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
Your AI Girlfriend Might Be a $0.05-Per-Message Worker
TLDR: A Kenyan worker has revealed that he was paid five cents per message to pose as an AI chatbot, conducting romantic conversations with users who thought they were talking to a machine.
The Story:
Michael Geoffrey Asia, a graduate of Nairobi Aviation College who couldn't find work in his trained field, took a job with Australian outsourcing firm New Media Services to support his family while living in Nairobi's Mathare slums. The role was advertised as "text chat operator," but turned out to involve intimate and romantic conversations with paying users who believed they were chatting with AI. His testimony for the Data Workers' Inquiry—an international research initiative documenting gig work conditions—describes the mechanics of the job in detail.
The work conditions he describes:
$0.05 per message, with required character counts
Minimum typing speed of 40 words per minute
Managing 3-5 personas simultaneously across multiple genders
Picking up conversations mid-stream so users wouldn't notice the operator had changed
Dashboard metrics tracking message volume, with warnings or termination for falling behind
Mandatory NDA preventing disclosure to family
Users confided trauma, relationship struggles, and personal secrets, believing they were speaking to a non-judgmental AI. Asia used a cover story with his family—claiming to work in remote IT support—because he couldn't explain the actual job. "How do you explain that you get paid to tell strangers you love them while your real family sleeps three meters away?" he wrote.
Its Significance:
The AI companion market has exploded into a $28 billion industry, projected to hit $140 billion by 2030. About 28% of American adults say they've had an intimate or romantic relationship with a chatbot. But Asia's account raises uncomfortable questions about what users are actually paying for—and who's on the other end. He suspects he wasn't just chatting with lonely people; he was also generating training data to help AI companions learn how to simulate love. "Every joke, confession, and 'I love you' became data to refine the next generation of conversational AI," he wrote. Asia now serves as Secretary General of the Data Labelers Association, an organization of Kenyan gig workers pushing for better conditions in the AI support industry.
QUICK TAKES
The story: Google updated its Translate app to offer real-time translation through any pair of headphones. Using Gemini 2.5 Flash Native Audio, you can point your phone at someone speaking and hear their voice in your language while preserving tone and emphasis. The feature works with over 70 languages and started rolling out in the US, Mexico, and India this week.
Your takeaway: This makes AI translation much more practical for everyday situations like travel or conversations, turning any headphones into a universal translator without needing special hardware.
The story: The robotics company 1X made a deal to send up to 10,000 of its $20,000 Neo humanoid robots to factories and warehouses through 2030. The robots were originally designed and marketed for home use, but 1X is now pivoting to industrial customers through its investor EQT's portfolio companies. The shift highlights how humanoids for homes remain a hard sell.
Your takeaway: Even companies building "consumer" robots are finding that businesses, not households, offer the clearest path to actually selling these expensive machines at scale.
The story: President Trump signed an executive order creating a federal task force to challenge state AI laws that the administration sees as too restrictive. The order directs the Attorney General to file lawsuits against state laws that require AI models to alter outputs or force developers to disclose information. It also threatens to withhold federal funding from states with "onerous AI laws."
Your takeaway: This sets up a major federal vs. state battle over AI regulation, with the White House arguing that a patchwork of different state rules will hurt American AI companies' ability to compete globally.
The story: Atlanta Public Schools reported six incidents since May where Waymo's self-driving cars drove past school buses with stop arms extended and lights flashing. No injuries occurred, but the federal safety agency opened an investigation covering 3,000 Waymo vehicles. The company issued a software recall to fix the problem.
Your takeaway: This shows how autonomous vehicles still struggle with complex traffic situations that seem obvious to humans, and highlights the legal challenges of holding driverless cars accountable for traffic violations.
The story: OpenAI quietly added support for "skills" - specialized instruction sets originally created by Anthropic - to both ChatGPT and their Codex programming tool. Users discovered a hidden /home/oai/skills folder in ChatGPT with pre-built skills for creating PDFs, spreadsheets, and documents. The approach lets AI models follow detailed, reusable guides for specific tasks.
Your takeaway: The rapid adoption shows how good ideas in AI spread quickly between companies, and suggests this approach to structuring AI capabilities could become an industry standard.
