Posts

Showing posts with the label AI Safety

Your Deleted AI Chat Isn't Gone: ChatGPT Keeps It 30 Days, Gemini Keeps Some for 3 Years

Image
You hit delete on a chat you wish you had never typed. Is it gone? Delete removes the chat from your screen in a second — but it does not remove it from the company's servers, or from the assistant's memory, for weeks, months, or in one case years. I checked what OpenAI, Anthropic, and Google actually promise in their own help pages, as of November 2026. All three agree on the headline number — 30 days — and disagree on almost everything that matters after it. The short version: your deleted chat is gone from your view at once, gone from the servers within about a month, and still alive in the assistant's memory until you delete that too. One of the three keeps some chats for three years even after you delete them. Figures are as of November 2026 unless noted. Delete Means Three Different Things Your screen, their servers, and the assistant's memory are three separate deletions — and the Delete button only does the first one properly. When you press delete, three t...

AI Voice Cloning Scams: 3 Seconds of Audio Cost One Mom $5,400 — and the Code Word That Stops Them

Image
Your daughter calls, crying. She says she has been kidnapped, and the man on the phone wants money — now. What do you do? You cannot tell a cloned voice from a real one by listening. The only test that works is a code word you agreed on before the call ever comes. In May 2026, a mother in Martinez, California, got exactly that call. Fortune reported that the voice on the line sounded exactly like her 37-year-old daughter, crying and saying she was scared. A man claimed a cartel had kidnapped her and demanded $20,000. She stayed on the phone for five hours and wired $5,400 to Mexico before she reached her daughter — who had been at work the whole time. This guide explains how little audio a scammer needs, what the newest official numbers actually say, how to set up a family code word in ten minutes, and what to do in the first hour if money has already moved. 3 Seconds Is All It Takes A three-second clip of someone talking is enough to build a voice clone that matches the orig...

What Can an AI Agent Actually Do for You? 5 Everyday Jobs That Work — and the 11% Line You Shouldn't Cross

Image
You have heard that AI agents can do your errands for you. But what does that actually mean on a normal Tuesday — and where does it stop? An AI agent can already sort your inbox, research a purchase, and chase a follow-up for you — yet only 11% of US consumers are willing to let one actually buy anything. The gap between those two numbers is the whole story. Agents are genuinely useful for preparation and follow-through, and genuinely unready for unsupervised spending. This post names five everyday jobs that work now, what they cost, and the line you should not cross yet. Figures are as of November 2026 unless noted. The one test that tells an agent from a chatbot A chatbot answers. An agent acts — it plans steps, uses tools, checks the result, and adjusts. Anthropic describes the difference as a loop: the system plans, acts, observes what happened, and repeats until the task is done or it needs you. Ask a chatbot to plan a trip and you get an itinerary; ask an agent and it can che...

How AI Chatbots Generate Answers — and Why They Sometimes Invent Facts

Image
A chatbot does not look up a fact and then write it down. It predicts the next word, then the next, based on patterns learned from large amounts of text. That single difference explains why an answer can read smoothly, sound certain, and still be wrong. Fluency is not proof. A confident tone, a tidy structure, and a plausible citation can all be produced by the same prediction process that produces a correct answer. The check has to happen outside the chatbot. What actually happens when you ask a question When you type a prompt, the model breaks the text into tokens — words or parts of words — and predicts which token is most likely to come next, given everything before it. It repeats that step until the answer ends. It learned those likelihoods from training text, not from a verified database of facts. That design works well for language itself. Grammar, common phrasing, and frequent facts appear often in training text, so the model reproduces them reliably. Rare or one-off facts ...