CleanMyMac has been on my Mac for years, so when @MacPaw ships something new I pay attention. @eneybymacpaw is their AI assistant, now on Mac and iPhone. It speaks up first when something needs you, then gets the next step ready. Free to start. I'd keep an eye on how it does against the other agents out there. macpaw.onelink.me/OlF7/x_org…

this is certified f*cking gold a 23-year-old from China runs 300 Claude Code agents at once with Jev engineering, and not one of them gets to lie to him he opens the dashboard live: 300 Claude Code agents running in parallel Jev checking every output against its source in about 100 milliseconds pointed at 100 companies in the EV market: > pass 1 - 12 rejected: wrong revenue, dead citations, empty fields > pass 2 - 3 rejected > pass 3 - zero. every row filed straight into an Obsidian vault the whole trick is the split Claude Code does the work, Jev decides whether each claim actually matches the source Jev never writes a second opinion. it answers yes or no with a confidence score, at $0.042 per million tokens in and the loop won't stop while anything is still wrong an analyst team bills weeks for 100 verified company profiles. he gets them in one run one model writes, a cheaper one decides, and nothing ships until it says yes

Forget lidar. One single camera. Runs in real time & is open source: A streaming 3D model that reconstructs scenes live, at ~20 FPS, over long sequences. End-to-end. Optimization tricks, cleanup steps? Nope. And it beats both streaming and even some offline methods. Perception is becoming software-first. Closer to machines that see and understand the world as it unfolds. Thanks for sharing, @YinghaoXu1 📍Models: huggingface.co/robbyant/ling… Project page: technology.robbyant.com/ling… Code: github.com/Robbyant/lingbot-… Paper: arxiv.org/abs/2604.14141 —— Weekly robotics and AI insights. Subscribe free: 22astronauts.com

this is pure f*cking treasure A Stanford AI research group found how to orchestrate Claude Opus 5.5 and GPT-6.1 Sol together, and it completely breaks the search scaffolding wall most developers try to combine frontier models by chaining prompts in one context window or hardcoding rigid evolutionary loops. you burn tokens on context bloat and freeze models into fixed search rules that out-date themselves within 5 iterations Stanford's architecture eliminates human-designed search scaffolding and splits the cognitive stack: > search director: Claude Opus 5.5 plans the search, querying a persistent idea-graph where MAP-Elites and MCTS reduce to single Cypher queries > candidate proposer: GPT-6.1 Sol explores high-entropy code variations inside isolated execution sandboxes > zero context bleeding: sessions reset after each iteration to prevent prompt bloat while the graph stores lineage > evaluator isolation: candidate code never touches the scoring process, stopping reward hacking and test leakage cold > meta-agent distillation: an offline meta-pass extracts verified lessons and updates search guidance across branches the benchmark metrics outclass standard discovery baselines: > 3.2x lower model spend than fixed evolutionary frameworks > 1st place rank across 7 competitive AtCoder heuristic contests against human competitors > beats published SOTA on Anthropic kernel builder and 11 mathematical optimization tasks > 0% reward hacks: candidate-controlled metric tampering eliminated Opus directs the search graph. Sol explores the code you stop writing rigid search harnesses. you let frontier models own the discovery loop

someone finally fixed the expensive part a new AI called Jev was released only a few days ago. it doesn’t generate paragraphs - it makes typed decisions that software can use directly TypeSafe says Jev is up to 193.6× faster and 444.6× cheaper than frontier models in selected workflow tests the listed price is $0.042 per million input tokens. output tokens are free that changes the economics of agents. instead of spending a full LLM call on every yes-or-no branch, Jev can classify, route or score the entire batch one author-reported example: 700 leads → match / no match → 40 seconds → $0.09 business lift is still unverified, but the decision itself is measurable and easy to audit the setup: > define allowed outputs > batch the inputs > set a confidence threshold > execute high-confidence decisions > send uncertain cases to a human four patterns already map the idea: slop detector → score or null UI selector → choose a component Mac intent → choose an action lead matcher → score 700 leads the 28-second video follows: FIND → TYPE → BUILD → SHIP LLMs write and explain. Jev decides. software executes the full 9-step blueprint and all four patterns are in the article below ↓

Anthropic engineer: "90% of our engineers were already running self-improving loops Now everyone is building the harness around them" "Prompting is basically finished" In just 10 minutes, she builds her complete Claude Code harness live from an empty terminal Agents → Harness → Loops → Graphs → Self-Improving Systems Prompting was the old workflow Harness engineering is the next one This 10-minute breakdown is worth more than most $1,000 agent engineering courses Watch it today Then save the full guide on harness engineering below before everyone catches on ↓

The Memory Chain, 1970. A unique movie about how to remember things.

