Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Stanford University

AI overview

Sign in and the AI will write an overview from our coverage.

Headlines · 56

  1. Elite students pursue AI startups over Big Tech and Wall Street internships

    A report from Xin Zhiyuan, citing The Wall Street Journal, says students from Ivy League schools, Stanford and MIT turned down internships at major technology companies and Wall Street investment banks for AI startup work at San Francisco Hacker Houses. Princeton student Charles Muehlberger reportedly founded an AI startup within four weeks, took a year off and pitched potential customers in Barcelona on deploying open-source AI models locally. 36Kr republished the article with authorization; it is not independent corroboration, and the supplied reports do not quantify the scale of the student migration.

    新智元 · 🔥 0
  2. SWE-sweep Tests AI Agents Finding Hidden Bugs Without Issue Hints

    Researchers from Meta Super Intelligence Lab, Harvard University, Stanford and other institutions introduced SWE-sweep, a software-engineering benchmark built from 100 real open-source repositories containing 4,100 historical vulnerabilities. Unlike conventional issue-driven benchmarks, it gives coding agents only complete repositories and requires them to discover and repair hidden defects without descriptions, error messages or vulnerability locations. DoNews reported that successful repairs must turn a hidden test from failing to passing while preserving the repository’s full original test suite, with the initial leaderboard run uniformly on mini-SWE-agent.

    DoNews · 🔥 0
  3. Altman Says Mathematicians Are Moving From Fear Toward Acceptance of AI

    According to media reports, OpenAI CEO Sam Altman told an audience of several hundred students at Stanford University that mathematicians’ attitudes toward AI have moved from disbelief and anger toward acceptance that AI can no longer be avoided in research and proof work. Altman described the progression as resembling stages of grief. His optimistic framing contrasts with mathematicians’ continuing concerns that AI systems may not show their derivations and leave researchers with difficult verification work. OpenAI’s announced solution to the Navier–Stokes equations and its claims about solving more than 700 other problems were not independently verified in the supplied material.

    36氪 人工智能 · 🔥 0
  4. HybridAgent paper mixes latent-space thinking with text tool calls

    A commentary published by Huxiu AI reports that a paper submitted to arXiv on October 13 (2610.18861v1), with authors drawn from Stanford, UIUC, Together AI and Meta, proposes HybridAgent, which lets a model decide within a single generation whether to reason in latent space without emitting text or to act in text for tool calls and outward delivery. The design rests on three parts: HybridLink, a residual adapter of roughly 10M parameters on a 4B model that switches between text and latent modes; a continuous working memory that carries mixed context across turns; and a pseudo-text rendering of latent reasoning that monitors can read. The authors self-report 2.7–3.7x cost savings, 1.8–2.6x speedup and a 14B model beating GPT-5.6-Luna by about 14.5 points on the single-point GAIA benchmark, and the article states none of these figures has been independently replicated.

    虎嗅 AI · 🔥 0
  5. Stanford/Cornell paper: AI job applications may push hiring toward experience

    A Chinese-language summary from Huxiu AI describes a preliminary Stanford and Cornell paper, “Can Labor Markets Function in the Age of AI? The Evaluation Bottleneck in Hiring,” by Itai Ashlagi, Ramesh Johari, Jon Kleinberg and Anushka Murthy. The model assumes AI lowers the cost of searching for jobs and producing applications while making application materials noisier, so firms rationally rely more on observable work experience; candidates with no experience but a genuine match are especially affected. The authors also analyze multi-stage hiring, where a cheaper intermediate assessment such as a work sample can provide new matching information before a costly full evaluation, potentially reopening screening that a single-stage process would close.

    虎嗅 AI · 🔥 0
  6. Proactive AI for Superintelligent Agents Introduces a 6A Framework

    A 203-page review titled “Proactive AI for Superintelligent Agents,” jointly developed by Google, UIUC, Stanford, UC Berkeley, and CMU, examines how AI agents can initiate useful work without waiting for a new instruction. It proposes a 6A framework—Awareness, Anticipation, Agenda, Arbitration, Action, and Adaptation—and reviews mechanisms, evaluation methods, permissions, and applications across 13 domains. The review also distinguishes intelligence, alignment, proactiveness, and autonomy, while noting that long-term maintenance and higher-level self-directed goal revision remain underdeveloped.

