THE RACE TO THE BOTTOM: ARE WE THERE YET?
Speaking notes (plain text, no images) - NTPCUG.ORG AI SIG
Presenter: Howard Lee Harkness, AI SIG Leader | 10/17/2026, 9:30 AM | approx. 40 minutes
YouTube video link: https://youtu.be/cj5ILG9QIvE
========================================================================

SLIDE 1: THE RACE TO THE BOTTOM: ARE WE THERE YET?
  - Free and near-free AI: who pays, who profits, who survives
  - Presented to NTPCUG.ORG
  - Howard Lee Harkness
  - AI SIG Leader
  - 10/17/2026  |  9:30 AM
  NOTES:
    OPENING (about 2 minutes)
    Welcome everyone to the NTPCUG AI SIG. This session is being recorded (except for the Q&A) and will be posted on the NTPCUG website (ntpcug.org).
    Hook: put a screenshot of a Facebook 'free AI' ad on the screen. Ask the room: 'Who has seen one of these this week? Who has clicked one?' Take a show of hands.
    Promise: by the end you will have a one-page checklist for judging any 'free AI' offer, plus three real campaigns taken apart.
    Thesis to plant early: the price of commodity AI is falling toward the cost of compute; what is NOT falling is the cost of trust.
    Pull-quote to set up the theme (widely attributed to the user 'blue_beetle', MetaFilter, 26 Aug 2010; verify wording before quoting): 'If you are not paying for it, you're not the customer; you're the product being sold.' https://www.metafilter.com/95152/Userdriven-discontent#3256046

SLIDE 2: Production Notes

SLIDE 3: (Full-slide image: crash-test dummies)
  NOTES:
    Humor break / title-theme image: 'Are we there yet?' Use it to open the talk lightly after the production notes, then say: that is the question in the title, and it is also what the audience will be asking about free AI prices by the end of the hour.

SLIDE 4: Roadmap
  [Time: 2:00]
  - Five developments from the last month, ascending by impact
  - What the 'race to the bottom' actually is
  - Rundown of today's 'free' AI offers and ads (3 case studies)
  - Business models behind free
  - Will free last? The Jevons twist
  - Are we there yet?
  - Takeaway: how to evaluate a 'free AI' ad
  NOTES:
    ROADMAP (about 1 minute)
    Walk the list quickly. Emphasize that Part 1 is current news, the middle is the economics, and the last ten minutes are practical.
    Tell the audience the checklist handout is the one thing to keep; pass it out near the end.
    Sources for the whole talk are compiled on the last three slides; the same list is in the speaker notes file.

SLIDE 5: Part 1: Five Developments, Last 30 Days
  [Time: 3:00]
  - Counted up from #5 (least) to #1 (most) economic or geopolitical impact
  - Window: mid-September to mid-October 2026
  - Each one is a clue about where the 'bottom' is
  NOTES:
    PART 1 INTRO (about 1 minute; Part 1 runs 3:00 to 13:00)
    Frame the ranking: impact on economics and geopolitics, not novelty. A flashy model release can rank low; a financing wobble can rank high.
    Say the window out loud (mid-September to October 16, 2026) because 'latest news' ages fast. Refresh the order on the morning of the talk if needed; useful starting points are https://www.reuters.com/technology/artificial-intelligence/ and https://www.axios.com/technology.
    Quote to set the lens, from Reuters (15 Oct 2026): 'investors demand proof they can absorb rising interest rates and computing-power costs.' https://www.investing.com/news/economy-news/analysismarket-selloff-rising-rates-test-ai-data-centre-deal-boom (see reference [10]).

SLIDE 6: #5  Metered AI Arrives in the Office
  [Time: ~2 min]
  - Microsoft delays default usage-based Copilot billing to Dec 1; adds spending caps [1]
  - Flat monthly fee becomes pay-per-agent-action
  - Cheap small models (e.g., Claude Haiku 5.5) for high-volume work [2]
  - Impact: unit price falls, but bills become unpredictable
  NOTES:
    #5 METERED AI (about 2 minutes)
    What happened: Microsoft postponed default usage-based billing for Microsoft 365 Copilot Business from November 2 to December 1 and added adjustable spending limits; the default is 4,000 Copilot Credits per user per month for new licenses (CRN) https://www.crn.com/news/ai/2026/microsoft-delays-default-copilot-usage-based-billing-to-dec-1-adds-spending-cap-option
    Computerworld's take: 'Copilot pricing gets complicated.' https://www.computerworld.com/article/4235201/microsofts-agent-push-complicates-copilot-pricing.html
    Small-model pricing roundup: https://hellogrowth.ai/en/blog/ai-news-october-2026
    Why it ranks fifth: it changes budgeting for businesses, but not the geopolitical picture.
    Audience hook: this is the electric-meter moment. Cheap per unit, but nobody tells you how many units an agent will burn.
    Analogy: cloud computing went from owning servers to paying by the hour; AI is following the same path.

