Curated reading atlas
Essays, interviews, talks, and standing references that connect artificial intelligence, science, economic development, and the cultural habits that quietly shape big systems. Every entry keeps its source trail, a short editorial summary, and a reading lens.
How machine intelligence actually works, where it falls short of animal cognition, and the long arc from Turing to neural machine translation.
The Wall Street Journal · World models
Yann LeCun argues that current AI still lacks the grounded world models, planning, and common-sense understanding that even animals use naturally.
Reading lensThe useful provocation is not that AI is weak, but that intelligence may require richer internal models than next-token prediction alone.
Synced Review · AutoML
An interview with Quoc Le about Google Brain, AutoML, and the long arc from deep learning research to systems that automate parts of model design.
Reading lensIt captures a moment when machine learning began turning its own tooling into a research object.
IEEE Spectrum · Fallible AI
IEEE Spectrum revisits Turing's interest in machines that can learn, adapt, and make mistakes rather than simply execute perfect fixed rules.
Reading lensFallibility can be a sign of learning capacity, not only a defect to be eliminated.
VentureBeat / Bloomberg · Embodied AI
Coverage of Facebook's robotics push frames robots as a way to study AI systems that learn from physical interaction, not only web-scale data.
Reading lensRobotics makes intelligence tangible: perception, action, uncertainty, and feedback all arrive at once.
The New York Times · Turing patterns
The article revisits Alan Turing's morphogenesis work: mathematical rules that help explain how natural patterns can emerge from simple interactions.
Reading lensTuring's legacy is not only computation; it is also the idea that form itself can be modeled.
Meta Engineering · Production AI
Meta Engineering describes Facebook's move from phrase-based translation to neural machine translation across thousands of language directions.
Reading lensThe story is about research becoming infrastructure: better models mattered only after they could run reliably at global scale.
WIRED · Civic tech
WIRED presents President Obama's challenge list for the technology industry, asking builders to address major public problems rather than only optimize products.
Reading lensThe enduring question is whether technical power can be steered toward civic capacity.
WIRED · Public AI
President Obama discusses AI, autonomous vehicles, labor, safety, and governance with MIT's Joi Ito and WIRED's Scott Dadich.
Reading lensIt is an early mainstream conversation about AI as public infrastructure, not merely a private product category.
TechCrunch · Latent language
TechCrunch explains a then-new Google translation result where multilingual neural models appeared to learn shared internal representations across languages.
Reading lensThe fascinating part is representation: models can discover abstractions that were never manually named.
MIT Technology Review Insights · Coordination, not scale
The argument is that the next step is horizontal rather than vertical, not a larger single model but many agents able to hold a goal in common, and that today they cannot: the piece cites a study putting the failure rate of seven open-source multi-agent systems between 41 and 87 percent, and calls the gap architectural rather than a prompting problem. Its proposed remedy is an "Internet of Cognition" layered over agent connectivity, with cognition state protocols for shared intent, a cognition fabric for shared context, and cognitive amplifiers with guardrails for shared reasoning.
Reading lensRead it as a vendor architecture proposal, not a finding. This is Insights, the custom content arm, produced in partnership with Outshift by Cisco, and not written by the magazine's editorial staff. The diagnosis still travels further than the remedy: coordination, not capability, is where multi-agent systems currently break.
Reward learning in the brain, the politics of scientific authority, the history of artificial light, and what still shapes human evolution.
MIT Technology Review · Reward learning
The piece connects reinforcement learning ideas with dopamine-neuron behavior, showing how reward prediction can illuminate both AI and biology.
Reading lensGood AI concepts often become more interesting when they are used as hypotheses about natural intelligence.
The New York Times / VnExpress · Culture and creativity
A culture-and-history essay asking how education, argument, migration, and community norms can shape unusually dense creative achievement.
Reading lensThe strongest reading is sociological: talent compounds inside cultures that reward learning, debate, and transmission.
Nature · Artificial light
Nature reviews histories of lighting technology and the social consequences that followed when night became increasingly engineered.
Reading lensTechnologies that look purely practical often redraw social life, work rhythms, and public space.
