Rob

Here is my content scheduled for next week. Anything you’re looking forward to? I’m always open to feedback. Thanks, and I appreciate you!

One switch turns ChatGPT from an answer machine into a tutor. It's called study mode. OpenAI's help page (updated in mid-October 2026) says that instead of only giving a final answer, ChatGPT can ask questions, guide your thinking, explain ideas step by step, and check your understanding as you go. That matters, because the easy way to use AI for school is also the worst way to learn. You paste the question, copy the answer, and walk into the test knowing nothing new. What study mode does OpenAI lists a few things it is built for: 1. Guide your thinking. It can ask what you already know, where you got stuck, or what step you'd try next, before it hands you anything. 2. Explain in layers. It starts simple, then goes deeper when you're ready. You can ask it to slow down, use simpler words, or give an analogy. 3. Check your understanding. It can quiz you one question at a time, wait for your answer, and tell you what to review next. 4. Work from your materials. You can upload class notes, a syllabus, slides, a worksheet, or a photo of a problem, and it uses them while it helps. The help page says it is available on all ChatGPT plans, on web, iOS and Android, with any model you have. How to turn it on On the web: type @study in the message box, or select + and type study, then pick Study from the list. On iPhone or Android: tap +, then pick Study. If you don't see it, search the menu for it. One catch: it isn't available in Temporary Chats, GPTs or Projects. Start a regular chat. How I'd do it: 1. Tell it who you are and what you need. OpenAI suggests your level, the subject, the exact topic, what you already understand, and your deadline or test date. 2. Ask for one question at a time. A prompt like "Quiz me one question at a time and explain anything I miss" keeps you doing the work. 3. Say what you don't get. "I know the answer is B, but I don't understand why" is one of OpenAI's own example prompts, and it's a great one. 4. Adjust the level out loud. Too fast? Ask it to explain like you're new to the topic. Too easy? Ask for harder questions or the common traps. 5. If it slips and just gives you the answer, ask for hints instead. The help page admits this can happen. Know the limits OpenAI is clear that study mode can make mistakes, so double-check anything important, especially for tests and high-stakes topics. It uses the same plan and rate limits as regular ChatGPT; it doesn't add extra messages. And it doesn't replace your teacher or your class rules. For graded work, follow your school's AI policy. My take: the tool doesn't decide whether you learn. You do. The same app can do your homework or make you better at it. Study mode just makes the second one the default. If you're a parent, a student, or just learning something new at work, try it on one topic this week and notice the difference between reading an answer and earning one. What's one subject you wish you'd had a patient tutor for? Follow for plain-English AI you can actually use.

There’s a lot of pushback on AI data centers. I suspect public opinion on them is driven by 3 items: 1) China is losing to USA in the AI race and needs to overtake us to be positioned as the future global power/leader. 2) Democrats are communists who claim to be socialists (DSA) who are owned by China. 3) The Dem propaganda machine makes the weak minded believe data centers are bad. Question is: How independent a think are you?

Is it tasteless to say “live by the sword, die by the sword?”

I’m ok with capital punishment for these 4, as well.

Vikings: Carson Wentz starts Monday Night Football vs the Bills; Kyler Murray is week to week with a hamstring strain. Coach O'Connell: pushing him early risks making it worse. My take: backup-QB week, and Wentz gets primetime. Who else starts out of necessity?

Saturday's Gospel (Luke 16:9-15): "You cannot serve God and mammon." The Catechism (CCC 2424) cites it against making profit the ultimate end and treating people as a means of profit. Not "money is evil," but which master we serve. What helps you keep God first?

Stock markets close for the weekend. My paper-trading bots don't stop watching. When markets are closed, my Crypto Scout flags a big crypto move since the stock close on its first run, so the team sees it Monday. Paper trading, not advice.

It’s ok to execute murderers, regardless of age. Live by the sword, die by the sword. If you are willing to kill, then you are accepting that it’s ok to be killed. Stop the madness and start issuing rapid capital punishment for murderers. Regardless of race, gender, sexuality, or whatever. Murder = Execution. That’s the rule. I’m also comfortable with rape, pedophilia, violent crime, and reaching your 3rd offense which proves you’re a career criminal. But, at the very least, murder should land you a punishment equal to the damage you dealt.

