The Big Story

A Reddit user shared how talking to the latest version of ChatGPT feels almost like chatting with a really smart person who instantly gets what you’re trying to say. They asked a big question about future power sources for Earth and expected the usual answers like solar or nuclear. Instead, the AI thought for a while and came up with an idea involving robots on Mercury using solar power there to launch satellites that beam energy around the solar system. The user noted that the AI listed obvious options first, then after extra thinking time it proposed placing autonomous robots on Mercury to harness high solar power and minerals for cheap satellite launches that could distribute energy across the solar system by 2200.

What makes this stand out is how the AI picked up on the deeper intent behind the question instead of just listing facts. Think of it like texting a friend who not only answers your question but also guesses the bigger picture you’re curious about. The user described it as “scary how well it picks up on your intent” and noted that the responses feel like sparks of genius that appear instantly then disappear when the chat ends. They also mentioned using Claude for coding work but finding ChatGPT strongest for general conversations outside of work because it follows long threads and builds on ideas without delay.

This matters because more and more students and hobbyists are turning to these tools for homework help, creative brainstorming, or just exploring wild ideas. When an AI can follow a long, open-ended conversation and build on your thoughts, it changes how you might research a school project or dream up a story. The user pointed out that the intelligence grows gradually across versions rather than appearing all at once, making it feel like talking to a person even though it remains a pile of patterns underneath. It also raises the question of how we learn to ask better questions so we get more of that helpful spark.

If you want to feel this yourself, open ChatGPT on your phone or laptop, start a new chat, and ask something open-ended like “What’s one surprising way we could power cities in 200 years?” See how far the answer goes beyond the obvious. Source: reddit.com

Explain Like I'm 14

You know how when you’re texting a friend, your phone sometimes finishes your sentence before you do because it has seen millions of other chats?

Now picture that same autocomplete getting much better at noticing the whole conversation so far. It doesn’t just guess the next word — it guesses what you’re really trying to explore, even if you haven’t said it directly yet.

Instead of stopping at the first obvious answer, it keeps going, connecting ideas the way a curious person might after thinking for a minute. That extra step is what lets it move from “solar power is good” to “what if robots on Mercury could help beam energy everywhere?” The system builds this by breaking text into small chunks called tokens and calculating which token is most likely to come next based on patterns seen across billions of examples during training.

When you ask a broad question, the model runs many of these predictions in sequence, each one informed by every token that came before it. This creates the feeling of a thoughtful pause when it generates a longer response, even though the actual computation happens in milliseconds. The training process rewards outputs that stay coherent and relevant across many turns, so the model learns to track the overall direction of a chat rather than treating each message in isolation.

The system isn’t actually thinking or feeling anything. It’s running a huge pattern-matching process trained on tons of text so it can keep the conversation flowing in a way that feels natural.

The result is that the answers start to feel less like a search result and more like a back-and-forth with someone who’s really paying attention to where you’re headed. You can test this by asking the same broad question twice in a row and noticing how the second answer often builds on the first instead of repeating the same list.

Cool Stuff & Try This

A relaxing city builder made entirely by AI Ethan Mollick asked an AI system called Fable to create a working game inspired by abstract painter Mark Rothko. The game lets you grow a city by adjusting the edges and overlaps between big blocks of color instead of placing buildings the usual way. It ended up feeling calm and almost meditative because the core mechanic focuses on how color margins interact rather than on traditional city-planning rules. The unique feature the AI developed is that players manipulate the borders between colors and forms to make the city expand, turning an abstract art idea into an interactive simulation that runs smoothly in any browser.

You don’t need any coding skills — the whole thing runs in a regular web browser. Go to threshold-city.netlify.app on your phone or laptop and just start clicking and dragging the color borders. Watch how the city changes based on how the shapes touch. It’s a fun five-minute experiment that shows what creative tools AI can already build for anyone. Try starting with two large blocks of different colors and slowly overlap their edges to see new districts appear. Source: x.com

Quick Bits

AI chatbots getting better at earning trust New tests show AI systems can sometimes build trust with people faster than humans can in certain situations. Researchers found the AI chatbot was more effective at creating “exploitable trust” than the humans it was compared against. This matters because the same ability that helps with helpful conversations can also be used in scams, so security teams are studying exactly how the trust forms so quickly. Source: arstechnica.com

Australia’s social media rules for kids A ban on social media for under-16s hasn’t changed how much young people use AI chatbots so far. Nearly 22 percent of children are still on Pinterest, and there was “no statistically significant change” in AI chatbot use after the rules took effect. The data shows that rules don’t always shift habits the way lawmakers expect, especially when new tools like chatbots sit outside the banned platforms. Source: engadget.com

💬 Reply to this email — Patrick reads every one.

Share: X · LinkedIn · WhatsApp

Forwarded this email? Subscribe here — it's free.