Nature · Research systems
A Nature essay tracing China's rise as a scientific power through institutions, national ambition, talent flows, and long-term investment.
Reading lensScientific strength is built as an ecosystem, not as a single laboratory success story.
YouTube · Public discourse
A public conversation with Brian Greene about science, belief, public reasoning, and the responsibility of scientists in a polarized culture.
Reading lensThe durable lesson is that science communication must connect rigor with patience and civic humility.
Nature · Trust
Nature argues that researchers should resist partisan reduction and clearly restate why scholarship matters to society.
Reading lensPublic trust is sustained by visible usefulness, intellectual honesty, and the courage to explain evidence under pressure.
World Economic Forum / The Conversation · Modern evolution
Laurence D. Hurst explains why human evolution has not stopped and may be shaped by mutation, culture, medicine, and demographic change.
Reading lensEvolution is not a finished prehistoric event; it is a slow background process still interacting with modern life.
The Things Inside / YouTube · Truth
A reconstruction of the famous Einstein-Tagore conversations on reality, consciousness, beauty, and whether truth is independent of human experience.
Reading lensIt is a rare dialogue where physics and poetry argue without trying to erase each other.
Vietnam's opening to the world, the rise of intangible capital, and what currency collapse looks like from inside a Caracas coffee shop.
World Economic Forum / Quartz · Globalization
The article highlights Vietnam's trade-to-GDP scale, export orientation, manufacturing integration, and the risks of relying deeply on global demand.
Reading lensVietnam's openness is a growth engine, but it also makes resilience and upgrading strategy non-negotiable.
Facebook / YouTube · Labor markets
A classroom-style argument about sweatshops, industrialization, worker choice, and the difficult trade-offs inside early-stage development.
Reading lensThe sharp question is whether moral judgment should start from ideal standards or from the alternatives actually available to workers.
World Economic Forum · Development
The article explains Vietnam's post-reform growth through exports, investment, poverty reduction, education, and participation in global supply chains.
Reading lensThe miracle is best read as compounding policy choices, not as luck.
Gates Notes · Capital without capital
Bill Gates reviews Capitalism Without Capital, emphasizing how software, knowledge, brands, networks, and other intangible assets reshape growth.
Reading lensWhen value becomes less physical, measurement, investment, and policy all become harder.
Bloomberg · Hyperinflation
Bloomberg's Cafe Con Leche Index turned hyperinflation into a concrete image: the price of an ordinary coffee jumping into the millions of bolivars.
Reading lensMacroeconomic collapse becomes legible when it is measured through the daily things people must still buy.
Quartz · Savings culture
Quartz examines Germany's savings culture as a blend of history, policy, household behavior, and attitudes toward debt and prudence.
Reading lensEconomic behavior is never purely numerical; it carries memory, institutions, and moral habits.
World Economic Forum · Institutional fragility
The piece traces Venezuela's crisis through oil dependence, policy controls, inflation, production decline, shortages, and institutional stress.
Reading lensResource wealth can mask fragility until governance, incentives, and monetary discipline break at the same time.
Dân trí · US trade lessons
A Vietnamese analysis of how countries navigate economic relationships with the United States, comparing lessons from South Korea and the Philippines.
Reading lensExternal opportunity becomes national advantage only when domestic strategy is clear.
Productivity gaps between countries, the four-day week, commencement advice worth rereading, and how research talent actually moves.
Business Insider · Four-day week
A large real-world four-day-workweek trial reported stronger retention, better well-being, and stable or improved business performance.
Reading lensProductivity is better framed as system design and focus quality than as the raw count of hours spent at work.
Thanh Niên · Diaspora and return
A profile of Professor Trương Nguyện Thành reflecting on leaving Vietnam, building an academic life in the United States, and returning home.
Reading lensIt is a migration story about obligation: what knowledge owes to the place that first formed it.
BBC · National productivity
The BBC asks why workers in some economies produce much more per hour, linking productivity to capital, management, skills, and organization.
Reading lensThe gap is rarely about laziness; it is usually about systems that either amplify or waste human effort.
