Hey there!
Welcome back inside the kitchen. 🔥
🤔 Your Questions, Answered
This week, FryAI readers submitted the following questions. Let’s break them all down:
What protective measures/settings should I take to ensure any data I upload to an LLM remains secure?
Answer: If you’re using ChatGPT, I’d do three things immediately:
Go to Settings → Data Controls → turn off “Improve the model for everyone”
Use Temporary Chat for anything sensitive.
Remove names, account numbers, passwords, API keys, customer information, or other identifying details before uploading anything.
OpenAI says chats are not used for training after you opt out, and Temporary Chats aren’t used for training and can be retained for up to 30 days for safety purposes. For company-confidential information, I’d move beyond a personal account entirely and use ChatGPT Business/Enterprise or another enterprise AI product with explicit data protections; OpenAI says those workspace inputs and outputs are excluded from training by default.
My rule: if leaking the document would create a serious problem, don’t put the raw version into a consumer chatbot.
Which is the most intelligent AI?
Answer: There really isn’t one “smartest AI” across every task, but some perform types of tasks better than others. So I wouldn’t ask “Which AI is smartest?” as much as “Which one is smartest for what I’m doing?”
For difficult reasoning, research, and complex professional work, GPT-6 Astra and Claude Fable 5.1 are currently among the strongest overall models; both sit at the top of Artificial Analysis’ current Intelligence Index. Claude Opus 5 is right behind them.
For coding or computer science, I’d lean GPT-5.6 Sol or GPT-6 Astra.
For deep writing and working through long documents for work, I’d lean toward any Claude model.
For fast everyday questions, Gemini 3.8 Flash or GPT-5.6 Luna are strong choices.
How close is AI to passing human intelligence, and how long do you think it will take for them to deem humans inefficient and unnecessary?
Answer: AI has already surpassed humans at some specific intellectual tasks, but there isn’t one clean finish line called “human intelligence.” Think about how you might be more intelligent than someone at one thing, but they are more intelligent at another. It’s the same with AI. Humans still have major advantages in things like adaptability, real-world judgment, social understanding, and operating independently across completely new situations. And it’s worth mentioning that becoming more intelligent does not automatically mean an AI develops a desire to eliminate humans. Intelligence and goals are different things. The bigger concern is increasingly autonomous AI pursuing badly designed goals in unintended ways. That’s something worth watching closely, but anyone giving you an exact countdown to ”superintelligence” is mostly speculating. That’s actually why I include the Singularity Meter in every daily FryAI email — it’s my way of tracking how much closer we seem to be getting, but it’s still a rough estimate.
🧠 Food for Thought
This week’s moral take: AI is really good at making uncertainty disappear, even when the uncertainty is still there.
That may be one of its most underrated risks.
Real life is full of situations where the honest answer is, “I’m not sure.” A doctor may need more tests. A manager may not know which strategy will work. A student may be working through an idea that does not have a clean answer yet.
AI is built to respond, though. So instead of leaving uncertainty alone, it often gives us a polished explanation, recommendation, or prediction.
And that can feel like knowledge.
The problem is that confidence and truth are not the same thing. A clean, polished answer can be entirely wrong.
As we get used to asking AI for answers, I think we also need to get better at asking a second question: How certain should I actually be about this?
Because sometimes the most intelligent answer is not a better explanation. It is knowing when there isn’t enough information yet.
💰 The Business Bite
(This is for educational purposes only… not investment advice)
This week’s business take: As AI moves off our screens and into the physical world, machines need the ability to see.
A chatbot can make a mistake and regenerate an answer. A robot in a warehouse, factory, or hospital does not have that luxury. It has to recognize objects, judge distance, spot defects, navigate around people, and react to a constantly changing environment in real time.
That makes machine vision a really interesting part of the physical AI boom. As robots get more capable, cameras and AI-powered vision systems become their eyes.
The company I’m watching: Cognex
Cognex has spent decades building vision systems that help factories inspect products, read labels, find tiny defects, and guide automated equipment. Now that expertise is moving directly into AI and robotics.
The company recently agreed to acquire RealSense, which makes 3D cameras that help robots understand depth and navigate physical spaces. Cognex is also rolling out AI vision systems that can perform complicated inspections directly on the factory floor without sending everything back to a separate computer.
That is what makes the company interesting to watch. Cognex does not need to build the winning robot.
It can sell the eyes to whoever does.
What was the temperature of this week's edition of Inside the Kitchen?
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.
📧 Email me directly if you have any questions or want to chat!