In partnership with

Analytics on Live Data Without Leaving Postgres

When analytics on Postgres slows down, most teams add a second database. Then come the pipelines, the sync jobs, and a copy of your data that's always a little behind.

TimescaleDB takes a different approach: extend Postgres instead of splitting away from it. Hypertables partition your data automatically as volume grows. Hypercore compression cuts storage up to 95%. Continuous aggregates keep dashboards live without re-querying everything.

CERN runs Postgres this way for sensor data from the Large Hadron Collider.

No split architecture, no pipeline lag, no new query language to learn. Same SQL, same drivers, same tools.

Start on Tiger Cloud and get $1000 in credits.

TODAY IN AI

4 things that happened while you were busy

1.  Washington and Beijing both reject an AI slowdown. Donald Trump called Jensen Huang live on stage at the All-In Summit on Monday, told the Nvidia CEO that AI safety fears are "a hoax," and said slowing down would hand the race to China. Huang agreed. Earlier that day the president had posted that AI needs no guardrails beyond a strong president. Across the Pacific, China's foreign ministry dismissed Amodei's essay as a tactic to contain the country's AI progress, which leaves the plan's second and third steps without a government willing to take them.

2.  Meta's Muse becomes the number two app in America. Meta's personal agent has climbed to second place among free iPhone apps in the US, behind only ChatGPT and ahead of Threads, WhatsApp and Facebook, with more than 83,000 downloads in its first days according to Sensor Tower. It is a fast start for an app that asks for your inbox, calendar and payment details, and a sign the privacy questions raised at launch are not slowing adoption.

3.  A Shanghai lab enters the frontier with an open-weight model. Shanghai Artificial Intelligence Laboratory released Atria Dawn Preview, an open-weight model pitched at researchers and built to produce verifiable, reproducible results. The lab's own benchmarks put it alongside Kimi K3 and Claude Opus 5 on several tests, though nobody outside the lab has confirmed that yet. Developers can try the API now.

4.  Sam Altman explains himself. Sam Altman published a long post arguing the world deserves confidence that labs will act responsibly, and naming two risks he says must be avoided for AI to play out safely. It reads as an answer to the skeptics who called last weekend's alignment with Amodei a marketing move.

FROM THE FRONTIER

Made with Chatgpt

Scientists mapped an entire fruit fly brain. Developers taught it blackjack.

The map. Researchers at HHMI Janelia and Google Research published a complete wiring diagram of a male fruit fly's brain: 166,000 neurons and 125 million connections between them, the largest such map ever built. They made it by using AI to stitch millions of two-dimensional microscope slices into three-dimensional reconstructions of every neuron, then tracing where each one connects. The whole thing is public.

The playground. Within days, developers started running the map as a simulation. Evan Sinclair Smith built NeuroCraft Fly, a Minecraft project that runs all 166,000 simulated neurons, using GPT-6 Astra to do the heavy lifting. Fan Pu Zeng, formerly of Jane Street, wired the fly to a blackjack table you can play against. Others have it solving a Rubik's Cube, playing Beat Saber and walking around in a small physical body. Google's own thread collects the best of them.

Why a fly. Because it is the right size. A fly brain is small enough to map completely with today's tools and complex enough to do real things: see, navigate, learn, court. The map includes the visual-motor pathways that let the simulated fly detect its surroundings and move in response, which is why it can play games at all. It is a working brain you can read line by line.

The ladder. Fly, then mouse, then monkey, then human. Each step is roughly a thousand times larger than the last, and the human brain has about 86 billion neurons against the fly's 166,000. Nobody expects a human map soon. What the fly gives neuroscience is a testbed: a complete circuit where you can check whether a theory about how brains compute actually holds when every connection is known.

The reframe. The blackjack demos are silly on purpose, and that is the point. A fully mapped brain has gone from a research artifact to something a hobbyist can run over a weekend, the same way open-weight language models did two years ago. The interesting question is not what the fly can play. It is what happens when the mouse map lands and the same crowd gets hold of it.

IN THE KNOW

What people are actually watching and sharing

The speakerphone moment. Huang was handed a foldable phone on stage, could not get the president onto speaker, and Trump joked that a man who designs the world's most complex chips cannot work a phone. Bloomberg has the exchange; the clip is everywhere.

Everyone dies, explained. A detailed video from March laying out how AI could realistically end humanity, built on the 2025 book If Anyone Builds It, Everyone Dies, has resurfaced on the back of the slowdown debate. Long, sober and worth an hour if you want the strongest version of the case.

Built with Astra, shipped for free. A developer's video game demo pulled 3M views, so he wrote up every step of how he made it with GPT-6 Astra. Skip the thread and just play Paper Route in your browser.

Siri, day one. iOS 27 started rolling out Monday with the Siri AI beta on board, which means the assistant Apple has delayed twice is finally on real phones. Expect a week of screenshots, and judge it on what it does with your own apps, not the demos.

Three tools worth a look. Oats is an open, local and free AI meeting note-taker, which is rare in a category full of subscriptions. Saydi does real-time voice translation for work calls, and Claras turns any YouTube video into a chat you can interrogate.

Turn Claude into a research assistant that remembers

Most AI research starts from zero every session. This setup fixes that. Open Claude, create a Project named for your topic, add the reports and notes you keep coming back to, and paste the instructions below into the project's instructions field. From then on every task in that Project starts with your sources loaded and your past work in memory, and Claude tells you what has changed since you last looked.

COPY AND PASTE THIS PROMPT

Act as my research assistant for [TOPIC]. Prioritize credible primary sources over commentary, and distinguish clearly between established facts and interpretation. Cite every source you rely on. Flag any place where the evidence conflicts rather than smoothing it over.


Organize every answer into four sections: Key Findings, Evidence, Implications, and Open Questions.


Remember useful context from previous work in this project and build on it rather than starting from scratch. When something I ask about has materially changed since my last analysis, say so first.


First task: research the latest developments in [TOPIC] and update my existing analysis with anything materially new.

Swap [TOPIC] with something narrow enough to track: "AI regulation in the EU", "competitors to our water filter store", "CRISPR delivery methods". Advanced tip: keep one Project per subject rather than one giant one, so the memory stays clean, and start each new task with "update my analysis" rather than a fresh question to make Claude diff against what it already knows.