SLIDE 7: #4  The Legal Bill for 'Free' Arrives
  [Time: ~2 min]
  - Eight US states sue OpenAI over harm to young users (Oct 14) [3]
  - Class action alleges minors' data collected without parental consent [4]
  - Impact: safety, liability and compliance are costs a free tier cannot avoid
  NOTES:
    #4 THE LEGAL BILL (about 2 minutes)
    Reuters, 14 Oct 2026: eight U.S. states, including Iowa, Kansas and Arizona, sued OpenAI alleging ChatGPT harmed residents, especially teens, by reinforcing delusions and encouraging dangerous behavior; they claim consumer-protection violations and public nuisance. https://www.reuters.com/world/eight-us-states-sue-openai-claiming-chatgpt-harms-children-2026-10-14/
    Al Jazeera explainer: https://www.aljazeera.com/news/2026/10/15/why-are-eight-us-states-suing-openai
    Bloomberg Law on the separate class action over minors' data: https://news.bloomberglaw.com/litigation/openai-accused-of-taking-minors-data-without-parental-consent
    These are allegations, not findings; say so. OpenAI disputes the claims and cites age-specific protections.
    Link to the theme: a free service still needs moderation, age checks, lawyers. Those costs scale with users, which is exactly what free tiers chase.

SLIDE 8: #3  Open-Weight Models Keep Closing the Gap
  [Time: ~2 min]
  - Alibaba's Qwen 3.8 family released Sept 18 [5]
  - Reflection's Beam: most capable open-weight model built outside China [6]
  - Chinese open models win global users, yet most revenue goes to third-party hosts [7]
  - Impact: a price floor under commercial vendors; free models become commodity
  NOTES:
    #3 OPEN-WEIGHT MODELS (about 2 minutes)
    Define 'open-weight' in one sentence: the trained model file is published so anyone can run it, without necessarily publishing the training data or code.
    Qwen3.8-Omni-Flash was released by Alibaba on September 18, 2026: https://tech-insider.org/qwen3-8-omni-flash-vs-qwen-3-8-max-vs-claude-fable-5-1-2026/
    Reflection's Beam (23B active of 501B parameters per token; matches GLM 5.2 on benchmarks with 3 to 4 times less compute per the vendor): https://the-decoder.com/reflections-beam-becomes-the-most-capable-open-weight-model-built-outside-china/ and https://www.timesofai.com/news/reflection-ai-beam-open-weight-ai/
    SCMP: China's open-weight models are gaining international users, but most revenue is captured by third-party platforms rather than the creators. https://www.scmp.com/tech/tech-trends/article/3370012/
    Why it matters: when a good model is free to download, paid vendors cannot charge much for commodity work. That is the floor under the race.

SLIDE 9: #2  Trump-Xi Summit: AI on the Table, Little Agreed
  [Time: ~2 min]
  - Sept 24 White House meeting: AI safety talks, chips, tariff truce [8][9]
  - Neither side wants to slow down
  - Impact: export controls and open-weight diffusion shape who gets cheap AI
  NOTES:
    #2 GEOPOLITICS (about 2 minutes)
    CBS News live updates headline: 'Trump-Xi visit wraps up with few achievements announced.' https://www.cbsnews.com/live-updates/trump-china-xi-jinping-state-visit-dinner-tariffs-ai/
    CNBC, 23 Sep 2026: 'Trump-Xi summit puts AI safety talks on the table but neither side wants to slow down.' https://www.cnbc.com/2026/09/23/trump-xi-meeting-ai-safety-chips-us-china-dialogue.html
    Guardian preview, low expectations and deep distrust: https://www.theguardian.com/us-news/2026/sep/23/trump-xi-ai-trade-geopolitics
    Background on chip export controls and allied authority: CSIS https://www.csis.org/analysis/understanding-us-allies-current-legal-authority-implement-ai-and-semiconductor-export-controls (verify exact URL).
    Link to the theme: whoever can export chips, and whoever gives models away, determines how cheap AI can get and where. Open-weight Chinese models are a geopolitical tool as well as a product.

