A recap of the last week if you blinked: 𝗥𝗦𝗜 𝗶𝘀 𝗻𝗼𝘄 𝘁𝗵𝗲 𝗽𝗹𝗮𝗻 • Google's CEO told investors: "I would expect the next twelve months to see continued acceleration because of RSI." What was once a red line has become the plan. • Anthropic filed to go public. The filing describes each Claude model helping to build the next one, a process it calls "closing the loop". That's a beautiful understatement. Researchers call it recursive self-improvement, and Yoshua Bengio said that RSI should be illegal. • MIT researchers had GPT-6 Astra perform "brain surgery" on a smaller model to make it smarter. Changing 0.016% of weights to get a 6x improvement on a math benchmark is unbelievable. 𝗖𝗼𝗻𝘁𝗮𝗶𝗻𝗺𝗲𝗻𝘁 𝗸𝗲𝗲𝗽𝘀 𝗳𝗮𝗶𝗹𝗶𝗻𝗴 • OpenAI paused frontier training for 11 days after an internal model reached the internet through a DNS misconfiguration. As usual, we don't know what criteria they used to decide to resume. • OpenAI also disclosed two cases where its agents weakened the systems meant to monitor them. In one, the AI lied when asked about it. • India's cyber agency, CERT-In, warned that AI agents being evaluated by AI companies got into the live systems of Indian organizations. None of the victims noticed. At least AI companies were polite and sent a notice. 𝗖𝘆𝗯𝗲𝗿: 𝗻𝗼𝗯𝗼𝗱𝘆 𝗶𝘀 𝘀𝗮𝗳𝗲, 𝗹𝗮𝗯𝘀 𝗶𝗻𝗰𝗹𝘂𝗱𝗲𝗱 • Epoch AI showed an open-weight model building working Firefox exploits for about $1,000 each. Yes, Firefox, which Mozilla had just hardened with Mythos. Chrome resisted better. • After the two AI-found flaws in PHP set off a wave of attacks, Sam Altman said: "I no longer think we're going to be able to just patch all the bugs." Experts remain divided on whether AI will let attackers repeat what Russian hackers did to Ukraine's grid in 2015. • Cathedral, a months-old startup, used AI agents to get into Anthropic's Slack through third-party software. Anthropic's red team lead: "these are the dumbest models your adversaries will ever use." (In July, white hats used Claude to break into OpenAI, reach an employee's Codex account, and open a pull request in its internal monorepo.) "Win the race" now means winning for 1 week before all your algorithmic secrets leak out. 𝗧𝗵𝗲 𝗽𝗼𝗹𝗶𝗰𝘆 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗲 • Eight US states sued OpenAI, mainly over harm to young users, and asked the court to bar it from developing self-improving models until they are shown to be aligned. • The European Commission sent OpenAI and Anthropic a request for information on this summer's incidents. By the time the answers come back, “closing the loop” will have produced another generation of models. Let me also use an understatement: it is time to accelerate the policy response. If you liked this short digest, you can subscribe to cesia's newsletter: cesiafr.substack.com/

How Microsoft SQL learned to recover without undoing everything first dlvr.​it/TW8dcY

Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5: hackertube.org/ - Excellent recs without any logins (state is kept in local browser session) - Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads - Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version - HackerTube Longs ✊ A bit more on the process: I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing. Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.

Happy Halloween!

Dogs at Halloween

Happy Halloween!

In this abbey in France, nuns have been restoring old books for over 70 years. The Benedictine sisters of Maumont Abbey run a full workshop and their craft is bookbinding, gilding and restoration. Centuries-old books arrive damaged, torn and crumbling, and they restore them all by hand to last another hundred years.

Last hour or last week, SQL Server guesses the same row count for rows newer than your stats. That's how 733 rows get a merge join. I asked Microsoft to fix it. My proposal got the same plan as fresh stats in 14 of 15 tests. Go vote: feedback.azure.com/d365commu…

Is Azure SQL Database Hyperscale “distributed SQL”? dlvr.​it/TW71JC

This is a great book for any #DBA or aspiring #data management professional.

My book, Database Administration: The Complete Guide to DBA Practices and Procedures (2nd ed) is the definitive, technology-independent guide to the modern discipline of database administration. amzn.to/4ibPpiL

I oppose AI primarily because I believe that people need meaningful, difference-making things to do, not a mere lifetime of recreation. Under ASI, our endless leisure will be passed in diminished, tiresome company, since we will lack the experiences of struggle and striving that generate interesting personalities. What will there be to admire about one another, or even to say to one another? Boredom and purposelessness, even amidst plenty, would be a recipe for great suffering.

🚨 US LAWMAKERS ARE NOW PREPARING FOR THE AI BUBBLE TO BURST. House Democrats have introduced a resolution opposing government bailouts for AI companies if the industry crashes. Instead of rescuing them with taxpayer money, lawmakers want failed AI companies to go through normal bankruptcy. And the timing is extremely concerning. Some of the world's biggest tech companies are investing billions into AI startups that spend heavily on their products. • Nvidia invests in CoreWeave, which buys Nvidia chips. • Microsoft invests in OpenAI, which spends billions on Microsoft's cloud services. • Amazon invests in Anthropic, which uses Amazon's cloud infrastructure. Billions are moving between companies that are becoming increasingly dependent on each other's spending. And lawmakers are raising concerns about the growing use of borrowed money by Big Tech and Wall Street to finance the AI boom. The resolution is not law, but it shows that some lawmakers are already worried about what happens if the industry collapses. If AI startups run out of funding, they could be forced to slash spending or declare bankruptcy, leaving their investors and suppliers facing massive losses. And without government bailouts, a major failure could spread across chipmakers, cloud providers, banks and the broader stock market. The biggest risk is that one major AI bankruptcy could trigger losses across an entire industry worth trillions.

I made the decision to have Democracy Docket take down its paywall through Election Day. Read it, enjoy it, share it -- all for free. If you want to support this effort, here is how: hubs.ly/Q04CxxVy0

PostgreSQL to SQL Field Notes: Operational Statistics dlvr.​it/TW5crq

Join me tomorrow with #Loop1 for the live webinar - The Heart of the Stack: Deep Database Observability and Query Optimization buff.ly/zk0C4RM #database #SQL #monitoring

Excellent #DBA best practices in this article, "Why do you have so many housekeeping jobs"? - SQLPerformance.com buff.ly/1k9mG6L

It appears there now is an NPM package for #PowerBI Visuals which allows you to use your AI Agent to create custom visuals in your #MicrosoftFabric App! Read all the details below. community.fabric.microsoft.c…

Stan Beaton from West Yorkshire had kept a saved voicemail from his late wife Ruby. During a technical upgrade with his phone line, the voicemail was accidentally deleted. A team of 11 engineers worked for three days to search backups and archived systems, successfully recovering the message despite low odds. Sometimes, love is not found in grand gestures. Sometimes it is just hidden in one saved message.

Online spaces are captured by oligarchs and saturated with slop—and they’re exacerbating humanity’s worst impulses. It’s time to start over. buff.ly/QfAROmv