Facebook / Harvard · Purpose
Mark Zuckerberg's Harvard commencement address focused on purpose, community, global connection, and the responsibilities of a generation shaped by technology.
Reading lensThe speech matters less as celebrity advice and more as an early platform-era argument about social obligation.
Gates Notes · Career advice
Bill Gates encourages graduates to look toward artificial intelligence, clean energy, biosciences, and work that reduces inequity.
Reading lensThe practical message is to choose fields where competence and social leverage can compound together.
Knote / HuffPost / Quora · Work culture
A set of readings around Germany's productivity culture, shorter work hours, training systems, focus norms, and institutional efficiency.
Reading lensThe core lesson is that disciplined boundaries can be a productivity asset, not a luxury.
The Conversation · Talent mobility
The Conversation argues that researchers in fast-moving technical fields should be free to move between academia and industry as opportunities evolve.
Reading lensTalent mobility is not only a hiring issue; it is part of how knowledge circulates.
National memory, civic responsibility, and the cultural habits that quietly decide how institutions behave.
Facebook · National memory
A reflective post centered on Vietnamese unity, civic memory, and the emotional force of seeing the country as one continuous historical community.
Reading lensRead it as a compact reminder that identity is not only geography; it is also shared responsibility.
Người Lao Động · Civic culture
A viral letter, reported by Vietnamese media, sparked debate about public behavior, social norms, and how overseas perspectives can challenge a community.
Reading lensThe value is in the conversation it opens: criticism becomes useful only when it leads to self-correction rather than defensiveness.
Facebook Note · Business ethos
A short Vietnamese reflection on doing business with discipline, public-mindedness, and a moral vocabulary rooted in national history.
Reading lensThe interesting tension is between entrepreneurship as private initiative and entrepreneurship as civic character.
Standing references kept for quick lookup rather than one-time reading: venue and journal rankings, bibliographic indexes, institutional league tables, integrity checks, assurance standards, and the documents that argue about how any of it should be used.
CSRankings · Research output
CSRankings provides a faculty-publication-based view of computer science departments, configurable by research area and time range. It is entirely metrics-based, counting publications by tenure-track faculty at a fixed set of selective conferences drawn from dblp, which makes it harder to game than survey or citation-based tables.
Reading lensRankings are most useful when treated as lenses for evidence, not as final judgments of quality. Narrow it to your own area before reading anything into the order.
CORE · Venue quality
The CORE list has grown into ICORE, an international collaboration between CORE in Australasia, the Italian groups GII and GRIN, and the Spanish computer science society SCIE. Venues are graded A*, A, B or C from citation rates, acceptance rates and the standing of the programme committee, with the 2026 round now in the searchable portal.
Reading lensVenue lists are helpful for orientation, but they should be balanced with community fit, review quality, and research goals. Note that the tiers move between rounds, so cite the edition you used.
SCImago · Prestige-weighted citations
SJR ranks Scopus-indexed journals by prestige-weighted citations, so a citation arriving from a highly cited venue counts for more than one from a marginal title. Each subject category is then cut into quartiles, Q1 for the top quarter down to Q4 for the bottom.
Reading lensAlways check which category a quartile refers to. A journal indexed under several Scopus categories can be Q1 in one and Q3 in another, and the promotional copy will quote the better one.
Clarivate · Impact Factor and indexing
Clarivate released the 2026 Journal Citation Reports in June 2026, covering 22,643 journals across 254 subject categories, with 521 titles receiving a Journal Impact Factor for the first time. The Master Journal List is the companion tool for checking whether a venue is genuinely indexed and in which collection.
Reading lensUse the Master Journal List before the Impact Factor. Whether a journal sits in SCIE, ESCI or nowhere at all decides far more than the decimal places do.
Elsevier · Four-year citation window
CiteScore averages citations over a four-year window rather than two, which makes it steadier than the Impact Factor and fairer to fields where work is cited slowly. The 2025 edition, released in June 2026, covers publications and citations from 2022 to 2025.
