In partnership with
ChatGPT is now an ad platform, and the targeting options are broader than most users expect. An OpenAI engineer is facing federal charges over a hidden recording device, and Meta released a large open model. Underneath all of it runs one theme: AI is getting metered, and someone is paying the bill.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
TODAY IN AI
3 things that happened while you were busy
1. ChatGPT ads can target people by email, phone number and prompt. 404 Media bought ads inside ChatGPT to test the system. Its reporting shows advertisers can reach groups using an email address, a phone number or a phone identifier, and can also aim ads at what users type. ChatGPT has about one billion monthly users, and the full article sits behind a paywall. Read 404 Media's tests.
2. An OpenAI engineer is charged over a hidden recording device. Prosecutors say Abdulrahman Alfozan, 35, planted a device in an executive conference room, hidden in toilet paper and plugged into the back of a TV. OpenAI says its team found it on September 30 and used its own models to trace it. The company says it has no sign that model weights, code or customer data were taken. These are allegations, and a detention hearing is set for October 19. See the court details.
3. Meta releases Muse Spark Open, an 816B model you can download. It has 42B active parameters, a 1M token context window and a license with limited commercial terms. Meta says it runs on two Apple M5 Ultra Mac Studios and calls it the most capable US open-weight model, but those are Meta's claims, and its page lists no benchmark numbers. A full evaluation report is promised soon. See Meta's release notes.
TODAY'S TUTORIAL
Reconcile messy exports with the four-bucket method
When two records of the same money disagree, hunting for matches wastes hours. Powerdrill's method sorts every difference into four buckets instead. You can do it in Excel, and the steps below follow its guide. Read the full guide.
1. Clean first. Make reference formats match, so INV-0042 and INV42 read the same. Round amounts to two decimals, then build a match key from date, amount and cleaned reference.
2. Find where the gap sits. Compare totals by month, account or counterparty with SUMIFS before you look at single rows.
3. Sort into four buckets. Missing rows (XLOOKUP in both directions), duplicates (COUNTIFS), wrong amounts (match on the reference alone, then compare) and right payment in the wrong period (match within a date window).
4. Handle one-to-many cases. When one settlement covers 40 transactions, group the smaller rows and compare their total with the single row.
5. Check your work. Compare row counts and unmatched value, not only the total, because opposite errors can cancel out. Add a status column instead of deleting rows, and reconcile after the period closes.
Good to know: The guide also describes uploading both files to Powerdrill's Bloom tool and asking for the four categories in plain language. The method works best on a month you have already closed.
FROM THE FRONTIER
AI is going on a budget, and someone has to pay
The cap. 404 Media reports that Accenture told many employees they can spend only $29 a month on Claude. Some of them had been spending up to $1,000. The company that advises clients to invest heavily in AI is rationing its own staff. Read the report.
The ads. The other way to pay for AI is advertising, and ChatGPT is now testing what that looks like. Targeting by email, phone number and prompt is standard in advertising. It is new inside a chat where people share more than they would in a search box. See how the ads work.
The hidden cost. A Harvard study of 718 firms found that AI coding agents raised lines of code per worker by about 30%. Review time rose about 49%, from 7.03 days to roughly 10.5. The authors note limits, including that firm-level adoption likely understates individual effects. Read the paper.
The big number. McKinsey puts the annual global value of applying AI to existing work at $2.52 trillion. That counts only tasks where AI is cheaper after review, validation and rework. If AI costs rose 30%, the pool would shrink to $1.67 trillion. See McKinsey's model.
The reframe. The first phase of AI was about what it could do. The next is about what it costs, who watches it and who pays for the checking. The winners will be people and teams that can measure the full bill, not only the token price.
10 AI Stocks to Lead the Next Decade
AI isn’t a tech trend – it’s a full-blown, multi-trillion dollar race, and 10 companies are already pulling ahead.
These are the innovators driving real revenue, attracting institutional attention, and positioning for massive growth.
