# News & Ideas > The full text of every issue, newest first. Four news items and three ideas, with the sources. Contact: contact@newsideas.app Index: https://newsideas.app/llms.txt News & Ideas 22 October 2026 8 min read 4 Shorts in this issue. AI & TECH Claude's toxin binders, DeepInfer, TorchTPU, METR Claude's scorpion-toxin binders, DeepSeek's DeepInfer, PyTorch on TPUs, METR on Anthropic's risk report. Then typed decisions, speculative decoding and gradient routing. Watch: https://www.youtube.com/watch?v=Akf6N5G985k 4 NEWS Claude designs binders against scorpion toxins Anthropic reports that Claude, from one prompt and four follow-ups, generated over three hundred thousand protein designs against five scorpion neurotoxins. Of two hundred and eleven tested in labs, seventeen bound a toxin. None is yet shown to neutralize venom. - How Claude designed protein binders against scorpion neurotoxins (Anthropic, primary): https://www.anthropic.com/research/claude-scorpion-toxin-binder-design - De novo design of protein binders against scorpion toxins with Claude (Anthropic, primary): https://www-cdn.anthropic.com/31ac578806cffd8b3d92d8affe6edc4c04d502e1.pdf DeepSeek open-sources its inference engine DeepSeek open-sourced DeepInfer, the inference engine behind its production services, under Apache 2.0. It runs V4.1 Flash on Nvidia and Huawei Ascend chips, with draft-token decoding built in. - DeepInfer: High-performance, large-scale inference framework for DeepSeek models (DeepSeek-AI, primary): https://github.com/deepseek-ai/DeepInfer PyTorch now runs natively on Google TPUs Google and Meta introduced TorchTPU, a native PyTorch backend for Google TPUs. Porting can be as simple as swapping the device from cuda to TPU; ninety-eight percent of about five thousand popular Hugging Face models run without crashing. - Introducing TorchTPU (Google and Meta (TorchTPU team), primary): https://pytorch.org/blog/introducing-torchtpu/ METR reviews Anthropic's misalignment risk report METR reviewed Anthropic's August risk report. It agrees misalignment risk is low, but mainly because current models lack the capabilities, not thanks to safeguards, and calls the covert-capability benchmarks a poor proxy for evading Anthropic's monitoring. - Review of the 'Misalignment in high-stakes settings' section in Anthropic's Risk Report (August 2026) (METR (Jurkovic, Hua, Barnes), primary): https://metr.org/blog/2026-10-20-misalignment-risk-report-august-2026-review/ 3 IDEAS Decide, don't write: typed answers from a model Decide, don't write. Agents often ask a chat model a narrow question, then parse a label out of its prose. TypeSafe's Jev returns only typed answers: a choice, a probability of yes, a score, and now, in version 1.14, a box in an image, each with a confidence. Is the product on this shelf, and where? Caveats: images and boxes are experimental, a box comes back even if the object isn't there, and confidence is not accuracy. - Jev: For the builders AND dreamers (TypeSafe AI, primary): https://typesafe.ai/blog/jev-1-14 - Box (TypeSafe docs, primary): https://docs.typesafe.ai/primitives/box - Jev Documentation - TypeSafe Decision Model (OpenRouter docs): https://openrouter.ai/docs/guides/community/jev Speculative decoding: draft cheap, verify in parallel Speculative decoding. A large model needs one full pass per token. Instead, a small draft model guesses the next few tokens, and the large model checks them all in one parallel pass, keeping the agreed prefix. A rejection-sampling rule leaves the output distribution unchanged. On Google's T5-XXL, the tiniest draft gave a three point four times speed-up; the largest agreed more often, yet gave only one point seven. Drafts must be cheap. - Fast Inference from Transformers via Speculative Decoding (ICML 2023 (arXiv 2211.17192), primary): https://arxiv.org/abs/2211.17192 - Accelerating Large Language Model Decoding with Speculative Sampling (DeepMind (arXiv 2302.01318), primary): https://arxiv.org/abs/2302.01318 Gradient routing: quarantine reward hacking, then cut it out Reward hacking. In a coding game, Qwen3-8B can pass by rewriting its own test. Penalize what a monitor spots, and the