Marketing Technology & Analytics Lead

Anthropic wants Claude operating real lab gear. Plus: Welcome, humans. Okay, so here's some dumb fun for your Sunday viewing pleasure: somebody turned the internet's favorite classic memes into one continuous stroll down "Meme Street": Now for something actually cool: someone built penombra, a handwriting notebook where you write with a stylus and Claude writes back on the page. It can read PDFs and ebooks, respond to what you annotate, explain passages, and quiz you. It runs on Android tablets with a stylus, and the creator is taking early-tester signups here. Pretty good timing: MIT says today's AI can now credibly complete most undergraduate assignments, which is either the end of homework or a strong argument for turning Claude into your tutor's tablet. Basically, it's Tom Riddle's diary, except it quizzes you on organic chemistry instead of unleashing a basilisk on your friend's sister. Here's what happened in AI today: * | Anthropic taught agents to operate real lab hardware * | Claude beat 28 humans at alignment research * | Z.ai's GLM-5.3 found 2,436 open-source bugs * | South Korea picked free AI for 52M people * | Make Claude cite every spreadsheet number Anthropic gave AI agents a common language for machines. AI agents to date have mostly lived inside browsers, terminals, and spreadsheets. Anthropic now wants the same kind of agent to walk into a lab and know how to use the machines. Here's what happened: * Anthropic and HHMI Janelia opened a research preview of the Model Hardware Standard (MHS), a shared interface for programmable lab and factory equipment. * Each device gets a standard driver with simple read/write commands plus plain-language tags describing what it does and its safety limits. * Early partners used MHS at Genentech, Carnegie Mellon, and QuEra; QuEra's agent-built script recovered a quantum laser's lock in 695 of 700 trials. Why this matters: Today, every microscope, plate reader, or robot arm tends to need its own custom AI software integration. MHS gives agents one interface for discovering equipment, sequencing work across devices, and turning successful procedures into scripts. The payoff is less integration glue to connect agent software with physical hardware. Anthropic says setups that usually take weeks or months can fall to hours or minutes, making round-the-clock automated experiments much easier to build. This is a fast takeoff scenario y'all. Check yourself: This is still a limited research preview, not an autonomous scientist in a box. Anthropic says Claude still needs expert oversight for physical reasoning, and MHS currently requires hardware with a programmable interface. But if the standard catches on, the next major agent platform may be the layer connecting models to the physical world. Researchers and manufacturers can apply for access here. FROM OUR PARTNERS Your prospects won't sign without proof of security. A proof of compliance request stalls deals and pulls engineers into audit prep. Vanta gets you compliant fast (SOC 2, ISO 27001, HIPAA+) and keeps you that way. Trusted by 16,000+ companies like Ramp, Harvey, and Writer. AI skill of the day: Make Claude show every number it touched. AI can sound confident about a workbook while quietly skipping the cells you care about. Claude for Excel can cite exact cells and highlight edits, so make it prove coverage before changing anything. * Ask for a coverage ledger: every sheet or range reviewed, skipped, or ambiguous. * Require cell-level citations for each conclusion and a log of every formula or value it proposes changing. * End with unresolved assumptions and a no-edit review pass; approve changes only after checking the cited cells. Review this workbook without editing it. List every sheet or range you inspected, cite the cells behind each conclusion, flag anything you could not verify, and show every formula or value you would change before I approve edits. Have a specific skill you want to learn? Request it here. Around the horn. * OpenAI is ending Cursor's direct model access on Nov. 12 as a reaction to SpaceX's acquisition of it, citing "trust" concerns and prior Musk-company contract breaches. Cursor users can still bring their own OpenAI API keys. * Researchers at Google and Purdue introduced SKILL.state, which keeps an agent's current structured state instead of replaying its full history; on a 100-step Gemini 3 Flash benchmark, it cut token use about 94% (65K vs. 1.06M) while accuracy rose from 0.91 to 0.94. * Pollen Robotics' Microduck (above) can train behaviors in simulation and transfer them to a real 25 cm biped. * Anthropic let Claude spend 48 hours and one GPU fixing 10 alignment failures (alignmnet = AI's "rules" for good behavior), beating 28 human researchers at the job; that said, a monitor did catch the AI test-gaming in 2.4% of roughly 1,600 runs. AI still gonna AI. * South Korea selected SK Telecom, Kakao, and KT for free domestic AI access for roughly 52 million residents, backed initially by 512 Nvidia B200 GPUs. * The EU AI Act entered its first transparency-enforcement phase, giving regulators access to company information and models while stricter high-risk-system rules arrive later. * Gemini Co-Scientist moved into real labs, helping guide materials, biology, and medical-reasoning experiments, including a technique that beat six frontier models in blinded physician review. Want absolutely EVERYTHING that happened in AI this week? Click here! Treats to try. * *Generate Music, Speech, and Sound Effects for creative projects, plus Firefly AI Assistant and leading