Posts

Showing posts with the label ChatGPT

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 Detectors Flagged 61% of Real Student Essays as AI — Can You Trust the Score?

Image
Did a detector just flag something you wrote yourself — an essay, a cover letter, a report — as AI-generated? A detector score is a probability, not proof. And the research shows it can be badly wrong in exactly the cases where it matters most. What the Studies Actually Found The story changed twice: the 2023 tools were unreliable, the best 2026 tools are far better — and the same three weak spots still break them. When ChatGPT went mainstream in early 2023, detection tools rushed to answer one question: did a machine write this? Three years of independent tests now give a clear answer — and it depends heavily on which tool, which text, and which year you mean. Study Tested Found OpenAI classifier, 2023 OpenAI's own tool Caught 26% of AI text; flagged 9% of human text; shut down July 20, 2023 Liang et al., Stanford, 2023 7 detectors, 91 TOEFL essays 61% of non-native essays flagged as AI Weber-Wulff et al., 2023 14 detection tools All below 80% accuracy; ...

1 in 4 AI Summaries Invents Something: How to Stop AI Hallucinating in Technical Documents

Image
Have you ever pasted a technical document into a chatbot, got a clean summary back, and later found a number in it that appears nowhere in the original? That invented number is the most common failure in AI summarization — and the fix is a workflow, not a cleverer prompt. This post shows you how often AI invents details when it summarizes a technical document, the four-step workflow that catches the inventions, the five invented details to check first, and when a quick check is not enough. Figures are as of November 2026 unless noted. How often does AI invent details in a summary? Even the best models invent something. The question is how often — and on what kind of task. Vectara's Hallucination Leaderboard is the most direct measurement we have. It hands each model more than 7,700 articles and asks for a summary using only the facts in the document, then scores the result with its own evaluation model, HHEM-2.3. In the October 22, 2026 update, 109 models ranged from a 1.8% ha...

How to Summarize a Long PDF With AI — Without Losing the One Number Your Boss Will Ask About

Image
You drop a 60-page report into a chatbot, ask for a summary, and get a clean one-pager back in seconds. Then your boss asks about one number — and it is not there. That is not a bug in your prompt. Summarization is compression, and without instructions about what must survive, the model compresses away exactly the details you needed most. This guide shows you the workflow that keeps them: pick the right tool for the document, extract before you summarize, and run a five-minute check before you trust the result. Figures are as of November 2026 unless noted. Why the details vanish — and why a bigger window does not fix it A summary is a smaller version of the document, so something has to go. The question is what — and the model decides that for you unless you tell it otherwise. Researchers at Stanford and Berkeley measured the problem in 2023. In their "Lost in the Middle" experiments, they hid the one document containing the answer among 19 distractors and moved it from ...

AI Tools for Job Seekers: What to Hand to AI and What to Check Yourself

Image
The short answer: let AI draft and rehearse, but keep the fact-checking and the final decision for yourself. Hand a chatbot your resume and the job posting and ask it to tailor the wording; let it run mock interviews with you out loud. Never let it invent experience you do not have, and never let an auto-apply tool submit on your behalf. That division matters more than which tool you pick, because the other side of the table is using AI too. How much AI is on the employer side? SHRM's 2025 Talent Trends survey of 2,040 HR professionals found that 51% of organizations use AI to support recruiting, and among those adopters, 44% use it to screen resumes — up from 26% using AI in any HR task just a year earlier. A separate August 2025 survey by Resume.org of 1,399 U.S. workers found 57% of companies already use AI to support hiring, with 79% of those using it to review resumes. LinkedIn's Future of Recruiting 2025 report adds that recruiters using generative AI save roughly 20% ...

AI Tools for Everyday Tasks: Which One to Start With (A Beginner's Guide)

Image
The short answer: start with one free general chatbot — ChatGPT, Claude, or Gemini — and use it for everything for a week. Check what AI is already inside the apps you pay for. Add a specialist tool only for the one recurring task that costs you the most time, and pay for nothing until a free plan limits you twice in one week. That advice runs against most "best AI tools" lists, which hand you thirty app names and no way to choose. The Federal Reserve Bank of St. Louis found that 54.6% of US adults aged 18–64 had used generative AI by August 2025 — up from 44.6% a year earlier — and that people now use it more for personal tasks (48.7%) than for work (37.4%). The question for most readers is no longer whether to try AI, but which tool to try first. St. Louis Fed AI adoption tracker Step 1: Pick one general chatbot and use the free tier For everyday tasks — drafting emails, summarizing documents, explaining things, planning — a general chatbot covers most of it. The th...

Writing Better Prompts for Work: A Practical Guide

Image
  A good work prompt answers five questions before the model starts: what is the task, who is it for, what context matters, what limits apply, and what format should come back. If you give those five, you usually get a usable first draft. If you give only a topic, you usually get a generic answer that needs another round of fixing. Google's own guidance for Gemini in Workspace makes the same point in plain terms: the best prompts provide the context, or the specific details, the tool needs to produce a great response. A prompt is not a search query. It is a short brief to a colleague who cannot see your desk, your deadline, or your last email thread. The five parts that cover most work prompts Task: Say the exact job to be done, using a verb: draft, summarise, edit, compare, or outline. Audience: Say who will read or use the result, and what they already know. Context: Give the background that changes the answer: the project, the decision, the previous discussion, or the materia...

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 ...