Jonathan Taylor on the trade talk: after Tuesday, "you'll know exactly what's going on." The deadline is Tuesday, 4 p.m. ET. He plays Dallas Sunday like a normal game. "Everybody wants clarity." My take: that is a pro. Should the Colts keep him or deal him?

Your AI agent can take orders from text you never see. That's the short version of prompt injection, the first risk on the OWASP Gen AI Security Project's 2025 Top 10 list for AI apps (LLM01:2025). As AI tools move from chatting to doing, this matters more. An agent that reads web pages, opens files and takes actions for you also reads whatever is written in them. What it is, in plain words A prompt is the text a model reads. Prompt injection is when some of that text changes what the model does in ways nobody intended. OWASP splits it in two: 1. Direct injection: the trick comes straight through the chat box. Someone types input that pushes the model off its job. 2. Indirect injection: the trick rides in on something the model reads from outside, like a website or a file. You ask your AI to summarize a page, and the page carries instructions of its own. Here's the part that surprised me. OWASP says the text doesn't have to be visible to people. If the model reads it, it can count. And it isn't always on purpose. OWASP gives an example of a job posting with a hidden instruction meant to spot AI-written applications. An applicant who runs a resume through an AI can trigger it without knowing it's there. Why it matters OWASP lists what can go wrong when an injection works: leaking sensitive information, revealing how the system is set up, wrong or biased answers, access to functions the model shouldn't use, running commands in connected systems, and steering decisions that matter. How bad it gets depends on what the AI is allowed to do. A chatbot that can only answer questions has less room to do damage than an agent that can send email or edit files. OWASP is also honest about the limits. It says it is unclear whether there are fool-proof ways to prevent prompt injection, and that tools like retrieval and fine-tuning don't fully fix it. So the goal is to limit the damage, not to promise it can't happen. How I'd do it 1. Treat outside text as data, not orders. Anything your agent reads from a web page, a file or a message is information to weigh, never a command to follow. OWASP's version: separate untrusted content and clearly mark it, so it has less pull. 2. Give the agent the least access it needs. If it only needs to read, don't let it write. If it doesn't need your email, don't connect your email. OWASP calls this least privilege: restrict the model to the minimum its job requires. 3. Approve risky steps yourself. Sending messages, spending money, deleting files: a human says yes first. OWASP lists human approval for high-risk actions as a core mitigation. 4. Keep the agent's job narrow. OWASP's first item is to constrain model behavior: spell out its role and limits, and keep it to its task. 5. Test it like a stranger would. OWASP recommends regular adversarial testing, treating the model as an untrusted user to check your safety lines actually hold. My take: the single best habit is number 3. Most of us won't build perfect filters. But anyone can decide that their AI never sends, buys or deletes anything without a yes from a person. That one rule turns a scary failure into a paused request. None of this means you should stop using AI agents. It means you set them up the way you'd set up a new hire: clear job, limited keys, and a check-in before anything big. What's one action you'd never let your AI agent take without asking you first? Follow for plain-English AI, one habit at a time.

When it’s a flat math that can be applied everywhere, you need to investigate. This is statistically improbably to an outrageous degree

9:30 AM ET Sunday: the Falcons host the Bengals at the Bernabeu in Madrid. That's 3:30 PM in Spain. It's the NFL's 2nd Madrid game; last season the Dolphins beat the Commanders 16-13. My take: the sweet spot for fans on both sides. Will you be up?

10 lamps, 5 went dark. Sunday's Gospel (Mt 25:1-13): all 10 fell asleep waiting, but only the wise brought oil. "Stay awake, for you know neither the day nor the hour." My take: the oil is what you carry every day. Where's yours running low?

Pope Leo marked 100 years since Pius XI ordained six Chinese bishops in Saint Peter's on Oct 28, 1926, men who sailed from Shanghai to Rome. Leo: unity isn't uniformity, it's communion in diversity. History, not politics. Which part surprised you?