Reading lensA longer window flatters slow-burning engineering venues and cools down fashionable ones. Read CiteScore alongside the Impact Factor rather than choosing between them.
Google Scholar · Conferences ranked as venues
Scholar Metrics ranks publications by h5-index over a rolling five-year window, and it is the only widely used venue metric that lists selected engineering and computer science conferences on the same table as journals.
Reading lensIn a field where the flagship venues are conferences, this is often the fastest honest answer to whether somewhere is a serious place to publish.
China Computer Federation · A / B / C tiers
The CCF catalogue sorts international venues into A, B and C tiers across ten computing domains. The seventh edition arrived in March 2026, adding 14 journals and 23 conferences, and it counts only full or regular papers, never short or workshop contributions.
Reading lensRead CCF next to ICORE rather than instead of it. The venues where the two lists disagree are usually the ones worth thinking about.
Community archives · Selectivity over time
Community-maintained tables track submissions and acceptances year by year. CVPR 2026 took 4,090 of 16,092 papers and ICML 2026 took 6,352 of 23,918, both close to a quarter, while robotics venues such as IROS run far higher and RSS far lower.
Reading lensAcceptance rate measures crowding, not quality. A robotics conference near forty per cent and a vision conference near twenty-five can carry identical weight inside their own communities.
Open-source trackers · Submission calendar
Open-source countdown boards for AI, vision, language and robotics deadlines, with time-zone handling and per-area filters. The Hugging Face space is the most actively maintained descendant of the original ai-deadlines project, while ccfddl covers the CCF-ranked venues.
Reading lensKeep two of them open. Any single tracker drifts out of date, and in conference-driven fields a missed deadline costs most of a year.
Schloss Dagstuhl · The substrate under the rankings
dblp indexes millions of computing publications with curated author disambiguation, clean BibTeX exports and ORCID links. It is also the data source CSRankings runs on, so an error on a dblp author page propagates straight into the rankings built from it.
Reading lensTreat your own dblp page as infrastructure you maintain, not as a passive record someone else keeps for you.
OurResearch · Open bibliometrics
Built by the non-profit OurResearch as a successor to Microsoft Academic Graph after it was retired in January 2022, OpenAlex now covers hundreds of millions of works, authors, institutions and venues under a CC0 licence with a free API.
Reading lensThis is what makes independent bibliometrics possible at all. When a ranking claims its data are reproducible, OpenAlex is usually the reason it can.
CWTS Leiden · Transparent bibliometrics
The Leiden Ranking covers roughly 2,800 universities across 120 countries using OpenAlex data, and reports impact, collaboration and open access separately in fields including mathematics and computer science. Both the underlying data and the source code are published.
Reading lensIts most useful feature is a refusal. There is no single overall score, which forces you to state which dimension you actually care about before you compare anything.
Nature Index · Output in selected journals
Nature Index counts institutional authorship across a fixed set of high-quality journals. The 2026 tables carry its largest methodological change since 2014: applied-science and social-science titles were added after surveying more than four thousand researchers, alongside a new article-level classification.
Reading lensThe counts are not normalised for size, so large institutions lead by construction. Read it as a map of where output concentrates, not of who does the best work.
QS / THE / ShanghaiRanking · Three league tables
The three best-known global tables disagree because they measure different things. QS leans heavily on reputation surveys, THE blends teaching, research environment and citations, and ARWU counts Nobel and Fields laureates alongside highly cited researchers. In the most recent editions the top place went to MIT, Oxford and Harvard respectively.
Reading lensThree methodologies, three winners. Whenever a single rank is quoted at you, the interesting question is which of these produced it.
SCImago · Research, innovation, society
SIR ranks more than ten thousand organisations, including universities, government laboratories, hospitals, companies and non-profits, on a composite of research at half the weight, patent-based innovation at thirty per cent and societal reach at twenty, drawing on Scopus and PATSTAT.
Reading lensThe only major table that places a corporate research laboratory and a university on the same axis, which is exactly the comparison an industry-facing researcher needs.