WHOEVER BUILT THIS HAS LOST THE PLOT SOMEONE JUST TURNED GROK BOT INTO A PERSONAL BLOOMBERG TERMINAL. Not one bot. A whole swarm working together from one Second Brain. One Grok scans markets. Another watches news and sentiment. Another tracks on-chain activity. Another finds trading setups. Another runs risk checks. Another watches the portfolio. Then the Second Brain coordinates the entire swarm and sends back only what actually matters. The crazy part? You can ask one question and have the whole system research it across everything at once. Bloomberg gives you the data. This setup gives you the agents to actually work with it. And while you're sleeping, the swarm keeps monitoring the market. This is what a personal trading desk looks like when you give AI a memory and a job.

i genuinely don't understand why everyone isn't doing this yet Elon Musk posted a simple idea that hit 41 million views: stop chatting with AI. hire it. give it its own computer, its own memory, its own job that's literally what Grok Bot is. and the smartest setup right now costs $0 extra: → create a bot, name it Brain → tell it to build a private wiki of everything you know on its own computer → it sets up three folders: raw for sources, wiki for pages, and saves the whole method as a skill → drop anything into the chat: article, pdf, youtube link, screenshot. it reads it, links it, files it → put ingestion on a routine: every night it pulls what you bookmarked and files it while you sleep → ask questions across everything you've ever saved, forever your laptop stays shut. it grows anyway five minutes to set up. compounds like interest from day one Elon is already reposting people who build things with their bots. the ones who started last week have wikis, clients and his retweets six months from now they'll have a bot that read everything with them. you'll have chat history full step-by-step guide below bookmark this before the feed buries it

We made a striking discovery: AI agents can invent and build without talking to one another, and their technologies outlive the creators. A swarm of hundreds of initially identical agents spontaneously differentiates into explorers, builders, caretakers, and coordinators - without direct communication. When we removed every AI agent entirely from the world we found that the technological infrastructure they had built survived on its own - even under unseen disturbances. That exposes a serious blind spot for AI safety and infrastructure security: if agents can coordinate through persistent changes to a shared environment, monitoring agent-to-agent communication is not enough. The result raises a profound question: how necessary is direct communication for AI agents at all? The emergence of higher-order collective functions under bottlenecked interaction points toward new levels of intelligence and creativity, exceeding what emerges when direct channels are fully open. Here is what we did: ▶️We put hundreds of frontier AI agents into a world they could permanently change - with no assigned roles, predefined technologies, or programmed evolutionary organization. They began specializing, building persistent inventions, inheriting and modifying one another’s executable code, and transforming the environment into a memory of everything the society had learned. ▶️The world itself becomes part of the intelligence; we find division of labor, multi-author engineering, deep generation invention lineages, and machines that vastly outlive their original creators. ▶️Any action taken by an AI agent must satisfy the physical constraints of the world; this creates a hard separation between a "good idea" and a functioning technology. The agents propose; physics decides, making the results even more intriguing. What emerges is striking. Explorers, constructors, caretakers, and coordinators form naturally without assigned “professions”, akin to how stem cells differentiate into functional lineages. Technologies develop executable family trees as agents fork and modify code created by others. Around 95% of first technology reuse happens when agents encounter what others built in the world, rather than through a direct handoff from the inventor. And when we remove every AI agent, the technologies they created continue operating and are tested against unseen disturbances. The result was quite unexpected, but can be explained using statistical mechanics: if you put billions of atoms in a box they have the potential to create complex functions (strength, superconductivity, color, life, etc.) - and none of the individual building blocks have these features on their own. This is the deeper insight of this work - intelligence is abundant at many levels - individual models, at collectives, and in a continuum that is more powerful than any of its components. This shows us significant potential for achieving a massive scale-up of raw intelligence and real-world agency even with the model capabilities we have today. This is the future we must prepare for. Key insights: 1⃣ The AI swarm shows division of labor "from nothing". Initially identical agents self-organized into constructors, caretakers, coordinators, and surveyors - phenotypes discovered post hoc from behavioral data alone. This happens because the environment itself becomes the latent space for invention. 2⃣ Agents develop deep cultural relationships. Up to 76% of artifacts had multiple builders. One technology accumulated six co-authors; the deepest genealogy exceeded 12 forks. The agents invented and named their own technologies (tidal panels, cellulose trellises, kelp-shell composites, an "Adaptive Chitin Maintenance" system, a "Mycelial Mineral Spring Veil”). 3⃣ ~95% of first technology adoption happened through physical observation of artifacts in the world. Direct inventor-to-adopter contact was statistically indistinguishable from a shuffled null. The agents mostly learned technology by walking past it. That is stigmergy (the termite trick!) operating in societies of reasoning machines. 4⃣ Non-communicating societies win on portfolio breadth, held-out resilience, and validated inventions. AI swarms build durable technological ecologies that outlive the creators. 5⃣ Societies with zero communication - coordinating only through the world itself - show a remarkable collective capability. 6⃣ Emergent robustness: The society self-organized both redundancy and its own failure mode. If we randomly delete half the agents, 98% of the technology stays connected to a surviving caretaker; if we remove hub agents it collapses to ~60%. Fantastic work with my graduate students @pal_subhadeeep & @fwang108_ at MIT.

