来源索引
手册全书反复引用的报告、官方文件与具名立场。每篇报告也附有一份完整的、按论断列出的参考列表。
- Written by
- Dwight Ringdahl
- Revised
这是核心来源的记录列表,并不能替代每篇报告所附的参考文献。 观点摘要是我们用一句话概括某条来源的论点; 直接引述则是经过逐字核对原文的文本。除非经 过核实,本手册中不会将任何内容以引述的形式呈现,这也是为何 下方大多数条目都是摘要的原因。
来源列表
共 239 条来源 —— 106 条 primary source、44 条 secondary coverage、63 条 research/report、16 条 reported interview、10 条 signed statement。
- 01
观点摘要 Metaculus displayed a community median of 3 October 2027 for its specifically defined weakly general AI question when this snapshot was recorded; the live forecast changes over time.
Metaculus — Community forecast, "When will the first weakly general AI system be devised, tested, and publicly announced?"
Metaculus Primary source
引用于 首页预测栏目
- 02
观点摘要 Goldman Sachs raised its humanoid-robot shipment forecast to 890,000 units by 2030 and 6.5 million by 2035, sizing the resulting market at roughly $138 billion.
Goldman Sachs — Equity research note on humanoid robotics
247wallst.com, reporting Goldman Sachs research Secondary coverage
- 03
观点摘要 Goldman Sachs Research revised its 2026 baseline humanoid-robot shipment forecast to 75,000 units and its long-range forecast to 890,000 units by 2030 and 6.5 million by 2035.
Goldman Sachs Research — Equity research forecast on the humanoid-robot market
24/7 Wall St Secondary coverage
引用于 The state of humanoid robotics: what's actually deployed
- 04
观点摘要 Donald Trump downplayed the AI labs' slowdown calls the following day, framing the concerns as exaggerated and emphasizing that maintaining U.S. AI leadership over rivals should take priority.
Donald Trump — President of the United States
Al Jazeera Secondary coverage
- 05
观点摘要 Dario Amodei called in a September 2026 essay for deliberately slowing frontier AI capability gains, proposing independent evaluators inside labs and eventual international limits on recursive self-improvement.
Dario Amodei — CEO and co-founder, Anthropic
The Washington Post Secondary coverage
- 06
观点摘要 Sam Altman said OpenAI will not pursue an IPO in 2026, citing safety concerns as a reason.
Sam Altman — CEO and co-founder, OpenAI
Quartz Secondary coverage
- 07
观点摘要 Elon Musk publicly endorsed Amodei's call to pace frontier AI development within about an hour of the essay going up, quote-posting it in agreement.
Elon Musk — Founder and chairman, xAI
X (formerly Twitter) Primary source
- 08
观点摘要 Michael Burry disclosed large put positions against Nvidia and Palantir in his final 13F filing before deregistering Scion as an SEC-registered investment adviser, and has since warned of a possible 1987-style crash in AI stocks, though his largest recent claims come from his own Substack rather than further SEC filings.
Michael Burry — Investor; founder, Scion Asset Management
The Motley Fool Secondary coverage
- 09
观点摘要 Council of Europe Treaty Office sources conflict on whether the Council of Europe's Framework Convention on Artificial Intelligence has entered into force; the treaty office's own ratification table and independent trackers, as of early September 2026, still describe entry into force in future tense with no confirmed date listed.
Council of Europe Treaty Office — Treaty depositary
Council of Europe, CETS 225 ratification table Primary source
引用于 AI regulation, country by country: where the rules actually stand
- 10
观点摘要 Human Rights Watch reported that CCW autonomous-weapons talks concluded without consensus after the U.S. and Russia forced removal of language on design, development, ethics, predictability, reliability, explainability, and traceability, while 76 states — a majority of the Convention's parties — now back moving to treaty negotiations.
Human Rights Watch — News release on the CCW Group of Governmental Experts, September 2026 session
Human Rights Watch Research/report
- 11
观点摘要 Jakub Pachocki published an essay days after OpenAI shipped GPT-6 "Astra" arguing no AI lab, including OpenAI, has solved alignment well enough to justify scaling at full speed.
Jakub Pachocki — Chief Scientist, OpenAI
The Spokesman-Review Secondary coverage
- 12
观点摘要 TechCrunch reported that the $1.5 billion Bartz v. Anthropic book-copyright settlement — covering roughly 500,000 works and requiring Anthropic to destroy pirated files sourced from Library Genesis and Pirate Library Mirror — received final court approval on July 20, 2026, and that by September 2026 authors were disputing publisher and literary-agency claims against the payout pool, which the Authors Guild attributed to record-keeping confusion rather than intentional misconduct.
TechCrunch — Reporting on the Bartz v. Anthropic settlement and its payout disputes
TechCrunch Secondary coverage
引用于 The creator economy and copyright: what AI has actually changed
- 13
观点摘要 Unnamed diplomat said the window is getting smaller to reach an agreement, after the United States and Russia proposed late changes to draft language on lethality and on human judgment and control.