    新智元 · 🔥 0
  7. Cammarata Says LLMs May Be Easier to Interpret, Drawing Pushback

    Goodfire researcher Nick Cammarata argues that large language models may be more understandable than previously believed, drawing on neuron-level experiments and a reinterpretation of Inception v1 as evidence that neural features can follow identifiable patterns. He reports that roughly 99% to 100% of final-layer neurons in Inception v1 could be relatively clearly explained, while preliminary Qwen3-32B tests found a main function for about 90% of neurons. MIT Technology Review reports that researchers including Chris Potts, Jacob Hilton, and Neel Nanda dispute the broader claim, noting that the work is an unreviewed working hypothesis and does not explain how neurons combine into circuits.

    MIT科技评论中文 · 🔥 0
  8. Anthropic scientist's talk asks whether a PhD is worth it if AGI arrives in three years

    MIT Technology Review Chinese reports that Anthropic scientist Jascha Sohl-Dickstein used a June talk at a Stanford physics-and-AI conference to press young researchers on how to plan academic careers if AGI could arrive within a few years. He described a graduate student who said he believed AGI was three years away yet still intended to complete a PhD, a postdoc and a faculty search, and argued that research topics should be chosen with fast-improving foundation models in mind. He urged researchers to adopt AI tools early for execution, but said they should keep writing drafts and judging results themselves, warning that outsourcing writing means outsourcing thinking.

    MIT科技评论中文 · 🔥 0
  9. WorldMemArena benchmark tests multimodal agent memory through action–world loops

    Researchers from UC Santa Barbara, Stanford, ETH Zurich, Johns Hopkins, Carnegie Mellon and J.P. Morgan Chase have jointly proposed WorldMemArena, a multimodal agent memory benchmark, according to a 机器之心 report. The paper, whose co-first authors Chengzhi Liu and Yuzhe Yang are both at UCSB, was accepted at NeurIPS 2026. WorldMemArena evaluates memory inside an observe–act–environment-feedback loop and splits long-term memory into writing, maintenance, retrieval and use stages, with more than 400 multi-session tasks, 24,258 QA pairs and 15,595 images or screenshots. In the paper's results, M2A reached 86.83% memory recall but only 50.14% final QA accuracy, while MemGPT scored 85.20% and 57.81%, and systems tended to append new memories rather than revise outdated ones and did not reliably use visual information in cross-modal reasoning.

    机器之心 · 🔥 0
  10. PhAI Labs releases DiscoveryRSI research agent and 27B model

    PhAI Labs, working with several universities and research institutions, proposed DiscoveryRSI and trained DiscoveryRSI-27B using a hypothesis-credit attribution mechanism, releasing the paper, code and models. The system stores evolving hypotheses and test plans in a research hypothesis tree and is paired with DFM-Bench, which evaluates hypothesis revisions, research paths and whole research episodes. The team reports that DiscoveryRSI's group scored 6.3% in a literature cross-check, the highest of five model groups, while the Claude Opus 5 group scored 2.3% and ranked last; it cautions that group sample sizes differ and the comparison is descriptive, so it does not prove the hypotheses were adopted or validated by later research.

    机器之心 · 🔥 0
  11. Neuroscientist Huberman: AI's real arms race is non-invasive brain read/write

    Stanford neuroscientist Andrew Huberman argued on a podcast that the real AI race is not model capability but non-invasive ways, such as ultrasound, to stimulate or suppress specific brain regions at precise moments and tune intensity to a user's live brain state. He expects commercialization to start with sleep monitoring and regulation, then extend to focus, faster learning and stress control, and eventually motivation. He also warned that strict safety mechanisms will be essential once devices can influence the whole brain; the claims are personal predictions without specific products, data or timelines.