models like Gemini, Runway, and Kling. Try today! * LLM Cliché Highlighter scans pasted text or a URL for common AI-writing tells and explains the pattern it matched. * Gemini Notebook Expert Intelligence turns eligible Google Play Books you own into sources you can question, quiz yourself on, or turn into audio overviews. * LightReel indexes roughly 10,000 new TikToks a day so marketers can search winning hooks, formats, and creative patterns instead of doomscrolling manually. * Hao AI Lab open-sourced FastH3 v1, a four-step MiniMax H3 distill that ran up to 14x faster on one Blackwell GPU, though its team says motion and fine detail still trail base H3. Try the checkpoint. * Z.ai open-sourced GLM-5.3 after post-training sharply improved coding and cyber performance; it says the model found 2,436 bugs across 269 open-source projects. Download the weights or run it on Tinker. * Tencent open-sourced Hy4 preview, a 770B-parameter model that activates 49B parameters per token and supports a 1-million-token context window. Download or run it here. Sunday special. Top 5 Stories of the Week * Hackers used Cursor to compromise seven real companies. * NVIDIA reportedly moved to buy Hugging Face for about $13B. * OpenAI's Jalapeño chip posted faster, more efficient inference results. * Anthropic's IPO could become one of tech's biggest raises. * AWS and NVIDIA planned 2 million more GPUs. Top 5 Tools of the Week * Construct turns repeatable work into scheduled agent workflows. * BrowserOS Neo gives Claude, Codex, and Cursor a local browser. * Atlaso shares one persistent memory across your AI tools. * Xirp gives coding agents your company's real system context. * Mem Agent tracks unfinished work and follows up automatically. New from the neuron: AI explained. New episode: Why AI agent security gets "almost infinite" once agents can act. Alice CEO and co-founder Noam Schwartz joins Corey and Grant to explain why model safety is only one layer, why prompt injection may never disappear, and why real agent security also has to cover tools, data, permissions, and policies. A chatbot can say the wrong thing. An agent can delete files, move money, change a database, or even influence another agent. It's about to get weird y'all. A cat's commentary. good problems only | / | That's all for now. If you want to get featured above, fill out the poll below and tell us how we did today! | What'd you think of today's email? | | * | Like a hit of catnip * | Good, not great * | It sucked | | Login or Subscribe to participate in polls. P.S: Love the newsletter, but only want to get it once per week? Don't unsubscribe - update your preferences here.

She studies cell lineage in the eye. Now she's leaving research lineage of her own. Award-winning biologist Connie Cepko, who advanced work on vision-threatening disease, passes the torch to scientific successors. Sy Boles Harvard Staff Writer Connie Cepko is not tired. She's not burnt out. And she's certainly no less fascinated by the complicated and beautiful human eye than she was when she started researching it some four decades ago. When Cepko, the Bullard Professor of Genetics and Neuroscience at Harvard Medical School, retires on July 31, it will be on principle: It's time, she says, to pass the torch of scientific discovery to a new generation. "I still love what I do," she said. "But it's time to step aside and make resources available for the up-and-coming junior people who want to do what I do, especially right now with a contraction going on because of the federal government." For decades, Cepko's lab has made major advances toward gene therapies for vision-threatening diseases such as retinitis pigmentosa and age-related macular degeneration (AMD) that affect millions. She has been an investigator with the Howard Hughes Medical Institute since 1994 and has won numerous prestigious awards, including the Bressler Prize in Vision Science and the Friedenwald Award for research in ophthalmology. Cepko didn't start out decades ago with treatments in mind. Her early passion was for understanding the development of the retina. In the 1980s, she used retroviral vectors to tag retinal progenitor cells and the process by which a common set of "mother" cells produced the stunning 120 cell subtypes involved in human vision. Later, researchers in her lab would demonstrate a remarkable oscillating pattern of cellular development, the first and only glimpse into the timeline of cell genesis at that level of precision. It was an unexpected phone call that turned her attention to therapeutics. "After about 20 years of doing developmental biology, I got a call from a guy named Alan Schwartz, who just coincidentally was the president of the U.S. Tennis Association," Cepko said. "His opening line was, 'What are you doing about blindness?' Just out of the blue. I'd never heard of this guy, and I'd never had anyone ask me that." "He asked what I would do if it was my own child. I had to own up to it. I said, 'Well, I'd probably work on it.'" Schwartz's grandson had been born with Leber congenital amaurosis (LCA), a rare genetic retinal disease characterized by the dysfunction of the photoreceptors - the rods and cones that capture light and make vision possible. Children with LCA are born blind or nearly blind. At the time, there was no treatment for LCA. Cepko explained that she was a basic scientist. She did the research that, she hoped, clinical or translational researchers would develop into therapies. Schwartz didn't buy it. "He asked what I would do if it was my own child," Cepko said. "I had to own up to it. I