model learns a hack the monitor misses, in about a quarter of answers. GRAFT, from MATS, FAR AI and Anthropic researchers, keeps the reward and changes what learns: flagged samples update only a sacrificial adapter. Delete it after training, and hacks of every kind nearly vanish. - Gradient Routing Localizes and Removes Unintended Behaviors in RL (arXiv, primary): https://arxiv.org/abs/2610.17062 Voice: synthetic (neural text-to-speech). PHYSICS & ENGINEERING Trapped antimatter, altermagnets Physics & Engineering. Caged antimatter, single-molecule kicks, attosecond X-rays, squeezed ice. Then altermagnets, W states and Carnot's limit. Watch: https://www.youtube.com/watch?v=cet518d5zuc 4 NEWS One antimatter particle, trapped for months Northwestern physicists have trapped a single positron, the electron's antimatter twin, and held it for months, cooled to about zero point three kelvin, in its quantum ground state. It could enable the most precise matter and antimatter test yet. - Scientists trap and tame a single particle of antimatter for months (Northwestern University, primary): https://news.northwestern.edu/stories/2026/10/scientists-trap-and-tame-a-single-particle-of-antimatter-for-months - Myers et al., Physical Review Letters 2026 (Physical Review Letters, primary): https://doi.org/10.1103/gbkc-6nlj Feeling the kick of a single gas molecule At Yale, a laser-levitated glass nanosphere in vacuum felt single gas molecules hit it, one kick at a time, down to ten to the minus twenty three newton seconds. Counting the kicks becomes a pressure gauge. - Optomechanical Detection of Individual Gas Collisions (arXiv, primary): https://arxiv.org/abs/2604.18371 Record-short hard X-ray flashes at SLAC SLAC's X-ray laser, LCLS, has produced the shortest high energy X-ray pulses ever measured at a free electron laser: 100 to 400 attoseconds. Energetic enough to see deep into atoms, fast enough to freeze electron motion. - SLAC's X-ray laser sets a record for shortest high-energy pulses ever generated by an XFEL (SLAC National Accelerator Laboratory, primary): https://www6.slac.stanford.edu/news/2026-10-20-slacs-x-ray-laser-sets-record-shortest-high-energy-pulses-ever-generated-xfel Squeezed past 222 GPa, ice loses symmetry UNLV physicists squeezed water ice between diamonds past 222 gigapascals. Instead of getting more symmetric, the ice switched to a lower symmetry structure, closer to conditions inside Uranus and Neptune. Published in Communications Chemistry. - Water As We've Never Known It: Physicists Test Water Beyond Earth-like Conditions (UNLV, primary): https://www.unlv.edu/news/release/water-weve-never-known-it-physicists-test-water-beyond-earth-conditions - Grande, Salamat et al., Communications Chemistry 2026 (Communications Chemistry, primary): https://doi.org/10.1038/s42004-026-02241-x 3 IDEAS Altermagnetism: a third kind of magnet Altermagnetism, a third kind of magnetism. In an antiferromagnet, neighbouring spins point opposite ways and cancel. Altermagnets cancel too, but the two opposite sublattices are related by a rotation of the crystal. So electrons moving in different directions see opposite spin splitting: zero net magnetization, yet spin polarized currents, like a ferromagnet. Predicted in 2022, seen in manganese telluride in 2024, and awarded the 2026 Europhysics Prize. - Beyond Conventional Ferromagnetism and Antiferromagnetism: A Phase with Nonrelativistic Spin and Crystal Rotation Symmetry (Physical Review X, primary): https://journals.aps.org/prx/abstract/10.1103/PhysRevX.12.031042 - Altermagnetic lifting of Kramers spin degeneracy (nature.com, primary): https://www.nature.com/articles/s41586-023-06907-7 - 2026 Europhysics Prize honors discovery of altermagnetism as a third fundamental class of magnetism (JGU Mainz, primary): https://press.uni-mainz.de/2026-europhysics-prize-honors-discovery-of-a-third-fundamental-class-of-magnetism/ GHZ vs W: two ways to share entanglement GHZ versus W: two ways for three photons to share one quantum state. In GHZ, all three are zero or all three are one; lose one photon and the rest are no longer entangled. In W, a single one is spread over all three; lose one and