    科技新报 · 🔥 0
  12. OpenAI math breakthroughs stir cryptography fears

    OpenAI and other labs' AI models have solved decades-old math problems, prompting cryptography experts to debate whether powerful AI could bypass private keys and break encryption. Whitfield Diffie, a co-inventor of public-key cryptography, said he expects major changes in mathematics and cryptanalysis but cannot gauge the timeline — a year may be too soon and five years may not be enough. Martin Hellman is more optimistic, while Johns Hopkins professor Matthew Green said public-key cryptography may be obsolete and predicted frontier AI could fundamentally break elliptic-curve cryptography within two years; Cloudflare's 2025 report says X25519 is used by most internet traffic. A Princeton computer science professor argued well-studied mainstream algorithms will likely stay safe, but less-studied ones, including many post-quantum candidates, could be broken or weakened; in July Anthropic said its AI nearly broke a NIST post-quantum candidate algorithm and had briefed the US government and industry partners beforehand. OpenAI did not respond on whether its internal models were used against cryptography.

    cnBeta · 🔥 0
  13. Stanford and Together AI propose parallel coarse-to-fine text generation

    Stanford University and Together AI researchers proposed Next-Concept Modeling (NCM), an approach that differs from autoregressive next-token prediction and diffusion-based generation. NCM first generates a coarse concept representation, then predicts finer levels of discrete text codes in parallel before decoding the result. In the paper’s experiments, NCM generated text 10 to 31 times faster than diffusion language-model baselines, with comparable or better quality on the tested tasks. It also outperformed autoregressive and diffusion baselines on difficult Sudoku puzzles with 30 clues, with an 86.1% complete-solution rate compared with 81.3% for FMLM+. The evaluated generator was about 100 million parameters, with experiments extending to 725 million parameters. Performance at larger scales, on exact-text tasks and on long-form generation still requires validation.

    36氪 人工智能 · 🔥 0
  14. Stanford economist: remote-work digital chatter became a trove of LLM training data

    Citing a Business Insider interview, TechNews reports that Stanford University economist Nick Bloom argues the surge in digital communication after people left offices created large amounts of material for training large language models, turning casual kitchen-table chat into Slack threads that could be fed to LLMs. The article says that as most people shift to hybrid work, impromptu coffee chats and desk-side discussions leave no record, making it a problem for companies to obtain more training data; it also notes that recording office conversations would raise legal and regulatory issues. Bloom separately estimates that remote work raised fertility rates in the United States and Nordic countries by 5%, and lists six ways the WFH era fundamentally changed society.

    科技新报 · 🔥 0
  15. Cryptographers weigh AI's threat to internet encryption

    MIT Technology Review China reports that a16z crypto recently released a discussion featuring Stanford cryptographer Dan Boneh, Georgetown computer science professor Justin Thaler and others on the potential threat that rapid gains in AI math capability pose to modern cryptography. The report notes that OpenAI previously released hundreds of mathematics research manuscripts generated by internal models, but the list did not include cryptography results; Boneh said the public results did not show an actual break of mainstream cryptographic schemes and argued frontier labs should disclose whether their strongest models tested deployed systems and what the results were. The discussion also covered Ethereum Foundation researcher Justin Drake's "bunker mode" proposal, parallel deployment of post-quantum standards, and "crypto agility," while industry remains divided on when attacks might occur and how migration should proceed.

    MIT科技评论中文 · 🔥 0
  16. Report: US developer postings up 15% since AI coding era, junior roles down 20%

    According to a CNBC report, Indeed Hiring Lab data shows U.S. software development job postings rose nearly 15% since Anthropic released Claude Code in February 2025, while overall hiring fell 7% over the same period. Growth was concentrated in senior and AI-related roles: senior positions accounted for 71% of software development hiring growth from May 2025 to May 2026, and roles with AI in the title made up 37%. Stanford's 2026 AI Index found employment for 22- to 25-year-olds in AI-related fields fell 16% since generative AI emerged, with early-career software developers down about 20%, while the U.S. Bureau of Labor Statistics projects software developer employment to grow 10% from 2025 to 2035, from roughly 1.72 million to 1.89 million.

    IT之家 AI · 🔥 0
  17. AI Replaces Interns: Where Will Future Experts Come From?