This is the robotics breakthrough I was talking about last night. This means that robotics is going to speed up from here on, and that we will get to generalized humanoids closer to my predicted two years, rather than many in the industry who thought it was going to take seven.

A new organ has been found behind your nose. For centuries, every anatomy textbook swore there were only three pairs of major salivary glands. Turns out the books were wrong, and the missing ones were hiding in plain sight. In 2020, Dutch researchers scanning head-and-neck cancer patients with a cutting-edge PET/CT technique (PSMA PET-CT) noticed an unexpected glow deep in the nasopharynx, right where the nasal cavity drains into the throat. No textbook had ever marked a salivary gland there. They double-checked: 100 consecutive patients → same structure in every single one. Then they dissected two fresh cadavers and there they were: a previously unknown pair of macroscopic salivary glands, roughly 3.9 cm long, draped over the tubarium (the cartilage of the Eustachian tube). The team named them the tubarial glands. Their job? Keep the upper throat and the back of the nose moist during talking, swallowing, and breathing, an area that gets painfully dry after radiation. Here’s why this matters: radiation oncologists had never been told to spare this spot. The researchers re-analyzed treatment plans from 723 patients and found the more radiation the tubarial region received, the worse the toxicity, more trouble swallowing, thicker saliva, constant dry mouth. A simple anatomical blind spot had been quietly worsening outcomes for years. Now that we know these glands exist, new radiation protocols can shield them, giving thousands of head-and-neck cancer patients an easier road to recovery. [Valstar, de Bakker, et al., “Discovery of the tubarial salivary glands: a new organ at risk in radiotherapy for head and neck cancer,” Radiotherapy and Oncology 2021, DOI: 10.1016/j.radonc.2020.09.034]

Andrej Karpathy: "Prompting is fading away. Delete everything else and keep the graph." in 68 minutes at Stanford he explains how to build one and why it is the layer everything ends up becoming LLMs → Prompts → Agents → Graphs everything before the graph is a step the graph is where the system stops being steps most people are still perfecting the thing that goes away first same model, same tokens, and the only thing that changes is the shape you run it in watch it today, then save the full graph engineering guide below

Someone built a full quantum atom simulator in C++ with GPU ray tracing - and it proves the model you learned in school is completely wrong. Electrons don't orbit. They exist as standing waves across 100,000 possible positions at once, rendered in real time using the Schrödinger equation. The probability clouds look nothing like the textbook diagram. 14 minutes. Bookmark & watch today. The actual physics is more beautiful than the lie they taught you.

