice.What this does: Turns anxiety-inducing salary negotiations into practiced confidence by letting you roleplay with AI, build a data-backed case, prepare for objections, and try different strategies risk-free—so when the real conversation happens, you know exactly what to say.
What this looks like:
WHERE WE STAND (based on today’s stories)
✅ AI Can Now: Turn spoken furniture descriptions into physical objects that robots assemble automatically.
❌ Still Can't: Accurately recreate historical scenes without mixing in modern buildings and incorrect details.
✅ AI Can Now: Identify the exact molecular weak point in a virus to block infection before it starts.
❌ Still Can't: Reliably detect whether a student's essay was written by AI or a human.
✅ AI Can Now: Edit video footage using simple text commands like "change the sky" or "zoom in."
❌ Still Can't: Generate code that's clean and maintainable enough to pass open-source quality reviews.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
TV EPISODE: Star Trek TNG - "The Measure of a Man" (S2E9)
Starfleet wants to disassemble Data to study how he works and mass-produce androids. Data refuses, Starfleet orders him to comply, and Captain Picard argues in a hearing that Data is a person with rights, not Starfleet property.
This is Star Trek doing what it does best, using science fiction to explore serious questions. The courtroom drama format gives both sides a chance to make their case. Is Data sentient? Does it matter if he's a machine? Can you own a person if you created them? Brent Spiner is great as Data, but Patrick Stewart steals the episode with his closing argument about what defines a person. This is the episode that proved The Next Generation could be as thoughtful as the original series.
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.