    This 0xAI "First Principles" commentary argues that as companies hand junior work — first-draft reports, client emails, data cleanup — to AI tools, they save on hiring and training but may sever the "learning by doing" path that turns newcomers into seasoned experts. It cites a Reuters program on AI and jobs and a Stanford University and ADP analysis of employment data on millions of US workers, which found that 22-to-25-year-olds in AI-exposed occupations lag their peers elsewhere by roughly 19 percentage points, a gap driven mainly by reduced hiring rather than mass layoffs of incumbents. The author also cites Brynjolfsson and colleagues' study of 5,179 customer-service agents, where an AI assistant raised overall productivity about 14% and gains for novices and lower-skilled workers reached 34%. The piece frames "hire ready-made talent instead of training" as a collective-action problem and suggests firms track how long a newcomer needs to handle complex tasks independently.

    虎嗅 AI · 🔥 0
  18. ByteDance Seed finds periodic blind spots in DeepSeek-V4 long-context retrieval

    A ByteDance Seed team with researchers from Princeton, Stanford and UC Berkeley systematically tested DeepSeek-V4 and V4.1 and describes a "phase sensitivity" effect in the paper "Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression": because CSA compresses the KV cache with overlapping windows of length 8 and stride 4, shifting the same information by a few tokens can sharply change whether it is retrieved. In a 128K needle-in-a-haystack test, the base V4-Flash model scored 69.92% in the best position group versus 29.69% in the worst, a 40.23-point gap; the post-trained V4-Flash-0731 narrowed this to 19.14 points, while V4-Pro showed 34.77 and 14.84 points for base and post-trained versions. The article says DeepSeek had already cut the compression stride from 4 to 2 in V4.1-Flash and added an Engram module, shrinking the gap to 6.1 points without removing the periodicity.

    36氪 人工智能 · 🔥 0
  19. Stanford-led team posts HealthFlux, a pan-modal health world model

    Stanford postdoctoral researcher Ying Kejun, with scholars from the University of Washington, Harvard Medical School and MIT, posted HealthFlux on the medRxiv preprint server, describing it as the first pan-modal human health world model, trained on the UK Biobank's long-term cohort of roughly 500,000 people. The model maintains a health state described by 384 values that evolves on its own when no new data arrive and is corrected immediately when new measurements come in, advanced by neural ODEs, with 11 data types — records, blood tests, about 3,000 plasma proteins, some 250 metabolites and abdominal and brain MRI — feeding the same state; a separate reader extracts disease risk afterwards. The team reports a five-year average AUROC of 0.832 across 1,010 diseases and death in about 60,000 held-out participants, versus 0.805 for the prior multimodal model RisQ, and says the score held at 0.837 when data was cut off for 20 years while Delphi-2M fell to 0.576. The authors note the cohort is largely white and lacks psychological scales, and that the health state index has no warning threshold, so causal evidence for interventions and prospective validation are still missing.

    MIT科技评论中文 · 🔥 0
  20. IROS 2026: Embodied AI turns toward engineering

    Leiphone AI reviewed the main-conference papers and workshop list already announced for IROS 2026 (Sept 27–Oct 1, Pittsburgh) and argues the field's technical路线 is converging on four themes: foundation models internalizing physics, hierarchical "big brain / little brain" architectures becoming standard, touch shifting from nice-to-have to core for dexterous manipulation, and a sharp fall in data-collection costs. The piece cites accepted work including ULTRA, HALO, LITHE, VOFA, Blind Dexterity, KineFuse and ViHaTeleop, saying discussion has moved from grand "general-purpose robot" narratives to concrete engineering issues such as physical correctness, real-time control and cost reduction, with Chinese teams increasingly visible.

    雷峰网 AI · 🔥 0
  21. MIT Technology Review roundtable: the researcher using AI to design viruses

    MIT Technology Review has published a subscriber-only Roundtable conversation in which senior AI reporter James O'Donnell interviews Samuel King, a bioengineering PhD candidate at Stanford University and the Arc Institute. According to the outlet, King used a generative AI model in 2025 to propose genetic blueprints for microscopic viruses; the report notes this is not yet an example of AI-generated life. The session was recorded on Friday, October 16, 2026, and is available to watch or listen, and it also covers King's selection as one of the outlet's Innovators Under 35.