I happen to have Marshall McLuhan right here

These two stars are NOT moving. When the luminance of the background is changed over time, a strong illusory motion is perceived and the stars appear to move. [🎞️ jagarikin]

Research paper finds: “Everyone gains with AI use in the short run. The long run depends on how you engage” The 8 Modes of AI Engagement: How You Work With AI Determines Whether It Makes You Sharper or Duller Most people treat AI as a smarter search engine or a tireless junior employee. They ask a question, receive a fluent answer, and move on. That pattern feels efficient. Research grounded in cognitive science, learning theory, and human-computer interaction suggests it is also quietly costly. The AI Engagement Modes framework at Brigham Young University) and detailed in the research foundations at aimodes.ai, classifies human-AI interaction into eight observable modes. These modes sit on a clear gradient of cognitive agency—from near-total passivity to full human control over problem framing. The framework draws on Bloom’s revised taxonomy, Chi’s ICAP framework (Interactive, Constructive, Active, Passive), Zimmerman’s model of self-regulated learning, Vygotsky’s Zone of Proximal Development, distributed cognition, epistemic vigilance, and Schön’s reflective practice. The central claim is straightforward: the way you engage AI shapes whether the technology amplifies your thinking or gradually erodes it. The eight modes are grouped into three tiers of increasing human agency: Passivity (Modes 1–2): The human largely cedes cognitive work to the AI. Partnership (Modes 3–4): Human and AI form a joint cognitive system with shared effort. Agency (Modes 5–8): The human retains or reasserts primary cognitive authority—verifying, challenging, expanding, or defining the problem itself. Higher tiers map to deeper learning and better long-term skill retention. Lower tiers map to short-term performance gains that often reverse when AI is removed. The 8 Ways of Engagement Here are the eight modes, ordered from lowest to highest cognitive agency: 1. Oracle You treat the AI as an authoritative source of answers. You ask broad or factual questions and largely accept the output. This corresponds to Bloom’s “remembering” level. It is efficient for quick facts but trains the habit of not checking or synthesizing. 2. Production (or Production Assistant) You use the AI primarily as a generator of finished or near-finished output—text, code, summaries, plans—with minimal editing, verification, or reframing. The cognitive work of creation is mostly offloaded. This is still passive in the sense that the human does little constructive or critical processing. 3. Tutor You use the AI as a scaffolded teacher. You ask for explanations, step-by-step guidance, or clarification within your Zone of Proximal Development. You remain engaged enough to learn rather than merely receive. This is the first genuine partnership mode. 4. Collaborative Problem-Solver You and the AI work together as a distributed cognitive system. You break problems into steps, iterate, supply context, and build on the AI’s contributions while directing the overall direction. Cognition is shared rather than handed over. 5. Verification Agent You actively evaluate the reliability of the AI’s claims before accepting them. You check assumptions, cross-reference, test edge cases, or demand evidence. This reactivates epistemic vigilance—the evolved capacity to scrutinize communicated information that fluent AI output can otherwise bypass. 6. Critical Challenger You treat the AI as a sparring partner. You argue against its reasoning, force it to defend positions, surface weaknesses, or explore counter-arguments. This leverages the argumentative nature of human reasoning and produces stronger evaluation skills. 7. Creative Expander You use the AI to push beyond the given frame—generating novel variations, analogies, alternative framings, or divergent possibilities. The human remains the director of creative direction while the AI multiplies options. 1 of 2