said, 'Well, I'd probably work on it.'" So she did. In the process of tracing retinal cell lineage, her team had identified some of the genes that, when mutated, lead to blindness. Researchers had identified hundreds of disease-related genes associated with blindness. Designing a bespoke therapy for each one - a process that can take years and millions of dollars - seemed impractical. So Cepko and her team decided to pursue a gene-agnostic approach. When a person loses vision, it's usually due to loss of function of the cone photoreceptors, which are essential for high-acuity vision and color perception. But in many conditions, including retinitis pigmentosa, the genetic mutation actually affects the rod photoreceptors, which are responsible for night vision. Therefore, Cepko reasoned, the cones had to be dying from a bystander effect: something in the retinal environment causing them to die independent of a genetic mutation. That meant she could potentially design gene-agnostic therapies to address those environmental conditions in any number of disease indications created by any of the hundreds of genetic mutations. Over time, she identified seven genes that could be inserted into the eye via an AAV vector (essentially, an engineered virus) to combat the oxidative stress, inflammation, and metabolic problems that contribute to cell death in many retinal diseases. It took years to thoroughly test each treatment in multiple models of disease to prove that the treatments really were gene-agnostic. "Connie is especially rigorous and thorough in the way she attacks a research question," said Grant Zimmermann, the managing director of business development at Harvard's Blavatnik Biomedical Accelerator, which has worked with Cepko to catalog about 40 inventions from the lab to select the most promising technologies for commercialization. "If she gets interesting results in one of her models, she'll repeat it using two, three, four different experimental variations before she convinces herself she's got the right answer." The data on her gene-agnostic therapies are promising. But demonstrating safety and efficacy in humans will take years. Cepko plans to stay involved as a consultant to those who are carrying the work forward. And she's keenly following her trainees' work as they move forward in their careers, blazing a path that began before Cepko and will follow after. Science proceeds slowly. The discoveries made by one researcher become the therapies developed by their trainees. The naming of one's mentors is the tracing of one's scientific genealogy. (For what it's worth, Cepko's was Richard Mulligan, Mallinckrodt Professor of Genetics and professor of pediatrics, emeritus, at Harvard Medical School and Phillip Sharp, a Nobel laureate, at MIT.) Cepko is just as proud of those who come after her: Emma West, now the co-founder and CEO of the biotech company Digital Biology, whose gifts with experiment design revealed the oscillating pattern of cell differentiation; Ryan Delgado, who is developing an even clearer window into cell lineage; Xiang Ma, whose work on extracellular vesicles could open up new avenues for delivering treatments. (Besides her scientific lineage, Cepko also notes she has a hugely important personal one, which includes daughters Leah and Ellie and her grandchildren.) "Connie has created a training environment in which people are encouraged to grow not only as scientists, but also as independent thinkers," said Ma, who has worked with Cepko for nearly 12 years. "Connie has been a role model for me in rigorously performing good science while being extremely humble," said Yunlu Sawyer Xue, a former postdoc in the Cepko lab. "She's super smart and has a great taste in research directions. I could see science coming out of her lab directly benefiting visually impaired patients in the next couple of years."

Training the next generation. Even with her many breakthroughs, Cepko is clear about what she is proudest of: "My trainees." She has mentored more than 100 graduate students and postdocs and is quick to credit them for the lab's achievements. Cepko's influence extends well beyond her lab. She helped launch and lead Harvard Medical School's Biological and Biomedical Sciences (BBS) Graduate Program - a role she held for more than a decade. She later co-founded the Leder Medical Sciences Program, now the Leder Human Biology and Translational Medicine, which she continues to co-direct today. Paying it forward. Although Cepko is retiring from Harvard and HHMI in July 2026, she plans to keep pushing the therapies she's helped develop for blindness toward the clinical stage. She also hopes to launch a one-on-one mentorship program pairing 12- and 13-year-old girls with female graduate students and postdocs. The goal is simple: two weeks in a lab over the summer, learning how science works and what a scientific career can look like. She is especially committed to reaching girls at the age when many turn away from STEM. It's no coincidence that she was 12 when a mentor first invited her into a lab - and changed the course of her life. Building tools to answer questions. When existing methods fell short, Cepko often built new ones. In the early 1990s, her lab devised the first barcoded viral libraries to track cell lineages in the brain. By giving cells unique genetic tags, her team could identify which cells descended from the same progenitor - even after they had migrated far apart. The approach opened new ways to study development in complex tissues. Media Contact: Halea Kerr-Layton, Media Relations Manager [email protected]