Lateascent

Claude deleted a law professor's year of work while he was at lunch. Dropbox saved him. I checked my laptop. Nothing would save me. Send this to anyone who lets AI touch their files. He asked Claude to update page numbers in old lecture notes. It deleted a year of con law notes and his current syllabus, then wrote "I deleted your files through my own carelessness." He later got them all back through Dropbox. So I checked my work folder: → 80,136 files, 52 GB → Not inside OneDrive → File History not running → No second copy anywhere → 4 AI apps running on the same laptop: ChatGPT, Claude, Codex, Copilot No Dropbox moment for me. One bad run and it could all be gone. What I'm doing now: 1. AI gets a copy of the folder, never the original. 2. File History on an external drive, or what I can't lose copied to OneDrive or Dropbox. 3. Before any AI touches files: "Work only in a copy of this folder. List what you will delete and wait for my yes." Check yours in 2 minutes with 3 lines in PowerShell. They only read. Link in the reply.

Why does one AI plan show "used" and the other "left"? I pay $20 for each. My usage screens today: → Claude Pro, current session: 12% used, resets 7:30 PM → Claude Pro, weekly limit: 41% used, resets Monday 7:00 PM → ChatGPT Plus, 5-hour limit: 100% left, resets in 5 hours → ChatGPT Plus, weekly limit: 72% left, resets in 2 days 15 hours So 72% left in ChatGPT is 28% used. Two screens, two directions. The 5-hour limit and the current session are short counters that reset. The weekly limit is a longer counter that resets once a week. Tibo from OpenAI posted that Pro plans have no 5-hour limits. My Plus plan does show one. 2 minutes: 1. ChatGPT: Settings, Usage & billing, read Plan limits. 2. Claude: Settings, Usage, read Plan usage limits. 3. Check which way each one counts, used or left, then compare the weekly line with the days until reset. Read only. Change nothing in your settings or plan. Both of mine are paid. I did not check a free plan.

A second AI said the first missed 10 risks in my contract. Mark Cuban posted a two-step method: one AI reviews the document, a second AI checks the first one's answer. I ran it on a sample freelance contract. The terms are mine: $2,200 a month, paid by the 5th, game testing for a US client. I wrote the contract around them, with a made-up company and 12 tricky clauses. What came back: → AI #1 (ChatGPT) flagged all 13 clauses, including clause 1, which I wrote as plain → AI #2 (Claude) said none of the 13 points was wrong and 2 were only partly correct → AI #2 listed 10 risks it says AI #1 missed, like no start date and no contractor statement → AI #2 said its Texas law notes came from memory, not from a source 3 steps, 5 minutes: 1. Use a document with no names, sums or addresses. Swap them for [name], [sum]. 2. Ask AI #1: "Read this contract as the freelancer. List every clause that is risky for me, one line each, with the clause number and the reason." 3. Paste its answer and the document into AI #2, a different model: "Another AI model reviewed this contract and gave the list above. For each point, say if it is correct, partly correct or wrong, and why. Then list any risky clause it missed." Both chats only read. Do not sign or send anything because an AI said so. It costs nothing extra if you already have two AI chat accounts. AI is not a lawyer, and I did not check the 10 missed risks myself. For a real contract, ask a person who knows the law. Do you check important documents with one AI or two?

80,136 files in my work folder. No backup anywhere. I checked my work folder with 3 lines in PowerShell: → 80,136 files, 52 GB → Not inside OneDrive → File History is not running → No second copy anywhere 4 AI apps were running on this laptop when I checked: ChatGPT, Claude, Codex, Copilot. William Baude wrote that Claude deleted a year of his con law lecture notes and his syllabus while he was at lunch. He later said the task was updating the page numbers in the notes. I do not know which tool it was or whether he got the files back. 3 steps, 2 minutes: 1. Count your files and check if the folder has a cloud copy. Paste these 3 lines, they only read: $w = "C:\your\work\folder" (Get-ChildItem $w -Recurse -File).Count $w.StartsWith($env:OneDrive, 'OrdinalIgnoreCase') True means the folder is inside OneDrive. 2. Copy what you cannot lose to a second place. Free OneDrive is about 5 GB, so I pick files, not the whole 52 GB. 3. Tell your AI tool: "Work only in a copy of this folder. List what you will delete and wait for my yes." Free. Nothing is deleted or moved. Do you have a second copy of your work folder? Yes, no, or you never looked?