two stay entangled. In 2025, Kyoto and Hiroshima physicists built the first one-shot measurement for W states, an optical Fourier circuit, right about 87 percent of the time. - Entangled measurement for W states (Science Advances, primary): https://pmc.ncbi.nlm.nih.gov/articles/PMC12428927/ - Measuring the quantum W state (Kyoto University, primary): https://www.kyoto-u.ac.jp/en/research-news/2025-09-16 Carnot's limit: temperature sets the ceiling Carnot's limit. In 1824 the engineer Sadi Carnot asked how efficient a heat engine can possibly be. The answer depends only on two temperatures, not on the fluid or the design: one minus cold over hot, in kelvin. With the cold side at three hundred kelvin, a hot side at four hundred kelvin caps you at 25 percent. At six hundred kelvin, 50 percent. At nine hundred kelvin, about 67 percent. That is why engineers keep pushing for hotter engines. - The Laws of Thermodynamics (The Feynman Lectures on Physics, Vol. I, ch. 44, primary): https://www.feynmanlectures.caltech.edu/I_44.html Voice: synthetic (neural text-to-speech). BIOLOGY & NATURE Lynx vs kites, laughing apes, CAR-NK cells Biology & Nature: a lynx turns on kites, laughing orangutans, sticky plankton, a fossil AI. Then the carbon pump, otters and kelp, CAR-NK cells. Watch: https://www.youtube.com/watch?v=fmdcJBE_fhQ 4 NEWS Drought turns a lynx against kites For three decades, Spain's Doñana Biological Station followed lynx and black kites. After a virus crashed the rabbits both hunt, lynx raided kite nests in drought years, taking up to 46 percent of nearly fledged chicks. The study is in Nature. - Drought and disease trigger novel interactions among top predators (Nature (Estación Biológica de Doñana-CSIC), primary): https://www.nature.com/articles/s41586-026-11194-z Orangutans prefer food found to laughter At Indiana University, six zoo orangutans heard recordings of orangutan laughter while digging for treats in one material, and a control sound for another. Later, given the choice, four of the six preferred the laughter material. The study is in Biology Letters. - Play vocalizations induce affective bias in reward learning and memory in orangutans (Biology Letters (Royal Society), Indiana University, primary): https://doi.org/10.1098/rsbl.2026.0404 The ocean's loner microbe has a sticky side Prochlorococcus, the most abundant photosynthetic organism in the sea, was thought to drift alone. MIT's Chisholm lab shows in Nature Communications that every strain they tested can also form biofilms, and attached cells release almost twice as much organic carbon. - The ocean's most abundant 'loner' microbe has a sticky second life (MIT News (MIT Civil and Environmental Engineering), primary): https://news.mit.edu/2026/oceans-most-abundant-loner-microbe-has-sticky-second-life-1020 - Biofilm formation and dynamics in the marine cyanobacterium Prochlorococcus (Nature Communications, primary): https://doi.org/10.1038/s41467-026-77598-7 A fossil AI questions the Big Five extinctions At the Geological Society of America meeting, Shuzhong Shen's team presented OneStrata, an AI trained on over 2.8 million fossil occurrences. Only two of the Big Five mass extinctions look like sharp global collapses. It's not yet peer reviewed. - OneStrata: A foundation model for deep-time biodiversity and mass extinction reassessment (Geological Society of America (GSA Connects 2026 abstract), primary): https://gsameetings.secure-platform.com/connects26/solicitations/103015/sessiongallery/schedule/items/96256/application/19884 3 IDEAS The biological carbon pump The biological carbon pump. In sunlit water, phytoplankton take up carbon dioxide as they grow, and animals eat them. Most of that carbon is recycled near the surface, but some sinks as marine snow: dead cells, fecal pellets and scraps of messy feeding. Below about five hundred meters, it can stay out of the atmosphere for a century or more. That's why Prochlorococcus riding sinking particles could matter for climate. - Biological carbon pump (Woods Hole Oceanographic Institution, primary): https://www.whoi.edu/ocean-learning-hub/ocean-topics/how-the-ocean-works/cycles/biological-carbon-pump/ Trophic cascades: otters, urchins and kelp Trophic cascades. A predator can shape plants it never eats. In the 1970s, Jim Estes dove around Alaska's Aleutian Islands. Where sea otters lived, kelp forests thrived. Where otters were missing, urchins covered the seafloor and little kelp was left. Otters eat urchins, urchins eat kelp, so losing the top link flips the system. Today's lynx and kites follow the same logic, rippling down the food web. - Emeritus and Distinguished Alumni Profile: Jim Estes Explains How Sea Otters Run the world (U.S. Geological Survey, primary): https://www.usgs.gov/centers/werc/news/emeritus-and-distinguished-alumni-profile-jim-estes-explains-how-sea-otters-run Programming cells from within: CAR-NK in vivo Programming cells from within. CAR cells are immune cells given a new receptor that recognises one target. Today they are usually engineered outside the body. An MIT team instead packed CAR instructions as messenger RNA into lipid nanoparticles aimed at natural killer cells. Inside the cell, ribosomes translate that RNA into the receptor. In mice, up to about 80 percent of NK cells took up the particles. It's preclinical research, not a therapy. - Reprogrammed NK immune cells target cancer and autoimmune disorders (MIT News, primary): https://news.mit.edu/2026/reprogrammed-nk-immune-cells-target-cancer-and-autoimmune-disorders-1013 Voice: synthetic (neural text-to-speech). MIND & SOCIETY The fog of war, a testosterone retest, focal points Mind & Society: the fog of war, a testosterone retest, Italians hearing Spanish, philosophy's new AI hubs. Then focal points, explanatory depth, memory as imagination. Watch: https://www.youtube.com/watch?v=4VYHZJm71pE 4 NEWS Fog of war: why defence looks like attack UCLA cognitive scientist Tao Gao built a silent two-player territory game. Players often read defensive moves as threats, and fights escalated. Just colouring the board into two regions, with no rules attached, meant fewer battles and longer peace. - Breaking through the 'fog of war' to find peace (UCLA Newsroom (University of California, Los Angeles), primary): https://newsroom.ucla.edu/releases/breaking-through-the-fog-of-war-to-find-peace Testosterone and gut answers: a big retest says no A 2017 study of 243 men suggested one dose of testosterone pushed people toward gut answers on the cognitive reflection test. Rival teams retested it together: 1,000 men, double-blind, in Psychological Science. No effect on the score. - Does Testosterone Affect Cognitive Reflection? Evidence From a Double-Blind, Randomized Controlled Study of 1,000 Participants (Psychological Science (open-access copy, UCL Discovery), primary): https://discovery.ucl.ac.uk/id/eprint/10227899/ Italians hearing Spanish overrate how much they get University of Chicago researchers had Italians listen to Spanish and pick, out of four options, what the speaker meant. They were right 40 percent of the time, but believed they were right 80 percent. A close language inflates confidence. - Can Italians Understand Spanish? Language Closeness Exacerbates the Illusion of Understanding (Psychological Science (full text, UChicago Multilingualism and Decision-Making Lab), primary): https://mdl.uchicago.edu/research/ren-2026-can-italians-understand-spanish-language-closeness-exacerbates-the-illusion-of-understanding/ Philosophers open two new AI research hubs LSE has launched the Global Priorities Research Group, funded by Good Ventures, to study transformative AI and how to weigh the interests of humans, animals and other minds. In Toronto, a new initiative studies artificial consciousness, without assuming machines have it. - The Global Priorities Research Group: About us (London School of Economics and Political Science (LSE), primary): https://www.lse.ac.uk/global-priorities/about-us - Artificial Consciousness Initiative (University of Toronto, primary): https://artificialconsciousness.ca/ 3 IDEAS Focal points: agreeing without talking Focal points. In 1960 the economist Thomas Schelling asked people where and when they would meet a friend in New York, with no way to agree in advance. Most chose noon at Grand Central Station. When we can't