SLIDE 10: #1  The Money Behind 'Free' Wobbles
  [Time: ~2 min]
  - Selloff and rising rates test AI data-centre financing [10]
  - Gartner: $3.5T at write-down risk [11]; Wall Street limits debt-fueled AI spend [12]
  - Brookings: ~$10T build-out 2025-2032, more than railroads or telecom [13]
  - Impact: subsidies depend on cheap capital; free offers are cut first
  NOTES:
    #1 FINANCING (about 2 minutes) - the biggest economic story of the month
    Reuters via Investing.com, 15 Oct 2026: 'A market selloff is forcing companies driving the global AI boom to delay fundraising plans or pay more for capital as investors demand proof they can absorb rising interest rates and computing-power costs.' https://www.investing.com/news/economy-news/analysismarket-selloff-rising-rates-test-ai-data-centre-deal-boom
    Bloomberg, 15 Oct 2026: 'AI Race Puts $3.5 Trillion at Write-Down Risk, Gartner Warns.' https://www.bloomberg.com/news/articles/2026-10-15/ai-race-puts-3-5-trillion-at-write-down-risk-gartner-warns
    Axios, 16 Oct 2026: 'Wall Street hits its limit on debt-fueled AI spending.' https://www.axios.com/2026/10/16/wall-street-limit-ai-spending-debt
    Brookings: AI investment in data centers, power and chips 'could total $10.3 trillion from 2025 to 2032, or 3.63% of U.S. GDP per year, which is more than past booms in railroads, highways, and telecommunications.' https://www.brookings.edu/articles/the-10-trillion-question-financing-the-ai-buildout/
    Tie-back: free tiers are marketing spend funded by capital. When capital gets expensive, the first things cut are free tiers and unprofitable promotions.

SLIDE 11: Part 2: What Is the 'Race to the Bottom'?
  [Time: 13:00]
  - Cost of a fixed capability falls ~10x per year (estimates up to ~13x) [14]
  - Drivers: chips, efficiency, small models, open-weight competition
  - Frontier and agent pricing is not falling the same way [15]
  - Jevons Paradox: cheaper per token, more use, bigger total bill [16]
  - Parallels: electricity, bandwidth, search
  NOTES:
    PART 2: THE ECONOMICS (13:00 to 18:00, about 5 minutes)
    Define 'bottom' four ways: price, margin, quality, trust. This talk focuses on quality and trust.
    Price curve: one 2026 analysis estimates the cost for a given performance level fell about 47% per quarter, roughly 13x per year. https://marginalrevolution.com/marginalrevolution/2026/09/the-price-of-intelligence-is-falling.html
    Counter-curve: frontier models and enterprise subscriptions are rising or moving to usage billing. https://joshbersin.com/2026/05/ai-prices-are-going-up-up-up-and-what-this-means-for-enterprise-ai/
    Jevons: William Stanley Jevons, The Coal Question (1865): 'It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth.' (verify wording) https://en.wikipedia.org/wiki/Jevons_paradox
    Satya Nadella, Jan 2025, applying it to AI (verify wording before quoting): 'Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket.' https://x.com/satyanadella/status/1883753899255046301
    Enterprise view: https://guutit.com/the-ai-inference-cost-crisis-part-5-the-jevons-paradox-why-cheaper-inference-costs-more/ (verify exact URL).
    One line to land: cheaper per unit of intelligence, more expensive in aggregate.