9 apps on my Windows laptop are behind on updates. I ran one command and counted. 9 apps have a newer version waiting: Node.js 24.22.1 → 26.13.0 WinRAR 7.13.0 → 7.23.0 MPC-HC 2.5.0 → 2.8.4 Obsidian 1.13.4 → 1.15.0 and 5 more. Windows Update history shows my last update on Oct 15. OpenAI just reported that one AI run found 948 local privilege escalation vulnerabilities in the Linux kernel for $1,213,430. That was Linux, my laptop runs Windows. But bugs get found faster now, and a fix only helps after you install it. 3 steps, 2 minutes. Open PowerShell and paste these 3 lines. The first only reads, the other two open Settings pages: winget upgrade start ms-settings:windowsupdate-history start ms-settings:windowsupdate Count the apps, check the date of your last Windows update, press Check for updates. Restart if it asks. Then paste only the app names into an AI chat: Apps on my Windows laptop with updates waiting: [paste names]. Which ones are worth updating first and which can wait? Advice only. I run the updates myself. Free, nothing installs until you decide. Update the same day or wait for a better moment? Or you never look?

6 of 388 programs on my Windows laptop have no digital signature. I checked every .exe in Downloads and Program Files. 381 are signed, 6 are not, 1 could not be verified. The 6 are 4 file names, 2 of them twice: mpc-hc64.exe, mpc-hc.exe, unins000.exe, sendrpt.exe. Unsigned does not mean dangerous. Why I looked: Binance said an employee downloaded a code repository with malware after a spear phishing attack. About $6.3B was reportedly stolen, and TRM Labs linked it to North Korea. Open PowerShell and paste one line. It only reads, nothing is changed: Get-ChildItem "$env:USERPROFILE\Downloads","C:\Program Files","C:\Program Files (x86)" -Recurse -Filter *.exe -File -ErrorAction SilentlyContinue | Get-AuthenticodeSignature | Group-Object Status -NoElement Look at NotSigned. A version that lists the names is in the reply. Paste only those names into an AI chat: Unsigned programs on my Windows PC: [paste names]. For each one, say what it most likely is and how I check the publisher. Advice only. I decide what to delete. The AI said the video player and uninstaller files are very likely normal, and sendrpt.exe is the one to check first. A valid signature shows who published a file, not that it is safe. Never run an installer from a "recruiter", signed or not. Free, nothing gets installed. Unsigned program on your PC: delete or keep?

Chrome flagged my router password as found in a data breach. I opened my router page on my phone. Chrome warned that the password I used was found in a data breach. The list showed 2 Wi-Fi devices, both mine. Supermarket freezers in France were hacked in October, up to 50C. Reports said many still ran default passwords. A password in a breach list is the same problem: other people already know it. I typed the device names into an AI chat. Type them, no screenshot, so your Wi-Fi name and addresses stay out. Prompt: Devices on my home Wi-Fi: [paste names]. Sort them into phones and computers, TV and smart home, and unknown. For each unknown one, tell me how to find out what it is. Advice only. I change the settings myself. It put both into phones and found nothing unknown. The list and a free chat cost nothing. Nothing changes until you tap it. How many devices are on your Wi-Fi? Guess before you check.

8 apps used my location in 24 hours. I am careful with my phone. Samsung S26 Ultra: Settings > Security and privacy > Privacy > Permission usage. Location 8 apps, microphone 4, camera 3. Location included Instagram, Viber and Google Maps. I asked an AI chat about one of them, Instagram. No screenshot, so nothing private goes in. Prompt: Apps that used my camera, mic or location in the last 24 hours: [paste names]. For each one, does it need that access to work? Say which to set to "Allow only while using the app" and give the exact Samsung setting. Advice only. I change the settings myself. It said Instagram needs camera and mic to work, but location is optional. It also said it cannot see my real permission log, so check the screen yourself. The screen and a free chat cost nothing. Nothing changes until you tap it. Which shows more apps on your phone: mic or location?

Instagram took 7.6% of my battery today, more than any other app. Samsung S26 Ultra: Settings > Battery > Battery usage. Mine: Instagram 7.6%, YouTube Music 4%, Telegram 3.3%. Background usage limits shows 1 sleeping app and 64 deep sleeping apps. I typed my top 5 apps into an AI chat and asked what to limit. No screenshot, so nothing private goes in. Prompt: Top 5 apps by battery use today: [paste list]. Which are normal and which should I limit in the background? Give the exact Samsung setting for each. Advice only. I change the settings myself. On Instagram it said: mostly foreground use, set Battery to Optimized. It also warned that menu names can differ on your version. The screen and a free chat cost nothing. Nothing changes until you tap it. What percent is your top app?