communicate, we coordinate on what stands out. The philosopher David Lewis built a theory of conventions on this salience. It's the coloured border in the UCLA game: a line that means nothing, until both sides know the other sees it too. - Convention (Stanford Encyclopedia of Philosophy): https://plato.stanford.edu/entries/convention/ The illusion of explanatory depth The illusion of explanatory depth. At Yale, Leonid Rozenblit and Frank Keil asked people how well they understood everyday things, like a zipper or a flush toilet. Then they had to write a step by step explanation, and their ratings dropped. The gap was large for devices and natural phenomena, smaller for facts like capital cities, and absent for procedures. Try it: explain how a zipper locks. Familiarity feels like understanding, just like Spanish for an Italian. - The misunderstood limits of folk science: an illusion of explanatory depth (Cognitive Science (via PubMed Central), primary): https://pmc.ncbi.nlm.nih.gov/articles/PMC3062901/ Is remembering a kind of imagining? Is remembering a kind of imagining? The causal theory says a memory counts only if it is causally linked to what you lived. Simulation theories disagree: one system imagines the future and rebuilds the past, and the causal link is at best incidental. Duke philosopher Felipe De Brigard, a 2026 MacArthur Fellow, found that imagining how a past event could have gone differently uses largely the same brain regions as remembering it. Memory errors become the price of a flexible mind. - Felipe De Brigard, Philosopher and Cognitive Scientist, 2026 Fellow (MacArthur Foundation, primary): https://www.macfound.org/fellows/class-of-2026/felipe-de-brigard - Memory (Stanford Encyclopedia of Philosophy): https://plato.stanford.edu/entries/memory/ Voice: synthetic (neural text-to-speech). Issue: https://newsideas.app/2026-10-22/ Contact: contact@newsideas.app --- News & Ideas 21 October 2026 2 min read A superconductor clue, a green Greenland, seahorse dads, first impressions. Then text drawn coarse to fine, the genetic code and the forgetting curve. BIOLOGY & NATURE All tracks (before 22 Oct 2026) A superconductor clue, ice-free Greenland, next-concept AI A superconductor clue, a green Greenland, seahorse dads, first impressions. Then text drawn coarse to fine, the genetic code and the forgetting curve. Watch: https://www.youtube.com/watch?v=PFAWHxCLlDY 4 NEWS A smoking gun for high-temperature superconductivity Physicists led by the University of Bristol report in Nature Physics that, in an electron-doped copper oxide, superconductivity comes from conduction electrons interacting with fluctuating spins. They call it half the riddle: the record-breaking hole-doped cuprates need a second, still missing ingredient. - Evidence for spin-fluctuation-mediated superconductivity in electron-doped cuprates (Nature Physics, primary): https://doi.org/10.1038/s41567-026-03454-6 - 'Smoking gun' may help solve mystery of high-temperature superconductivity (Phys.org / University of Bristol): https://phys.org/news/2026-10-gun-mystery-high-temperature-superconductivity.html Cold War samples reveal a warmer, ice-free Greenland Sediment pulled from under the Camp Century base in the 1960s shows the site was ice-free about 416,000 years ago, with summers 4.7 degrees Celsius warmer than today. Similar warmth after the last ice age didn't melt the ice there, so how long warming lasts may matter as much as how hot it gets. - Forgotten samples from Greenland's Cold War base reveal warmer summers during ancient ice-free period (Phys.org / University at Buffalo): https://phys.org/news/2026-10-forgotten-samples-greenland-cold-war.html How seahorse fathers clear their embryos' waste Male seahorses carry embryos in a brood pouch, and ammonia waste must get out. A team at Sophia University found ammonia-transporting Rh proteins in the pouch lining and in blood cells, suggesting the waste travels through the father's bloodstream to his gills. - Male seahorse pouch proteins may help remove toxic waste from developing embryos (Phys.org / Sophia University): https://phys.org/news/2026-10-male-seahorse-pouch-proteins-toxic.html First