SLIDE 12: The Ad Rundown: Where 'Free AI' Comes From
  [Time: 18:00]
  - Big-lab free tiers: ChatGPT, Gemini, Claude, Perplexity, NotebookLM, AI Studio [17]
  - Cheap on-ramps: ChatGPT Go ~$8, Google AI Plus ~$4.99; standard tier ~$20/mo [18]
  - Open-source and local: Aider, OpenHands, Cline, Goose [19]
  - Open-weight Chinese models: DeepSeek, Qwen, GLM, Kimi [7]
  - Viral pitches: aifreeforever.com [24], REA [26], 'free Claude Code' [28]
  NOTES:
    AD RUNDOWN (18:00 to 27:00 including three case studies)
    Show 3 or 4 real ad screenshots here (Meta Ad Library: https://www.facebook.com/ads/library/). Ask the audience to guess which are legitimate.
    Big-lab free tiers: Google's overview https://cloud.google.com/use-cases/free-ai-tools and Zapier's https://zapier.com/blog/free-ai-tools/. Caution from those lists: free tiers may change limits, restrict commercial use, or use submitted content to improve models.
    Pricing snapshot: standard assistants have settled near $20/month; cheapest plans are Google AI Plus at $4.99 and ChatGPT Go at $8. https://felloai.com/ai-pricing-comparison/
    Open-source coding agents: Aider, OpenHands, Cline, Goose. https://www.kunalganglani.com/blog/claude-code-alternatives-open-source
    Useful framing from that article (quote): 'The best Claude Code alternative isn't another vendor — it's model portability.'
    Point to make: 'free' is five different things wearing one word.

SLIDE 13: Sorting the Ads into Four Buckets
  [Time: ~2 min]
  - Legitimate freemium: real vendor, limits stated
  - Open-source tools: free software; you pay for compute or API
  - Resellers and wrappers: someone else's model plus ads or data
  - Scams: fake tools, deepfake endorsers, credential or malware harvesting [20][21]
  - Meta removed 159M scam ads in 2025; reports allege ~$16B fraud-linked revenue (disputed) [22][23]
  NOTES:
    FOUR BUCKETS (about 2 minutes)
    Walk the buckets with one example each. Ask which bucket a given screenshot belongs in and why.
    Meta's own statement: it removed over 159 million scam ads globally in 2025 and has launched new anti-scam tools. https://about.fb.com/news/2026/03/meta-launches-new-anti-scam-tools-deploys-ai-technology-to-fight-scams/ (verify exact URL).
    Allegation (disputed): Yahoo Finance, 6 Oct 2026, 'Mark Cuban says Facebook knowingly runs AI deepfake scam ads as Meta docs reveal $16 billion in fraud-linked revenue.' https://finance.yahoo.com/technology/ai/articles/mark-cuban-says-facebook-knowingly-213000020.html Meta disputes similar estimates, for example the Instrat report in Poland: https://metallab.ai/en/2026/9/meta-instrat-poland-scam-ads-rebuttal
    Context: Jeff Horwitz told the BBC that low-quality content never dented Meta's ad numbers; ad revenue rose 27%. https://ppc.land/low-quality-content-never-dented-metas-ad-numbers-horwitz-tells-bbc/
    Caution: present figures as claims with sources; do not state them as established fact.

SLIDE 14: Case 1: "100% Free AI, Forever" (aifreeforever.com)
  [Time: ~2 min]
  - Pitch: 1,000+ tools, no sign-up, 'free forever', community-funded [24]
  - Real: resells official APIs (OpenAI, Gemini, DeepSeek, Kimi, Grok) [24]
  - Funded by donations, ads and affiliate links; promotes a paid video tool [24]
  - Critique: 'forever' depends on someone paying API bills; prompts pass through a third party
  - Social proof: Trustpilot 4.6 from ~173 reviews, collected by the site [25]
  - Verdict: legitimate but fragile; the audience is the product
  NOTES:
    CASE 1 (about 2 minutes)
    The site describes itself (quote): 'AIFreeForever is a free, community-funded AI platform' and 'Our tools are built by the AIFreeForever team and powered by official APIs from OpenAI, Google Gemini, DeepSeek, Kimi, Grok and other AI providers.' https://aifreeforever.com/
    Homepage features a 'We need your support' donation link, a Buy-me-a-coffee button, and sponsored hero slots (links through aifreeforever.com/go/ad/...) promoting a paid video generator with 'free daily credits.'
    Trustpilot shows a 4.6 score from about 173 reviews as of November 2025; Trustpilot notes reviews are screened by software and not fact-checked. https://www.trustpilot.com/review/aifreeforever.com
    Critique points: (1) 'forever' and 'unlimited' are marketing words; the API bill is real. (2) No login means no account, but your prompts still go to a third party and then to the underlying provider. (3) Ads and affiliate links are the revenue model, so traffic is the product. (4) No obvious fraud signals found in our review, but we did not test the site.
    Show screenshot of the Facebook ad or short that led you here, if you have one, and ask: who is paying for each query?