Stanford HAI · Annual measurement
The annual AI Index compiles capability benchmarks, investment, adoption, policy and workforce data into one citable reference. The 2026 edition adds dedicated chapters on AI in science and AI in medicine, and reports organisational adoption near ninety per cent against governance that is visibly lagging behind capability.
Reading lensThe most useful single citation when a claim about AI progress needs a defensible number attached to it.
Arena · Human preference as a metric
Arena ranks language models by blind pairwise votes fed into an Elo-style rating, now spanning millions of votes and hundreds of models, with separate boards for vision, code and image generation. It rebranded from LMArena in January 2026.
Reading lensGaps of under roughly ten Elo points are noise. Read the confidence intervals before you read the order.
Hugging Face · A lost commons
Papers with Code closed in July 2025, taking with it around fifteen hundred leaderboards and the links between papers and their implementations. Hugging Face now carries the nearest replacement through its papers feed, model hub and hosted leaderboards, but the unified task-level view has not been rebuilt.
Reading lensA reminder that community infrastructure is not permanent. Archive the benchmark table your paper depends on instead of linking to it and hoping.
DORA / CWTS / CoARA · Using metrics responsibly
Three standing statements on how research should be evaluated. DORA rejects the Journal Impact Factor as a proxy for individual quality and has thousands of institutional signatories, the Leiden Manifesto sets out ten principles for using indicators, and CoARA coordinates the reform commitments made since 2022.
Reading lensThe natural companion to every other card in this section. Rankings answer where, never how good, and these documents say so precisely enough to cite in a committee.
Think. Check. Submit. · Before you submit
A short checklist for judging whether a journal or conference can be trusted, paired with the DOAJ whitelist of vetted open access journals. The working sequence is to confirm indexing, verify the ISSN at the ISSN Portal, check COPE or OASPA membership, then look up two or three members of the editorial board.
Reading lensPredatory venues routinely claim a DOAJ listing they do not hold. Verify outward from the directory, never inward from the journal's own page.
Crossref / PubPeer · After publication
The Retraction Watch database was acquired by Crossref and made a public resource, so retractions and corrections now surface through the Crossref API and an openly licensed dataset covering tens of thousands of records reaching back to the 1970s. PubPeer hosts the post-publication comment threads that often precede a retraction.
Reading lensWorth running across a reference list before submission. A retracted citation is cheap to catch beforehand and expensive to explain afterwards.
Bộ KH&CN / HĐGSNN · Domestic journal standards
Vietnam now maintains two lists that matter. The Ministry of Science and Technology issued the first national list of journals meeting Vietnamese science standards in April 2026 under Decision 2244/QĐ-BKHCN, recognising 341 titles, while the State Council of Professors continues to publish the annual scored list used for academic titles and promotion.
Reading lensDomestic careers are still measured against these lists rather than against Scopus quartiles alone. Check both before deciding where a paper goes.
RTCA / SAE / ISO · Certification baseline
The certification stack that dependable autonomy work eventually meets: ARP4754A for aircraft and system development, DO-178C for airborne software and DO-254 for airborne hardware, ISO 26262 with its ASIL A to D scale on the automotive side, and ISO 23247 for the manufacturing digital twin framework, whose part on the digital thread was published in 2026.
Reading lensNone of these were written with learning-enabled components in mind. That gap is precisely where much of the current advanced air mobility assurance research sits.
ORCID / ROR / Crossref · Who wrote what, where
A researcher accumulates several identifiers issued by unrelated organisations, an ORCID iD, a Scopus Author ID and a Web of Science ResearcherID, which can be linked to each other but are never merged into one. ROR does the same job for institutions and DOIs for the works themselves.
Reading lensLink them once and keep ORCID authoritative. Every downstream system, from dblp to funding agencies, resolves ambiguity by asking it first.
arXiv / OpenReview · The open record
arXiv remains the default preprint venue for computing, but since late 2025 its computer science category no longer accepts survey or position papers unless they have already cleared peer review, a direct response to a flood of machine-generated submissions. OpenReview holds the public review threads for most major AI conferences.
Reading lensOpenReview is badly underused as a reading source. For a contested paper, the reviews often teach more than the paper itself.