impressions of faces, without trait words Instead of rating faces as trustworthy or competent, more than 3,000 people at Caltech simply arranged faces by how similar their personalities looked. Five dimensions emerged: approachability, warmth, competence, morality and youthfulness. It maps our biases, not anyone's real personality. - Judging faces without personality labels uncovers key patterns in first impressions (Phys.org / Caltech): https://phys.org/news/2026-10-personality-uncovers-key-patterns.html 3 IDEAS Next-concept modeling: text drawn coarse to fine Next-concept modeling. Chatbots usually write one token at a time. A Stanford paper takes the image trick of next-scale prediction to text. An encoder compresses a passage into levels of codes, from a short coarse sketch to fine detail. A transformer then predicts each level in parallel, about twelve steps instead of one per token. At around a hundred million parameters it matches or beats diffusion language models while running ten to thirty one times faster. - Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text Generation (arXiv, primary): https://arxiv.org/abs/2610.14162 - HazyResearch/ncm: Next-Concept Modeling code (GitHub (HazyResearch, Stanford), primary): https://github.com/HazyResearch/ncm - Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (arXiv, primary): https://arxiv.org/abs/2404.02905 The genetic code: life reads in triplets The genetic code. A gene is a message written with four letters. It's copied into messenger RNA. Then a ribosome reads it three bases at a time. Each triplet, or codon, calls for one amino acid. Four letters in groups of three give sixty four codons for just twenty amino acids, so the code is redundant, and three codons mean stop. The chain then folds into a protein, built from those left-handed amino acids we met yesterday. - Genetic Code (NHGRI Talking Glossary, primary): https://www.genome.gov/genetics-glossary/Genetic-Code - From DNA to protein - 3D (yourgenome (Wellcome Sanger Institute) (YouTube)): https://www.youtube.com/watch?v=gG7uCskUOrA The forgetting curve, and why reviews are spaced The forgetting curve. In the eighteen eighties Hermann Ebbinghaus memorised lists of nonsense syllables, then measured how much time he saved when relearning them later. Forgetting is fastest right after learning, then slows down. In 2015, Murre and Dros replicated it with results close to the original. That curve is why this show reviews ideas after one, three, seven, twenty one and sixty days. - Replication and Analysis of Ebbinghaus' Forgetting Curve (PLOS ONE, primary): https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0120644 - The forgetting curve (McGill Teaching and Academic Programs (YouTube)): https://www.youtube.com/watch?v=UpAq9EZpL5A Voice: synthetic (neural text-to-speech). Issue: https://newsideas.app/2026-10-21/ Contact: contact@newsideas.app --- News & Ideas 20 October 2026 2 min read AI and the Riemann zeros, a leaky core, a cosmic hamburger, faster trees. Then mirror molecules, momentum and Hume. PHYSICS & ENGINEERING All tracks (before 22 Oct 2026) AI's quasi-Riemann proof, Earth's leaky core, chirality and Hume AI and the Riemann zeros, a leaky core, a cosmic hamburger, faster trees. Then mirror molecules, momentum and Hume. Watch: https://www.youtube.com/watch?v=lWiSqj2VoUs 4 NEWS An AI proof of the quasi-Riemann hypothesis OpenAI's model proved the quasi-Riemann hypothesis: all the nontrivial zeros of the zeta function lie in a narrower strip, between one twelfth and eleven twelfths. Experts quoted by Quanta largely agree it's correct, and Stanford's Jared Lichtman calls it the most significant AI-for-math achievement so far. - AI Proved the Quasi-Riemann Hypothesis. What Happens Now? (Quanta Magazine): https://www.quantamagazine.org/ai-proved-the-quasi-riemann-hypothesis-what-happens-now-20261019/ Earth's core may be leaking hydrogen Some deep volcanic lavas, from places like Iceland, carry hydrogen unusually poor in deuterium. A Science Advances study simulated the early magma ocean and found it can't explain that signature. The likely culprit: hydrogen leaking out of Earth's core for billions