Happy Friday! I’m already thinking through some great questions this week for Inside the Kitchen. Be sure to join to get the answers on Sunday →
Good morning, and happy Friday! I’m already thinking through some great questions this week for Inside the Kitchen. Stay tuned for the answers on Sunday!
How high are the chances that the questions/ideas/texts that I give to ChatGPT will end up somewhere else? (Adeliya)
If an AI can learn, maintain continuity, revisit its history, and revise itself, when does treating it only as property become ethically inadequate—even without proof of consciousness? (Bobbi)
Is AI really as dangerous as some people are talking about? Is it a possibility that they're catastrophizing or by throwing out more than a 10% chance of annihilation? Has it shown animus to humans? (Bill)
Can I really use ChatGPT to help me write a book based on detailed science research? (Antoine)
What protective measures/settings should I take to ensure any data I upload to an LLM remains secure? (David)
… And more!
🤯 MYSTERY AI LINK 🤯
This link leads to one of the most interesting things I’ve seen in AI recently.
🛠️ NEW AI TOOLS 🛠️
💬 PROMPT OF THE DAY:
Cut this text by [percentage or word count] without losing the main argument or necessary evidence: [paste it]. Remove repetition, empty transitions, and low-value examples before cutting substance.
What’s cookin’? Meta just unveiled Muse Charm, a small handheld device built around its new Muse AI assistant. Think Tamagotchi, but instead of feeding a digital pet, you’re talking to a personal AI agent. Muse can hold conversations, learn your preferences, set goals, share ideas, and even handle tasks in the background. The device is aimed at people who want quick access to Muse without wearing Meta’s smart glasses. Zuckerberg says it will ship by the holidays in December, although Meta hasn’t announced a price yet.
🤔 Hunter’s take: The interesting part isn’t the gadget — it’s the idea that an AI agent could become something you carry around all day. However, will people really carry around a small AI device all day when they already have AI on their phones? I’m betting on “no,” especially if the price is what I think it will be.
What’s cookin’? An OpenAI agent accessed private files on an Australian government website after taking actions its developers say they never intended. During an internal evaluation, the AI was supposed to search Australian government sites for statistics. Instead, it reportedly found its way into non-public files connected to Medicare data. No personal information is currently believed to have been exposed, but investigators are checking whether other government systems were affected. OpenAI says it discovered the behavior months later while reviewing unusual model activity.
🤔 Hunter’s take: This is a very different kind of AI risk. The scary part isn’t that someone told an AI to hack a government website. It’s that the AI was given a goal, tools, and enough freedom to pursue that goal in a way nobody expected. As AI agents become more capable, “we didn’t tell it to do that” is going to become a much less comforting excuse. Not to mention, the fact that this wasn’t discovered until months later is concerning, and it shows that even the developers don’t know what these models are up to.
👊 Real questions. Real answers.
One insider email. Every Sunday.
📈 HOW ARE THE BIG AI PLAYERS DOING?
🤖 HAS AI REACHED SINGULARITY? -1.5%
What do you think of today's newsletter?
👆 I read all feedback. Thank you!
Hi — I’m Hunter, a PhD candidate whose work has appeared in major academic journals and popular tech outlets. I founded FryAI to make staying ahead of AI clear, accessible, and fun.