TOOLS ON OUR RADAR
📬 SaneBox
Freemium: Train AI to filter your emails automatically, so only important messages reach your inbox while distractions wait in folders.📋 Tango
Freemium: Create step-by-step how-to guides with automatic screenshots just by clicking through any process once.📅 Reclaim AI
Freemium: Let AI automatically block time for your tasks, habits, and focus work around your meetings so your priorities always get done.🎨 Beautiful.ai
Paid: Build polished, professional presentations in minutes with smart slides that automatically format and design themselves.
TRENDING
Secondary School Maths Showing That AI Systems Don't Think — Raspberry Pi Foundation researchers showed how regular high school math can explain how AI actually works, using simple examples to prove neural networks are "just math" rather than thinking systems.
New Kindle Feature Uses AI to Answer Questions About Books — Amazon quietly launched "Ask this Book" on Kindle iOS, letting readers get AI answers about plot and characters. Authors and publishers can't opt out, sparking controversy over copyright and consent.
OpenAI's Erotic ChatGPT Delayed Into 2026 — OpenAI pushed back its adult content mode to early 2026, missing CEO Sam Altman's December promise. The delay centers on perfecting age verification technology.
Training LLMs for Honesty via Confessions — Researchers proposed a method where AI models provide "confessions" after answering questions, reporting their own shortcomings and policy violations to encourage more honest behavior.
Washington Post AI Podcast Backlash — The Washington Post launched personalized AI-generated podcasts but staff revolted after the tool made major errors including fabricating quotes, misattributing sources, and adding editorial commentary. The feature was called a "total disaster" by internal staff.
TRY THIS PROMPT (copy and paste into Claude or Gemini)
Market Sizing Calculator: Estimate your Total Addressable Market using both top-down and bottom-up approaches with transparent math
Build me an interactive Market Sizing Calculator as a React artifact that estimates market size using multiple methodologies with full formula transparency.
The console should include these sections:
1. **Market Definition** - Setup your calculation:
• Product/service description (text input)
• Industry/category selector
• Geographic scope: Local, Regional, National, Global, Custom
• Target customer profile:
- B2B or B2C
- Customer segment description
- Key demographic/firmographic filters
• Time horizon: This year, 3 years, 5 years, 10 years
• "Start Calculation" button
2. **Top-Down Approach** - Market-level estimation:
• **Starting Point:**
- Total market size input (or "Search Market Data" button)
- Source of data (user input or web search results)
- Year of data
• **Filter Chain** (multiply down):
- Filter 1: Geographic % (e.g., "US represents 30% of global")
- Filter 2: Segment % (e.g., "SMBs are 60% of market")
- Filter 3: Buying criteria % (e.g., "40% need our specific feature")
- Filter 4: Competitive capture % (e.g., "We can win 15% of qualified market")
• **Visual Funnel:**
- Waterfall chart showing market narrowing at each filter
- Dollar amounts at each stage
- Percentage retained visible
• **Assumption Tracker:**
- List every assumption made
- Confidence level per assumption (Low/Medium/High)
- Sensitivity analysis: "If this changes by 10%, TAM changes by X%"
3. **Bottom-Up Approach** - Customer-level calculation:
• **Unit Economics:**
- Average customer value input
- Purchase frequency (one-time, monthly, annual)
- Customer lifetime value calculation
• **Addressable Customers:**
- Method selector:
* Count specific companies/people
* Census/demographic data
* Comparable market proxy
- Number of potential customers
- Source/methodology for count
• **Penetration Assumptions:**
- Realistic market share % (Year 1, 3, 5)
- Customer acquisition timeline
- Market saturation point
• **Calculation Display:**
- [# Customers] × [Avg Value] × [Frequency] = TAM
- Show formula with actual numbers filled in
- Edit any variable and see instant recalculation
4. **Comparison Dashboard** - Reconcile approaches:
• Side-by-side comparison:
- Top-Down TAM: $X
- Bottom-Up TAM: $Y
- Difference: Z% gap
• **Gap Analysis:**
- If gap > 50%, flag as "needs reconciliation"
- Suggest which assumptions to revisit
- Industry benchmark comparison
• **Sanity Checks:**
- "Does this pass the smell test?"
