Good morning. Three weeks ago on ARQIV we asked whether AI agents would replace your apps. This week OpenAI shipped GPT-6, and ChatGPT now answers some questions with a working mini-app instead of a paragraph.
Also inside: the electricity race behind AI, a Nobel week with a lot of brain science in it, and the quantum noise that may be holding the universe together.
In today’s Dispatch
- 🤖 AI: ChatGPT stops just talking: GPT-6 now builds you buttons, maps and mini-apps
- 💻 Tech: Google signs a 3.6-gigawatt power deal, with new nuclear output
- 🚀 Space: A “phoenix planet”? Old Hubble data hints at a world born from a dead star
- ⚛️ Quantum: “Cosmic lockdown”: quantum noise may keep the universe from collapsing
- 🎲 Surprise Science: Nobel Week: Nobel for the light switch in the brain: optogenetics wins Medicine
- 🔭 Futurism: What’s Coming: Apple’s smart-home push reportedly lands next Tuesday
🤖 AI
ChatGPT stops just talking: GPT-6 now builds you buttons, maps and mini-apps
OpenAI rolled GPT-6 into ChatGPT on Oct 7 with Intelligent UI. Answers can now arrive as interactive charts, forms, maps and small single-purpose tools, such as a road trip plotted on a map, a retirement calculator or a bill splitter. Paid tiers get it first; Free and Go users follow.
Why it matters: This is the "agents replace apps" idea showing up in a real product. If the assistant builds the calculator you need on the spot, a lot of single-purpose apps start to look optional.
Source: TechCrunch
Quick hits
- Meta and Sierra proposed the Personal Agent Protocol, a sign-in standard that defines what your AI agent may do at a store or bank. Shopify, Stripe and Walmart are backing it. (The Next Web)
- ChatGPT passed 1.2 billion weekly users, and OpenAI will test labelled visual ads next to image generation in the US. (TechCrunch)
- Mistral previewed Large 4, a reported 1-trillion-parameter mixture-of-experts model, and says the open weights will follow after red-teaming. (VKTR)
From ARQIV
Watch the latest ARQIV episode
Bold questions meet real science. New deep dives on AI, the brain, and the future — every episode on YouTube.
💻 TECH
Google signs a 3.6-gigawatt power deal, with new nuclear output
Google contracted for about 3,590 megawatts from Constellation Energy, one of the largest corporate power purchases in the US. Around 890 MW comes from upgrades that get more out of 11 existing nuclear reactors, with first power in 2028. The rest is grid supply.
Why it matters: AI is now limited by electricity as much as by chips. Big Tech is locking up decades of power, including nuclear, to keep its data centers running.
Source: Reuters via TimesLIVE
Quick hits
- Schneider Electric agreed to buy design-software maker PTC for about $22.6B, betting on AI for factories. (The Next Web)
- SpaceX is reportedly raising $40B in debt to pay for an order of Nvidia AI chips. (BNN Bloomberg)
- Finland ordered Google to pause work at two data-center sites while it checks whether an environmental assessment was skipped. (Al Jazeera)
🎬 New on ARQIV
Is This the End of Traditional Apps? The Future of Mobile
GPT-6’s mini-apps make this one timely. We go through weather, games, calendars, smart homes, parking and rental cars, and look at where AI agents help, where apps still win, and what happens when the AI gets it wrong.
▶ Watch on YouTube or listen on Spotify · Apple Podcasts
🔬 SCIENCE
🚀 SPACE
A “phoenix planet”? Old Hubble data hints at a world born from a dead star
Re-reading a 1999 Hubble observation of a white dwarf 270 light-years away, astronomers matched dozens of mystery signals to the element niobium. Their best explanation is a planet that formed from debris the star shed as it died, and whose atmosphere is now being stripped back onto the star. It is still a candidate, not a confirmed planet.
Why it matters: Planets might get a second life around dead stars, and this one turned up in data that sat in an archive for 27 years.
Source: NASA via Newswise
⚛️ QUANTUM
“Cosmic lockdown”: quantum noise may keep the universe from collapsing
Physicists modelling the early universe found that decoherence, the same environmental noise that ruins qubits, can suppress quantum tunnelling. That could keep a field stuck in a higher-energy "false vacuum" state. The authors stress this does not yet show that our own vacuum is safe.
Why it matters: The effect that makes quantum computers so hard to build may also help explain why reality is stable at all.
Source: University of Portsmouth
🎲 SURPRISE SCIENCE: NOBEL WEEK
Nobel for the light switch in the brain: optogenetics wins Medicine
Karl Deisseroth, Peter Hegemann and Georg Nagel share the 2026 prize for light-gated ion channels. Hegemann and Nagel found an algae protein that fires an electrical signal when hit by blue light. Deisseroth put it into neurons, so individual brain cells could be switched on and off with light.
Why it matters: Optogenetics lets researchers test which brain circuits cause which behaviors, which matters for work on depression, addiction and memory. It ties directly into our mental-health episodes.
Source: AP via NBC New York
Quick hits
- Physics: Francis Halzen, for IceCube, the neutrino detector built into the Antarctic ice, and its discovery of high-energy “ghost particles” from space. (AP via OPB)
- Chemistry: Henri Kagan and Kenso Soai, for showing how a small excess of left- or right-handed molecules can amplify itself. It is a clue to why life uses only one handedness. (Al Jazeera)
🔭 FUTURISM: WHAT’S COMING
Apple’s smart-home push reportedly lands next Tuesday
Oct 13 (reported): Bloomberg-sourced reports point to a home hub with a square screen, a new HomePod mini and a new Apple TV, with M6 Macs to follow later in October. Apple hasn’t confirmed any of it.
Source: Macworld
The US sets a date for a useful quantum computer
2028: A Department of Energy roadmap targets an error-corrected machine that can do real science by 2028. Progress will be judged on chemistry and fusion results, not qubit counts.
Source: Quantum Computing Report
Artemis III becomes the last big rehearsal before a Moon landing
Late 2027: Orion will practice docking with test versions of SpaceX’s and Blue Origin’s lunar landers in Earth orbit. A crewed landing attempt is planned for 2028.
Source: SpacePolicyOnline
📼 From the ARQIV vault
Cracking the Brain Code: Why Mental Health Is So Hard to Treat — this week’s Nobel tool, optogenetics, is part of how researchers are trying to answer the questions in this episode.
From ARQIV: Building something in AI, science or tech? Pitch yourself as an ARQIV guest. Apply here →
Which would you hand over to an AI first: your calendar or your thermostat? Hit reply and tell me. I read every one.
