DEV Community: 박태인

How to find the subscriptions you forgot you're paying for (without linking your bank)

박태인 — Fri, 26 Jun 2026 08:18:29 +0000

The average person wastes over $200 a year on subscriptions they forgot about. A free trial that started billing. An app you used once. A service you meant to cancel. Here is how to find them - without handing your bank login to another app.

Search your email for the words that bills use. In your inbox, search each of these one at a time: "receipt", "your subscription", "renews", "payment", "invoice", "free trial". Receipts you forgot will surface fast.
Check the app stores directly. On iPhone: Settings, tap your name, then Subscriptions. On Android: Play Store, profile, Payments and subscriptions. A surprising number hide here.
Look at your card statement for round, repeating amounts. Scan two or three months. The same charge on the same date each month - $9.99, $14.99 - is almost always a subscription. Note the merchant name and search it.
Watch for the annual ones. The worst offenders bill once a year, so they slip past a monthly scan. Search your email for the same month last year.
Cancel at the source. Always cancel from the company's own account page or the app store, not a random "cancel my subscription" website.

Doing this once a year quietly gives most people back a couple hundred dollars.

The reason people don't do it is that it is tedious, and the apps that automate it want your bank login - which feels invasive, because it is. That trade-off is the problem I am working on.

SubAudit does the same thing using only your forwarded receipt emails. No bank connection, ever. You forward the receipts, it builds one clear list of every recurring charge with the price, the renewal date, and a direct cancel link.

Less power over your money, far less risk. If that trade sounds right, it is in early access here (first 100 free):

https://panels-charles-occupational-varied.trycloudflare.com/s/subaudit

Either way - run the five steps above this week. The money is already yours.

How to tell if a text message is a scam: a 5-second check

박태인 — Fri, 26 Jun 2026 08:16:04 +0000

Scam texts now look real. They copy your bank, a delivery company, even a family member. Here is a simple check you can run before you tap anything.

Check the sender, not the name. A real bank does not text from a personal mobile number. Long random numbers, foreign codes, or email-to-text addresses are a bad sign.
Look for urgency. "Your account will be closed in 24 hours." "Final notice." Real institutions do not rush you by text. Urgency is the scammer's main tool - it stops you from thinking.
Never tap the link. Hover or long-press to preview the address. If it is not the company's exact official domain, it is fake. Scam links often add words: yourbank-secure-login.com is not yourbank.com.
They ask for something a real company already has. Your full card number, your password, a one-time code. No legitimate company asks you to confirm these by text.
Verify on your own. Do not call the number in the text. Open the official app or type the website yourself. If it is real, the alert will be there too.

If even one of these is off, treat it as a scam: do not reply, do not tap, delete it.

The hardest part is that this happens to the people least equipped to run the check - older parents, often alone, often trusting. They lose billions a year to texts that look exactly like the real thing.

That is the problem I am working on. ScamShield lets anyone forward a suspicious text, email, or call and get an instant answer - scam or safe, why, and what to do - simple enough to set up for a parent.

If that would help you or someone you look after, it is in early access here (first 100 free):

https://panels-charles-occupational-varied.trycloudflare.com/s/scamshield

When in doubt, slow down. Urgency is the scam.

The 30-second citation check before you submit

박태인 — Fri, 26 Jun 2026 04:50:42 +0000

Do this for every reference AI helped you write.

Search the title on Google Scholar.
Resolve the DOI separately at doi.org. If it points to a different paper, stop.
Match author, year, and journal to the record.
Open the source. Find the exact line that supports your claim.
No supporting line? Cut the citation.

Steps 1 to 3 are now automated. Free tools - CiteTrue, SwanRef, Citely - check existence and metadata against CrossRef and PubMed in seconds.

Steps 4 and 5 are still manual. No tool reliably checks whether a real paper supports your specific claim. That is the error that gets papers rejected and retracted.

I am building a tool for steps 4 and 5. Early access, first 100 get a free audit of one paper:

https://least-visited-lottery-fitness.trycloudflare.com

Open your sources. AI assists; you stay accountable.

Every AI citation raises two questions. Tools answer only one.

박태인 — Fri, 26 Jun 2026 04:44:39 +0000

Question 1: Does the paper exist?
Question 2: Does it support your claim?

Free tools answer Question 1. They check every reference against CrossRef, PubMed, and OpenAlex in seconds. Use one — CiteTrue, SwanRef, or Citely.

No tool answers Question 2. And Question 2 is where papers get retracted: a real, correctly formatted reference attached to a claim the source never made. AI produces this often, because it finds a paper on the right topic and staples it to your sentence.

The fix is manual, for now. Open the source. Find the exact line that backs your claim. If there is no line, cut the citation.

I am building a tool that answers Question 2 — it reads the source and flags where it does not support your claim. Early access, first 100 get a free audit of one paper:

https://least-visited-lottery-fitness.trycloudflare.com

Open your sources. AI assists; you stay accountable.

Fake citations in published papers jumped 12x in three years

박태인 — Fri, 26 Jun 2026 04:41:07 +0000

In 2023, 1 in 2,828 PubMed-indexed papers contained a fabricated reference. In early 2026, it was 1 in 277. That is a 12-fold increase in three years.

The cause is direct: researchers use ChatGPT, Claude, and Gemini, and these models invent references that look real.

Two things are true at once.

Fabricated references are easy to catch. Free tools — CiteTrue, SwanRef, Citely — check every reference against CrossRef and PubMed in seconds.

One error is caught by no tool: a real paper cited for a claim it never made. A reviewer catches it. So does a retraction.

What to do if you use AI in research:

Run your reference list through a free checker.
Open every source behind a key claim and find the line that supports it.
No line, no citation.

I am building a tool for the second problem — an audit that checks whether a cited paper actually supports your claim. Early access, first 100 get a free audit of one paper:

https://least-visited-lottery-fitness.trycloudflare.com

Open your sources. AI assists; you stay accountable.

Half of ChatGPT's citations are wrong. One half is dangerous.

박태인 — Fri, 26 Jun 2026 04:39:14 +0000

A Deakin University study tested GPT-4o''s citations. 56% were fabricated or contained errors. For less-researched topics, the fabrication rate approached 30%.

There are two kinds of wrong.

The paper does not exist. Easy to catch. Free tools flag it in seconds.

The paper exists, but does not say what you claimed. No tool catches this. A reviewer does. So does a retraction.

The second kind is dangerous because everything looks correct: real author, real DOI, correct year. Only the claim is wrong.

How to protect yourself today:

Run your reference list through a free checker — CiteTrue, SwanRef, Citely.
Open every source behind a key claim and find the exact line that supports it.
No line, no citation.

I am building an audit for the second problem — it reads the source and flags where it does not support your claim. Early access, first 100 get a free audit of one paper:

https://least-visited-lottery-fitness.trycloudflare.com

Open your sources. AI assists; you stay accountable.