| | Welcome, humans. | Okay, so hereās some dumb fun for your Sunday viewing pleasure: somebody turned the internetās favorite classic memes into one continuous stroll down āMeme Streetā: | | 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 todayās AI can now credibly complete most undergraduate assignments, which is either the end of homework or a strong argument for turning Claude into your tutorās tablet. Basically, itās Tom Riddleās diary, except it quizzes you on organic chemistry instead of unleashing a basilisk on your friendās sister. | Hereās what happened in AI today: | šŗ Anthropic taught agents to operate real lab hardware š° Claude beat 28 humans at alignment research š° Z.ai's GLM-5.3 found 2,436 open-source bugs š° South Korea picked free AI for 52M people š Make Claude cite every spreadsheet number
| ā¦and a whole lot more that you can read about here. | | | šŗ 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 yāall. | 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. | |
|
| |
Watch the on-demand demo | |
š 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?Ā Request it here. | |
š° Around the Horn | | OpenAI is ending Cursorās direct model access on Nov. 12 as a reaction to SpaceXās acquisition of it, citing ātrustā 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 agentās 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 Roboticsā 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 = AIās ārulesā 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! | |
|
|
| *Generate Music, Speech, and Sound Effects for creative projects, plus Firefly AI Assistant and leading models like Gemini, Runway, and Kling. Try today! LLM ClichƩ 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.
| |
š Sunday Special | | Top 5 Stories of the Week | | 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 with Noam Schwartz, CEO and co-founder of Alice. Click the image above to watch on YouTube |
| New episode:Ā Why AI agent security gets āalmost infiniteā 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. Itās about to get weird yāall. | Watch and/or Listen on your favorite platform: YouTube | Spotify | Apple Podcasts | P.S: Weāre trying to hit 50K subscribersĀ on YouTube this year. Click here to help! | | |
 | good problems only |
| | | Thatās 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? | |
|
| P.S: Love the newsletter, but only want to get it once per week? Donāt unsubscribeāupdate your preferences here. |
|
