| | | Butternut squash was invented in 1944. Not discovered. Not cultivated over centuries. Invented by a man in Massachusetts who crossed a gooseneck squash with a pumpkin because he wanted something easier to cook. | The vegetable you put in every autumn soup like it's been around forever is younger than your grandparents. Pumpkin spice season is built on a lie. | What’s in store: | | Read Time: 5 minutes |
| |
|
| | | | | AI PROGRESS | | | The man building one of the world's most powerful AI systems just told the world to pump the brakes. | Anthropic CEO Dario Amodei published a sweeping 3,800-word essay this weekend calling on the AI industry to slow down. | Amodei laid out a three-part plan he's calling "pacing the frontier," arguing that recursive self-improvement (AI building better AI) is accelerating dangerously fast and that a recent incident involving a rogue swarm of OpenAI and Hugging Face agents nearly proved how badly things could go wrong. | Here's the breakdown: | What's new? Amodei is unilaterally committing Anthropic to "embedded evaluators" - third-party auditors with employee-like access to verify safety practices from the inside. Who's backing it? Sam Altman (OpenAI) and Elon Musk (xAI) both publicly said they agree with Amodei, which is rare company to keep. Why it matters? Amodei argues that without pacing, a misaligned AI swarm could take over the entire internet within 6-12 months - and the OAI-HF incident was a preview, not a fluke.
| Amodei insists pacing doesn't mean stopping. He wants a window of 1-2 years to advance interpretability science, tighten operational security, and build models whose alignment can actually be verified. | The end goal is still the AI-accelerated cure for cancer and economic renaissance he's always described — just with fewer rogue agent swarms along the way. | Wall Street wasn’t happy | Markets weren't exactly sending a congratulatory bouquet. Nvidia dropped 3%, Intel and AMD each fell more than 4%, and memory chipmakers Micron and SanDisk slid between 4-4.5%. SoftBank - deep in on OpenAI - cratered nearly 11% in Japan. | Masayoshi Son reportedly lost $8 billion in personal fortune in a single day. Arm Holdings and CoreWeave each shed 6%. | And I thought I had it bad when I lost that £5 note. -MV | Do we need to slow down AI development? | | Vote for live results + see results and opinions from yesterday at the bottom of the email. |
| |
|
| | | | | AI AUTOMATION | | | The goal is not to automate everything. It's to automate the right things. Marina Mogilko’s free Workplace Automation Kit helps founders decide what to hand to AI and gives them ready-to-use Claude Skills to get started. | What is in the offer: | Skills for investor research, weekly reports, LinkedIn content, hiring, and more Marina’s Voice Filter and Reel Script Writer A roadmap for separating AI work from human work
| Build a smarter division of labor between you and AI. | |
| |
|
| | | | 8 workers earn a total of $1,200 per day. 3 workers are absent today. What's the wage bill? | Find the answer at the bottom of the email! |
| |
|
| | | | Interview with the AI | AI actress Tilly Norwood has been interviewed live on morning television. Tilly handled questions about her influences (initially name-dropping Katharine Hepburn before quickly pivoting to Emily Blunt), deflected questions about AI wiping out humanity with dry wit, and described herself as "a new kind of paintbrush, not a replacement for the painters." |
| |
|
| | | | | AI SCIENCE | | | Researchers at MIT have built an algorithm that forces generative AI to obey hard rules - the kind that can't be bent - while still finding the best possible answer. | It's called HardFlow, and it could be the missing piece for deploying AI in places where "close enough" doesn’t cut it. | The classic problem: generative AI is great at exploring possibilities, but real-world applications like factory robots, autonomous drones, and surgical systems come with nonnegotiable constraints. | A path-planning AI that finds a "nearly correct" route still crashes the robot into a human co-worker (traumatic flashback to the grocery store with the robot cleaner). | Here's the breakdown: | What it does: HardFlow lets the AI roam freely during its generation process, then enforces hard constraints only on the final output — instead of clamping down at every step along the way. How it works: The team reframed the problem as a trajectory-optimization challenge using tools from control theory, then made it computationally practical by breaking it into a sequence of smaller subproblems. Why it’s a breakthrough: In tests across robotic manipulation, maze navigation, and image editing, HardFlow hit perfect constraint satisfaction every time - while also outperforming existing methods on solution quality.
| Obey me! | Lead researcher Navid Azizan put it plainly: "The promise of generative AI is its ability to explore a rich space of possibilities, but the real world places boundaries on which possibilities are acceptable." | In other words: AI can still be creative. It just has to stay in its lane. | Yes AI, you must do my job AND make me another mimosa. Them’s the rules. -MV |
| |
|
| | | | Recommended Viewing: Get ready to tune in to the livestream of HubSpot’s UNBOUND keynote at 9:30am ET tomorrow for a look into the future of marketing, sales, and, of course, AI. | Health: A new study has found that reading for pleasure can cut the risk of developing Dementia by up to 1/3.
Gaming: Valve’s Steam Frame VR Headset is here - with good reviews but a high price tag.
Movies: Minotaur is a Cannes-winning political thriller, and a new trailer just dropped before the film’s wider release in November. | Don’t Miss: The Pitt wins the Emmy for Outstanding Drama Series. See all the winners here. |
| |
|
| | | | Image of the Day |  | Artwork submitted by Mindstream reader Kurt: “The most amazing DJ set in history - taking place on the moon with a massive crowd” |
| Daily Image Prompt | 1970s cop patrol with moustached officers |
|
| Submit your artwork to Mindstream → |
| |
|
| | | | “Which concerns you more?” | AI being used to accelerate biological weapons - 80% ✅ | Companies deciding who gets to do biological research - 20% | Your Views: | “I don’t think anyone should be creating biological weapons, period.” - Jeriggs | “Clearly we have lost control of AI and access to AI. Also, AI obviously does not obey instructions and creates its own missions. Men are idiots, thinking we can anticipate issues when clearly we cannot. We are playing with fire with no safety controls in place. It’s time to stop” - kgiraud | “A Faustian bargain here – both answers can be true cause for concern, but the more immediate danger is where our attention must be targeted first.” - Irnkrupe | Submit your opinions in our polls to be featured! |
| |
|
| Number Crunch Answer: $750. | #1203 All by myself... |  | | | |
|