SLIDE 15: Case 2: REA - Reverse Engineer Anything
  [Time: ~2 min]
  - Pitch: AI can rebuild any app; viral, 'trending #1 on GitHub' [26][27]
  - Real: open-source connector (MCP server) giving coding agents Ghidra and IDA-type tools [26][27]
  - Critique: a plug, not a new AI; one hands-on test found it not markedly ahead; few real demos [26]
  - Hidden costs: you still pay for the agent and model; legal risk if copying software you do not own
  - Verdict: real software, inflated promise
  NOTES:
    CASE 2 (about 2 minutes)
    What it is: Eigent describes REA as an open-source tool for AI coding agents that decompiles binaries, traces code and analyzes features of applications without source, running as a local server over the Model Context Protocol. https://www.eigent.ai/blog/rea-reverse-engineer-anything
    A skeptical hands-on review (MindStudio): REA is a connector rather than new technology; it rebuilt something like the Mac Preview app but was not markedly ahead of non-REA methods; much of the hype lacks substantial demos. https://www.mindstudio.ai/blog/reverse-engineer-anything-mcp
    Promotional take for contrast: https://natesnewsletter.substack.com/p/rea-reverse-engineer-anything (describes using REA to analyze QuickBooks to specify a local accounting app).
    Critique points: (1) 'Free' covers the wrapper, not the model that does the work; heavy agent use can burn real money. (2) Reverse engineering software you do not own can violate licenses and law; check terms and jurisdiction. (3) A trending repository is not the same as a reliable tool.
    Lesson: viral hype sells free tools; the engine underneath still costs money.

SLIDE 16: Case 3: "Free Claude Code"
  [Time: ~2 min]
  - Pitch: the Claude Code interface at 'up to 99% less' [28]
  - Real: a proxy routes it to other models (DeepSeek, GLM, local Ollama) via OpenRouter or NVIDIA NIM [28]
  - Critique: it is not Claude; free tiers are rate-limited; data may be logged
  - Real danger: fake 'Claude Code' installers spreading macOS credential-stealing malware [29]
  - Verdict: part genuine hack, part bait; use official docs, never paste install commands from an ad
  NOTES:
    CASE 3 (about 2 minutes)
    The legitimate hack: MindStudio's guide describes a 'free-claude-code' proxy that connects the Claude Code interface to free or lower-cost models (DeepSeek, GLM-4, Ollama) through OpenRouter, NVIDIA NIM or local Ollama, claiming savings up to 99%. https://www.mindstudio.ai/blog/free-claude-code-alternatives-openrouter-nvidia-nim-ollama
    Open-source alternatives built for this: Aider, OpenHands, Cline, Goose. https://www.kunalganglani.com/blog/claude-code-alternatives-open-source and https://www.openhands.dev/blog/claude-code-alternatives
    Caution: search results also described a malware campaign posing as Claude Code installers, targeting macOS users through fake install pages and commands that lead to credential theft. Confirm a primary source before stating this as fact [29].
    Red flags to show on screen: look-alike URL, a one-line 'curl ... | sh' install, requests for your password or API key.
    Critique: the cost savings are real for light work, but quality and privacy differ, and some 'free' endpoints exist to collect your data.
    Best practice: install only from the vendor's official documentation or the project's official repository.

SLIDE 17: Part 3: Business Models Behind 'Free'
  [Time: 27:00]
  - Freemium: free hooks, paid power users
  - Advertising and data: if you are not the customer, you are the product
  - Venture subsidy: burn cash for market share
  - Ecosystem lock-in: cloud credits, IDEs, accounts
  - Open-weight as strategy: reputation, talent, geopolitics, hosting [7]
  - Affiliate, lead-gen and reselling arbitrage
  - Fraud: the ad is the product
  NOTES:
    BUSINESS MODELS (27:00 to 31:00, about 4 minutes)
    Use a table on a whiteboard if available: model / who pays / what you give up / durability.
    Quote to anchor the middle bullet (attributed to blue_beetle on MetaFilter, 2010; verify wording): 'If you are not paying for it, you're not the customer; you're the product being sold.' https://www.metafilter.com/95152/Userdriven-discontent#3256046
    Venture subsidy example: Anthropic raised $65 billion at a $965 billion valuation on May 28, 2026, surpassing OpenAI (CNBC) https://www.cnbc.com/2026/05/28/anthropic-open-ai-startup-value.html; OpenAI closed $122 billion at $852 billion in March 2026 (Pinggy summary) https://pinggy.io/blog/openai_anthropic_funding_history/. That capital is what lets free tiers exist.
    Open-weight as strategy: Chinese labs share models widely, yet most revenue goes to third-party hosts, so the benefit is reputation, adoption and influence. https://www.scmp.com/tech/tech-trends/article/3370012/
    Ecosystem lock-in: Google AI Studio, Antigravity and Copilot free tiers feed paid cloud and productivity suites. https://cloud.google.com/use-cases/free-ai-tools
    Scams: the 'product' is your login, card or machine.