of years. - Zhang et al., Science Advances 2026 (Science Advances, primary): https://doi.org/10.1126/sciadv.aei4756 - Earth's core may have been leaking hydrogen for billions of years (Phys.org): https://phys.org/news/2026-10-earth-core-leaking-hydrogen-billions.html A cosmic hamburger where giant planets form With the ALMA telescope, astronomers spotted the earliest phases of giant planet formation in Gomez's Hamburger, a huge disk of gas and dust seen almost edge-on. It's a rare benchmark for testing how planets form far from their star. - Cosmic 'hamburger' offers rare edge-on view of where giant planets may take shape (Phys.org / NRAO): https://phys.org/news/2026-10-cosmic-hamburger-rare-edge-view.html CRISPR rewires poplar chromosomes in one generation And in biology, University of Georgia researchers used CRISPR to trigger large chromosome rearrangements in poplar trees in a single generation, a shortcut that could cut decades from tree breeding. - CRISPR shortcut could cut decades from chromosome design in poplars (UGA Research News, primary): https://research.uga.edu/news/crispr-shortcut-could-cut-decades-from-chromosome-design-in-poplars-uga-study-shows/ 3 IDEAS Chirality: molecules with a left and a right hand Chirality. Some molecules come in two mirror-image forms, like your left and right hands: same parts, same distances, yet you can't superimpose them. Usually the cause is a carbon atom bonded to four different groups. And the two forms can behave differently. R carvone smells of spearmint. Its mirror twin, S carvone, smells of caraway. Life picks one side: proteins are built from L amino acids, DNA and RNA from D sugars. Why just one hand? That's the puzzle behind this year's chemistry Nobel. - What is chirality and how did it get in my molecules? (TED-Ed (YouTube)): https://www.youtube.com/watch?v=71GjsRnsoL8 - Carvone (Wikipedia): https://en.wikipedia.org/wiki/Carvone - Homochirality (Wikipedia): https://en.wikipedia.org/wiki/Homochirality Conservation of momentum: how rockets push on nothing Conservation of momentum. Momentum is mass times velocity. In an isolated system the total never changes, because every internal push comes as a third-law pair: equal, opposite, at the same instant. An astronaut floating in space throws a wrench forward, and drifts backward. A rifle kicks back when the bullet leaves. And a rocket needs no air to push against: it throws exhaust backward at high speed, and gains the same momentum forward. It's yesterday's third law, added up over time. - Conservation of Momentum (NASA Glenn Research Center, primary): https://www1.grc.nasa.gov/beginners-guide-to-aeronautics/conservation-of-momentum/ - Vol. I, ch. 10: Conservation of Momentum (The Feynman Lectures on Physics (Caltech), primary): https://www.feynmanlectures.caltech.edu/I_10.html Hume's problem of induction Hume's problem of induction. We expect the future to resemble the past. But why? Not by logic: Hume notes that the sun not rising tomorrow implies no contradiction. And not by experience, because arguing from past successes assumes the very thing we want to prove. Bertrand Russell's chicken makes it vivid: the man who fed it every day at last wrings its neck. Hume adds that we never observe causation itself, only one event constantly followed by another. Correlation, again. - The Problems of Philosophy, ch. VI "On Induction" (Project Gutenberg, primary): https://www.gutenberg.org/files/5827/5827-h/5827-h.htm - The Problem of Induction (Stanford Encyclopedia of Philosophy): https://plato.stanford.edu/entries/induction-problem/ Voice: synthetic (neural text-to-speech). Issue: https://newsideas.app/2026-10-20/ Contact: contact@newsideas.app --- News & Ideas 19 October 2026 2 min read Quantum errors, a chemistry Nobel, fungi that trade, AI benchmarks. Then Newton, superposition and retrieval practice. PHYSICS & ENGINEERING All tracks (before 22 Oct 2026) Google Willow, Chemistry Nobel, Newton, retrieval practice Quantum errors, a chemistry Nobel, fungi that trade, AI benchmarks. Then Newton, superposition and retrieval practice. Watch: https://www.youtube.com/watch?v=jR0x6fxDZ0o 4 NEWS Google cultivates magic states