    MIT Technology Review AI · 🔥 0
  22. Stanford's Fei-Fei Li team open-sources OpenWAM world-action model framework

    A Stanford team including Fei-Fei Li, Jiajun Wu and Ehsan Adeli published OpenWAM, an open framework for composable world-action models that aims to disentangle the many ways video prediction and action generation can be coupled in this emerging class of robot policies. OpenWAM starts from Alibaba's open Wan2.2-5B video model, continues pretraining on roughly 3.34 million robot-and-human interaction videos (nominally about 14,600 hours) for 14 days on 32 B200 GPUs, then joins a 5B video expert with a 2B action expert via a shared Mixture-of-Transformers architecture with four configurable interaction orders (VTA, ATV, Joint, Decoupled). The authors report 98.6% average success for VTA across the four LIBERO subsets and 92.1% / 91.9% average success for VTA and Joint on three real dual-arm Franka FR3 tasks, plus a LIBERO-Long-CF counterfactual dataset; training, evaluation and composition code and pretrained video weights are open-sourced.

    36氪 人工智能 · 🔥 0
  23. TPCA Show 2026 spotlights energy-efficient AI, advanced packaging

    Taiwan's Printed Circuit Association (TPCA) says its 27th TPCA Show, running Oct 20–22 at Nangang Exhibition Hall 1, will focus on "Energy-Efficient AI: From Computing to Systems and Applications," with more than 680 international brands, 1,840+ booths and a projected 70,000 professional buyers. The co-located 21st IMPACT-IAAC conference opens Oct 19 at the Grand Hilai Taipei and is expected to feature nearly 350 technical presentations and more than 1,500 industry and academic attendees, expanding for the first time to two Nangang halls. Sessions centre on AI computing and system integration, next-generation packaging and substrates (glass substrates, panel-level packaging, hybrid bonding, chiplets), and thermal management, reliability and smart manufacturing.

    科技新报 · 🔥 0
  24. MIT's Delimitrou applies ML to curb data centers' energy drain

    Christina Delimitrou, a newly tenured associate professor in MIT's Department of Electrical Engineering and Computer Science and a member of CSAIL, and her group apply machine learning to resource management in large-scale data centers, redesigning cloud systems and server architectures so operators can squeeze more computing power out of existing hardware. She says many large computing systems ran at only about 15 percent utilization, with considerable bloat especially on the software side, and that removing that bloat without hurting performance would reduce the need to build new data centers. The group has also built Seer, which uses deep learning to anticipate and prevent web application failures, and Ditto, which mimics a proprietary application's structure and performance so academic teams can study commercial systems they cannot access.

    MIT News · AI · 🔥 0
  25. 2026 Nobel Prize honors optogenetics pioneers

    QbitAI reports that Karl Deisseroth, Peter Hegemann, and Georg Nagel received the 2026 Nobel Prize in Physiology or Medicine for discoveries involving light-gated ion channels and optogenetics. Their work connected light-sensitive proteins from green algae with precise, millisecond-scale control of selected neurons, helping neuroscience move from observing correlations to testing causal links between neural activity and behavior.

    量子位(原生 RSS) · 🔥 0
  26. Apple paper studies how users negotiate ontological boundaries in personal sensing systems

    Apple Machine Learning Research and Stanford University authors published "Negotiating Ontological Boundaries in User-Authored Personal Sensing Systems." They built two open-ended Wizard-of-Oz probes that let participants experience training a personalized machine learning system on phenomena they define themselves, then ran a week-long exploratory study. The authors identify four sites of ontological boundary negotiation — the boundaries of a phenomenon, the subject as part of relations, what counts as signal versus noise, and the objectivity of data — and offer design starting points for supporting such negotiation.

    Apple Machine Learning Research · 🔥 0

← Previous

Experience and discussion from the community

Share my Stanford University experienceAsk about Stanford University

Nobody has shared their experience with Stanford University yet.