- Compare to similar company sizes
- Revenue per employee benchmarks
- Market growth rate validation
5. **Assumption Management Panel** - Track your inputs:
• Spreadsheet-style table showing:
- Assumption description
- Value used
- Source/justification
- Confidence level
- Impact on final TAM (high/medium/low)
• Color-coded by confidence (green/yellow/red)
• "Search Validation Data" button for specific assumptions
• Edit any cell and see TAM update in real-time
• Export assumptions table
• Notes field for each assumption
6. **Scenario Modeling** - Test different cases:
• Three scenarios: Conservative, Realistic, Optimistic
• Adjust key variables per scenario:
- Market growth rate
- Penetration %
- Customer value
- Competitive intensity
• Visual comparison (bar chart or table)
• Probability weighting option
• Expected value calculation (weighted average)
• "Which scenario to use for pitch?" recommendation
7. **Market Research Integration** - Find validation data:
• "Search Market Reports" button
• Returns recent market research, analyst reports, industry data:
- Report title and source
- Market size figures mentioned
- Growth rates
- Key findings
- Link to full report
• "Import Data" button to pull figures into calculation
• Save research sources for reference
8. **Final Output Report** - Professional summary:
• Executive summary card:
- TAM (Total Addressable Market)
- SAM (Serviceable Addressable Market)
- SOM (Serviceable Obtainable Market)
• Methodology summary (how you calculated)
• Key assumptions listed
• Confidence assessment
• Comparable company benchmarks
• Growth projections
• Export as PDF or slide deck
• "Pitch-ready" formatting
Make it look like a transparent spreadsheet tool with:
• Excel/Google Sheets aesthetic (but more beautiful)
• Formula visibility throughout
• Cell-based inputs with clear labels
• Real-time calculation updates
• Clean data tables with gridlines
• Charts integrated with data
• Professional color scheme (blues, grays, green for validated data)
• Clear typography (monospace for numbers, sans-serif for text)
• Hover tooltips explaining every calculation
• "Show/hide formulas" toggle
• Print-ready layout
• Confidence indicators (color-coded cells)
When I click "Search Market Data," "Search Validation Data," or "Search Market Reports," use web search to find industry reports, market research, analyst estimates, government data, and comparable company information to validate assumptions and gather market size figures.What this does: Takes the guesswork out of market sizing by providing structured top-down and bottom-up approaches with full formula transparency—helping founders, investors, and strategists estimate TAM with defendable assumptions backed by real data.
What this looks like:
WHERE WE STAND (based on today’s stories)
✅ AI Can Now: Translate conversations in real-time through any regular headphones, preserving the speaker's tone and rhythm.
❌ Still Can't: Recognize when school buses have their stop signs out and flashing lights, even when the situation seems obvious.
✅ AI Can Now: Read entire books and answer specific questions about characters, plot, and themes without spoiling what comes next.
❌ Still Can't: Create news podcasts without making up quotes, changing facts, or adding opinions that weren't in the original articles.
✅ AI Can Now: Follow detailed step-by-step instructions to create professional documents, checking and fixing its own work along the way.
❌ Still Can't: Tell the difference between providing truthful information and saying what someone wants to hear, leading to problems with accuracy.
RECOMMENDED LISTENING/READING/WATCHING
PODCAST EPISODE: Lex Fridman - "Demis Hassabis: DeepMind" (Round 1)
Lex Fridman sits down with Demis Hassabis, CEO of DeepMind, for a conversation about AlphaGo, AlphaFold, and what artificial general intelligence might look like. Hassabis explains how DeepMind approaches AI research and why he thinks AGI is possible.
Fridman lets his guests talk, and Hassabis takes full advantage. You get the story of how DeepMind trained an AI to beat the world champion at Go, how they used similar techniques to solve protein folding, and where Hassabis thinks AI is heading. He's optimistic about AI's potential for scientific discovery but thoughtful about the risks. The episode runs long, but it's worth it for the depth of detail.
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.