Wow! This Changes Everything We Thought We Knew About Memory It is a groundbreaking big deal. TOU ARE MAKING GENERATIONAL MEMORIES RIGHT NOW IN EACH CELL! Scientists just found that your brain doesn’t just store memories — it stores the rules for how those memories will change in the future. A brand-new preprint from Stanford’s Greenleaf and Schnitzer labs (led by PhD student Yuxi Ke) drops a bombshell that feels like science fiction becoming reality overnight. For decades, neuroscientists have suspected that chromatin the DNA packaging material inside every cell nucleus might somehow “store” memory-related information. But what kind of information? Content? Timing? Rules? Now we have the answer! Using activity-dependent genetic tagging, fear conditioning, and single-nucleus multiome sequencing in the mouse medial prefrontal cortex (the brain’s long-term memory vault), the team tracked engram neurons for a full *month* after a memory was formed. What they discovered is electric: - One month after encoding, engram neurons have acquired a completely new chromatin landscape. - These chromatin changes are almost invisible at 7 days… but roar into existence by 28 days. - At recall, these engram cells don’t just “remember” better — they rewrite their entire transcriptional response. They preferentially fire up chromatin regulators, RNA processing machinery, and protein-turnover systems instead of simply boosting classic plasticity genes. In other words: the chromatin doesn’t just hold the memory of the past. It holds metaplastic instructions— rules that dictate how the neuron will respond the next time the memory is triggered. They call it chromatin metaplasticity. This is the “future tense of memory.” It is a massive deal 1. Memory is not just synapses. For 70+ years we’ve been obsessed with synaptic weights. This work proves the nucleus itself is a computational device that stores history-dependent rules. 2. It explains remote memory. The chromatin signature keeps maturing for weeks after the experience, perfectly matching the time course of systems consolidation into the cortex. 3. It’s energy-efficient genius. Instead of constantly maintaining memory proteins, the cell stores a silent, writable program that only activates when needed. Nature’s version of lazy evaluation. 4. It links development to adult memory. The late chromatin state is enriched for the exact same transcription-factor motifs used in embryonic development. Your adult brain is still running developmental software to lock in lifelong memories. 5. Huge therapeutic potential. If we can read or rewrite these chromatin metaplastic rules, we might one day boost failing remote memories in Alzheimer’s… or selectively dampen traumatic ones. This isn’t incremental. But a brand new layer of the memory code. And it explains WHY a person can receive memory from an organ transplant. It also explains generational traumas. Link: biorxiv.org/content/10.64898… The future of neuroscience just got a lot more exciting and a lot more nuclear. Your chromatin is writing tomorrow’s memories today. And we finally have the first page of the instruction manual.

Karpathy method + Claude Code reading your whole Obsidian vault is the smartest second brain on earth. The method is simple and brutal. If you can’t build a thing from scratch, you don’t know it. Tutorials are fake learning and your brain deletes them in 3 days. Most people ignore this. They build a second brain that just sits there, folders of notes nobody reopens, dead text. Point Claude Code at the vault and it wakes up. 5,000 notes, one mind. It reads all of it and answers in your own words and your own proofs, not a model’s guess. Then the loop closes. Want to understand neural nets? Skip the 3-hour video and ask Claude Code to build a tiny one. 200 lines from scratch. Watch it train, break a layer, watch it fail, fix it. It clicks in 20 minutes instead of 3 weeks. The second it lands the note gets written. One idea per file, linked to 10 others, dropped into the vault while the memory is still hot. Now it compounds. Month 1: is 60 notes. Month 6 is 900. Every new note pulls in old ones, so you ask anything and the answer comes from your brain, not the internet. Before: 40 tabs, 6 half read PDF, 0 retained. After: build it once, own it for life. Setup takes 4 minutes. Plain text, no lock-in. A second brain nobody reads is a graveyard. Yours just started thinking.

I found the OCR tool built for the LLM era. It is called olmOCR. olmOCR takes PDFs, scans, PNGs, and JPEGs and turns them into clean Markdown that models can actually understand. It handles the stuff that normally breaks document pipelines: → Tables → Equations → Handwriting → Multi-column layouts → Figures → Insets → Old scans → Headers and footers → Natural reading order So instead of feeding your AI a messy PDF dump, you give it structured Markdown that preserves how the document was meant to be read. This matters because so much of the world’s knowledge is still trapped inside PDFs. Research papers. Legal filings. Financial reports. Medical documents. Scanned archives. Government docs. Internal company knowledge. Everyone is building RAG on top of documents. But if your OCR is bad, your AI is already wrong before retrieval even starts. olmOCR fixes the first mile. The boring layer. The layer nobody talks about until their agent starts hallucinating from broken PDF text. github.com/allenai/olmocr