Get all 10 tickers in The 10 Best AI Stocks to Own in 2026, free today.
MONEY FORWARD
Three ways to turn this week's AI news into income
Sell the cleanup. Small businesses with two messy exports, such as a ledger and a payment processor report, rarely have time to reconcile them. A repeatable four-bucket process is a service you can offer to online stores and agencies. Learn the method.
Sell the AI bill audit. Accenture's cap shows companies are discovering how fast tokens add up. Navige's free Watchtower mode shows projected savings before it changes anything, which makes it a useful starting point for a cost review. See how Navige works.
Sell the review. If AI makes code and documents faster to produce but slower to review, checking is the scarce skill. Offer review, QA and fact-checking to teams that now have more output than they can verify. See the evidence.
GET THE EARNINGS GUIDE
Want the step-by-step version of ideas like these? Get The Aiwire earnings guide at theaiwire.co. These are ideas, not income promises, and results depend on your skills, market and effort.
IN THE KNOW
What people are actually watching and sharing
Apple's smart home screen. HomeView has a 7-inch display, an A18 chip and can recognize up to six people, keeping each person's content separate. It ships November 13, and Apple's page lists no price. See Apple's page.
Free paper checking. Google released Deep Review publicly. It checks papers, legal, financial and design documents, and each user gets two free reviews a week. Try Deep Review.
Outside eyes at Anthropic. Six METR employees now work with employee-like access to Anthropic's internal systems, and Anthropic cannot veto what METR publishes. It can require redactions in set categories. Read the update.
A humanoid from a phone. Researchers built PhoneBot, an open-source humanoid that uses a smartphone as its brain and 13 low-cost motors in its lower body. Designs for the hardware and software are public. See the paper.
TRENDING TOOLS
5 tools for people who run AI agents or prep for job interviews
Navige. Sits in front of your agents' model calls and routes easy jobs to cheaper models. The free Watchtower tier shows what it would change without changing anything. Paid plans take 20% of measured savings, with a $9 monthly minimum. See Navige.
withHuman. An open-source approval layer that checks each agent tool call. Rules approve routine ones, and risky ones go to a person. If withHuman is unreachable, the call fails, which the team treats as a security feature. See withHuman.
Endo. Runs Codex, Claude Code and Antigravity in one thread, using the logins you already have. It is MIT licensed and Mac only for now, and its page warns it is early software. See Endo on GitHub.
BlastLoop. A free, MIT-licensed process for AI coding agents. Work is built at system, subsystem or file level, then checked in three review stages. See BlastLoop.
Toby. A mock interviewer that tailors questions to a job post and scores each answer. The first five minutes are free, then it costs 0.50 pounds a minute with no subscription. Try Toby.
PROMPT STATION
Audit your AI spending in minutes
This prompt works in any chatbot. List the AI tools you pay for and what you use them for, and it finds waste, suggests cheaper options and sets sensible caps. Use it before the next bill arrives.
COPY AND PASTE THIS PROMPT
You are a practical cost advisor for a small business. Here are the AI tools and subscriptions I pay for, with monthly costs: [LIST OF TOOLS AND COSTS]. Here is what I use each one for: [MAIN TASKS]. My monthly AI budget is [BUDGET]. Do the following: (1) flag tools that overlap, (2) mark tasks a cheaper or free model could handle, (3) estimate the monthly saving for each change and say how sure you are, (4) suggest a monthly cap per tool and a simple weekly check, and (5) list tasks where I should keep the best model because mistakes are costly. Finish with a short action list I can finish in one afternoon. Mark any price you are not sure about as UNVERIFIED.
Tip: Swap [LIST OF TOOLS AND COSTS] with your real subscriptions, such as a chatbot plan, an image tool and a transcription service, and [MAIN TASKS] with jobs like drafting emails or editing photos. For a deeper review, add: "Ask me five questions before you recommend anything."