on Willow Google just demonstrated magic state cultivation on its Willow chip, cutting errors by about 40 times. It's published in Science. Why it matters: magic states are the bottleneck for universal, error-corrected quantum computing. - Bringing the magic: Google breaks open a bottleneck in quantum computing (Google Quantum AI, primary): https://blog.google/innovation-and-ai/models-and-research/quantum-computing/magic-state-cultivation-quantum-computing/ Nobel 2026: how a reaction picks one mirror image The 2026 Nobel Prize in Chemistry goes to Henri Kagan and Kenso Soai. They explained how a reaction can choose just one of a molecule's two mirror-image forms. That matters for the origin of life, since biological molecules share one handedness, and for making drugs. - Nobel Prize in Chemistry 2026 (The Nobel Prize, primary): https://www.nobelprize.org/prizes/chemistry/2026/press-release/ Fungi trade with plants, no brain needed On a Quanta podcast, biologist Toby Kiers explains how soil fungi trade nutrients with plants, and pick the best partners, all without a brain. - How do fungi keep the planet healthy? (Quanta Magazine): https://www.quantamagazine.org/how-do-fungi-keep-the-planet-healthy-20261015/ Making AI benchmarks a science And in AI, Moritz Hardt calls benchmarks the error machine learning made, and explains how to make AI benchmarking a real science, not a dark art. - He Wants AI Benchmarking To Be a Science, Not a Dark Art (Quanta Magazine): https://www.quantamagazine.org/he-wants-ai-benchmarking-to-be-a-science-not-a-dark-art-20261016/ 3 IDEAS Newton's third law: why the horse still moves Newton's third law. When A pushes on B, B pushes back on A with an equal and opposite force, at the same instant, along the same line. The catch: the two forces act on different bodies, so they never cancel each other. Take the horse and cart. The cart speeds up because the horse's pull beats the friction on the cart. The horse moves forward because its hooves push the ground backward, and the ground pushes the horse forward. And, as Feynman shows, the third law plus the second gives you conservation of momentum. - Vol. I, ch. 10: Conservation of Momentum (The Feynman Lectures on Physics (Caltech), primary): https://www.feynmanlectures.caltech.edu/I_10.html Superposition and sparse autoencoders Superposition. A language model must represent far more concepts than it has neurons, so it packs them as nearly orthogonal directions that overlap. That's why single neurons are polysemantic. A sparse autoencoder unpacks them into a much larger dictionary, where each unit tends to mean one thing. In Claude 3 Sonnet, Anthropic found a Golden Gate Bridge feature: clamp it at 10 times its maximum, and the model drifts to the bridge. The catch: in the 34 million feature version, about 65 percent of features were dead. And a new paper shows that features picked by correlation often fail to predict what an intervention does. Correlation isn't causation. - Scaling Monosemanticity (Anthropic, primary): https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html - Toy Models of Superposition (Anthropic, primary): https://transformer-circuits.pub/2022/toy_model/index.html - What Do Sparse Autoencoder Feature Discovery Scores Identify? From Activations to Interventions (arXiv, primary): https://arxiv.org/abs/2610.15850 Retrieval practice beats rereading Retrieval practice. Pulling something out of memory strengthens it more than reading it again. Rereading feels like knowing. Retrieving forces you to rebuild. In Roediger and Karpicke's 2006 study, people read a text, then either reread it or wrote down everything they remembered. After five minutes, rereading won: 81 percent versus 75. After two days, testing won: 68 versus 54. After one week, testing again: 56 versus 42. What works in the short term is not what works in the long term. - Test-Enhanced Learning (Psychological Science (author copy, Purdue Learning & Memory Lab), primary): https://learninglab.psych.purdue.edu/downloads/2006/2006_Roediger_Karpicke_PsychSci.pdf Voice: synthetic (neural text-to-speech). Issue: https://newsideas.app/2026-10-19/ Contact: contact@newsideas.app