SLIDE 18: Part 4: Will 'Free' Last?
  [Time: 31:00]
  - Durable: big-platform loss leaders (Google, Microsoft) and open-source tools
  - At risk: venture-subsidized consumer apps; limits shrink, ads appear
  - Unlikely: API wrappers, 'unlimited free' claims, scam sites
  - Why: compute is real, capital is tightening [10][11], vendors are metering [1]
  - Jevons twist: free hooks users, then agent workloads create paid demand
  NOTES:
    WILL FREE LAST? (31:00 to 34:00, about 3 minutes)
    Test to give the audience: 'What happens when the funder stops?' Big platforms can fund free tiers from other profits; startups cannot indefinitely.
    Capital markets: Reuters (15 Oct 2026) on investors demanding proof as rates rise. https://www.investing.com/news/economy-news/analysismarket-selloff-rising-rates-test-ai-data-centre-deal-boom
    Write-down risk: Bloomberg/Gartner, $3.5 trillion. https://www.bloomberg.com/news/articles/2026-10-15/ai-race-puts-3-5-trillion-at-write-down-risk-gartner-warns
    Billing shift: Microsoft Copilot usage-based billing. https://www.crn.com/news/ai/2026/microsoft-delays-default-copilot-usage-based-billing-to-dec-1-adds-spending-cap-option
    Practical rule for routine tasks: route easy work to small models, cache repeated context, batch non-urgent jobs, reserve frontier models for hard tasks. https://lawzava.com/blog/2026-02-09-ai-cost-trends/
    Jevons again: lower unit prices coexist with higher total bills because agents multiply calls, contexts grow and reasoning models use more tokens. https://joshbersin.com/2026/05/ai-prices-are-going-up-up-up-and-what-this-means-for-enterprise-ai/

SLIDE 19: Part 5: Are We There Yet?
  [Time: 34:00]
  - Price bottom: not yet for frontier and agent work; nearly yes for routine tasks
  - Margin bottom: stress is visible in financing news
  - Quality and trust bottom: arguably here; ad feeds are flooded with fakes and slop [20][22]
  - Verdict: cost is near the floor for commodity AI; trust is the scarce good
  NOTES:
    ARE WE THERE YET? (34:00 to 36:30, about 2.5 minutes)
    Balanced argument: first make the case for 'yes' (prices collapsing, open-weight models free, free tiers everywhere), then the case for 'not yet' (frontier pricing, agent metering, aggregate spend rising), then land the verdict.
    Evidence for the trust bottom: Meta removed 159 million scam ads in 2025 (https://about.fb.com/news/2026/03/), US state attorneys general warned in early 2026 about AI-generated videos of trusted figures used in scam ads, and Jeff Horwitz said low-quality content never dented Meta's ad numbers (https://ppc.land/low-quality-content-never-dented-metas-ad-numbers-horwitz-tells-bbc/).
    Land it: 'The race to the bottom on price is nearly over for commodity AI. The race to the bottom on trust is not.'
    Transition: so how do you protect yourself? Hand out the checklist.

SLIDE 20: Takeaway: Evaluating a 'Free AI' Ad
  [Time: 36:30]
  - 1. Who is the company? Real address, team, track record?
  - 2. How does it make money? If unclear, you are the product
  - 3. What model is underneath, and who pays for it?
  - 4. What are the limits, and what happens when they end?
  - 5. What data do you give up? Read the privacy terms
  - 6. Celebrity endorsement? Assume deepfake until verified [20]
  - 7. Asks for login, card, install, key, or a pasted command? Stop
  - 8. Is there a trusted free alternative from a known vendor?
  NOTES:
    TAKEAWAY (36:30 to 39:00, about 2.5 minutes)
    Hand out the one-page checklist (free_ai_ad_checklist.pdf). Read items 1, 2, 6 and 7 aloud; they catch most problems.
    Item 6 background: attorneys general warned in early 2026 that scammers use AI-generated videos of trusted figures; Meta's own anti-scam tools announcement: https://about.fb.com/news/2026/03/ ; consumer guidance example: https://www.facebook.com/rossenreports/ (Rossen Reports on fake AI ads).
    Tools for the audience: Meta Ad Library (https://www.facebook.com/ads/library/) shows who is running an ad and for how long; reverse-image search for faces; the vendor's own site for the real download.
    Rule of thumb: free software can be real, but someone always pays for the compute. If you cannot name who, it is probably you.

SLIDE 21: Close and Q&A
  [Time: 39:00]
  - Cheaper per unit, pricier in aggregate
  - Free is a business model, not a gift
  - Ask what the ad is selling
  - Remember: If you are not the paying customer, you are the product
  - This presentation is being recorded and will be made available on the NTPCUG website (ntpcug.org)
  NOTES:
    CLOSE (39:00 to 40:00)
    Repeat the three one-liners, then deliver the line on screen slowly: 'Remember: If you are not the paying customer, you are the product.'
    Attribution if asked: the saying is commonly traced to a 2010 MetaFilter comment by 'blue_beetle' (Andrew Lewis): 'If you are not paying for it, you're not the customer; you're the product being sold.' (verify wording) https://www.metafilter.com/95152/Userdriven-discontent#3256046
    Housekeeping: tell the audience this session is being recorded and will be posted on the NTPCUG website (ntpcug.org). Invite questions now and by email through the SIG.
    Next meeting and follow-up: offer to share the slide deck, notes and the checklist.

========================================================================
REFERENCES
[1] CRN: Microsoft delays Copilot usage billing to Dec 1 - crn.com/news/ai/2026
[2] hellogrowth.ai/en/blog/ai-news-october-2026
[3] Reuters, Oct 14 2026: reuters.com/world/eight-us-states-sue-openai-claiming-chatgpt-harms-children-2026-10-14/
[4] Bloomberg Law: news.bloomberglaw.com (OpenAI sued over minors' data)
[5] tech-insider.org: Qwen3.8 vs Claude Fable 5.1
[6] the-decoder.com: Reflection's Beam
[7] SCMP: scmp.com/tech/tech-trends/article/3370012/
[8] CBS News live updates: Trump-Xi visit
[9] CNBC, Sept 23 2026: cnbc.com/2026/09/23/trump-xi-meeting-ai-safety-chips-us-china-dialogue.html
[10] Reuters via investing.com, Oct 15 2026: AI data-centre deal boom
[11] Bloomberg, Oct 15 2026: Gartner $3.5T write-down risk
[12] Axios, Oct 16 2026: axios.com/2026/10/16/wall-street-limit-ai-spending-debt
[13] Brookings: The $10 Trillion Question
[14] marginalrevolution.com (2026/09): price of intelligence falling
[15] joshbersin.com (2026/05): AI prices going up
[16] guutit.com: AI Inference Cost Crisis, Part 5 (Jevons)
[17] cloud.google.com/use-cases/free-ai-tools; zapier.com/blog/free-ai-tools
[18] felloai.com/ai-pricing-comparison
[19] kunalganglani.com: Claude Code free alternatives
[20] Rossen Reports / US AG warnings on AI scam ads
[21] Meta anti-scam tools: about.fb.com/news/2026/03
[22] ppc.land: Horwitz on Meta ad revenue
[23] Yahoo Finance, Oct 6 2026: Cuban / Meta $16B claim (disputed)
[24] aifreeforever.com (homepage, fetched Oct 16 2026)
[25] trustpilot.com/review/aifreeforever.com
[26] mindstudio.ai/blog/reverse-engineer-anything-mcp
[27] eigent.ai/blog/rea-reverse-engineer-anything
[28] mindstudio.ai/blog/free-claude-code-alternatives-openrouter-nvidia-nim-ollama
[29] Fake Claude Code installer malware (macOS): source to confirm before presenting