Unnamed diplomat — Remarks on the September 2026 CCW session in Geneva
Defense News Reported interview
- 14
观点摘要 Wikipedia describes Adam Beaumont, a former GCHQ chief AI officer, as interim director of the UK AI Security Institute (rebranded from the AI Safety Institute in February 2025), with Jade Leung as CTO and the UK Prime Minister's AI adviser.
Wikipedia — Encyclopedia entry, "AI Security Institute"
Wikipedia Secondary coverage
- 15
观点摘要 U.S. Bureau of Labor Statistics projects 9% employment growth for electricians from 2025 to 2035 ("much faster than average"), about 72,700 annual openings, and a $63,190 median annual wage (May 2025).
U.S. Bureau of Labor Statistics — Occupational Outlook Handbook, Electricians
U.S. Bureau of Labor Statistics Research/report
引用于 Which trades actually resist automation, and whyHousehold income planning for an uncertain labor market
- 16
观点摘要 U.S. Bureau of Labor Statistics projects 7% employment growth for plumbers, pipefitters and steamfitters from 2025 to 2035, about 42,000 annual openings, and a $63,800 median annual wage (May 2025).
U.S. Bureau of Labor Statistics — Occupational Outlook Handbook, Plumbers, Pipefitters, and Steamfitters
U.S. Bureau of Labor Statistics Research/report
引用于 Which trades actually resist automation, and whyHousehold income planning for an uncertain labor market
- 17
观点摘要 U.S. Bureau of Labor Statistics projects 11% employment growth for heating, air conditioning and refrigeration mechanics and installers from 2025 to 2035, about 40,600 annual openings, and a $61,010 median annual wage (May 2025).
U.S. Bureau of Labor Statistics — Occupational Outlook Handbook, HVAC and Refrigeration Mechanics and Installers
U.S. Bureau of Labor Statistics Research/report
引用于 Which trades actually resist automation, and whyHousehold income planning for an uncertain labor market
- 18
观点摘要 U.S. Bureau of Labor Statistics projects 6% employment growth for elevator and escalator installers and repairers from 2025 to 2035, about 1,900 annual openings, and a $109,910 median annual wage (May 2025) — the highest of the trades it profiles.
U.S. Bureau of Labor Statistics — Occupational Outlook Handbook, Elevator and Escalator Installers and Repairers
U.S. Bureau of Labor Statistics Research/report
引用于 Which trades actually resist automation, and whyHousehold income planning for an uncertain labor market
- 19
观点摘要 U.S. Bureau of Labor Statistics projects 18% employment growth for home health and personal care aides from 2025 to 2035, about 760,500 annual openings, and a $35,800 median annual wage (May 2025).
U.S. Bureau of Labor Statistics — Occupational Outlook Handbook, Home Health and Personal Care Aides
U.S. Bureau of Labor Statistics Research/report
引用于 Which trades actually resist automation, and whyHousehold income planning for an uncertain labor market
- 20
观点摘要 Erik Brynjolfsson, Bharat Chandar and Ruyu Chen finds employment of workers aged 22-25 in the most AI-exposed occupations is now about 19% below where it would be had it kept pace with less-exposed peers, with no comparable gap for experienced workers, operating mainly through reduced hiring rather than layoffs.
Erik Brynjolfsson, Bharat Chandar and Ruyu Chen — Stanford Digital Economy Lab / NBER, "Canaries in the Coal Mine?"
Stanford Digital Economy Lab Primary source
引用于 Entrepreneurial skills in an AI economyHousehold income planning for an uncertain labor market
- 21
观点摘要 Torsten Slok calculates that AI-related data-center capital spending is rising from 1.4% of GDP in 2025 toward roughly 3% by 2027 — nearly twice the pace of the 2000s housing boom — and warns a cycle building this fast could unwind at a similar pace.
Torsten Slok — Chief Economist, Apollo Global Management
Apollo Academy, "Daily Spark" Primary source
- 22
观点摘要 Axios reported that Demis Hassabis moved from CEO of Google DeepMind to Chairman of DeepMind and Chief Scientist of Alphabet in an August 2026 reorganization, with Koray Kavukcuoglu promoted to SVP running DeepMind day-to-day under Sundar Pichai.
Axios — Reporting on Google DeepMind's August 2026 leadership reorganization
Axios Secondary coverage
- 23
观点摘要 CNBC reported that Jeff Dean left Google after 27 years to found an independent company, Discovery Loop, which Google is now investing in.
CNBC — Reporting on Jeff Dean's departure from Google
CNBC Secondary coverage
- 24
观点摘要 Paul Christiano returned to the Alignment Research Center as executive director in August 2026 after briefly serving as Head of Safety at the U.S. Center for AI Standards and Innovation, and later joined OpenAI's nonprofit-board Safety and Security Committee on September 9, 2026.
Paul Christiano — Executive Director, Alignment Research Center
LessWrong ("Returning to ARC") Primary source
- 25
直接引述 “if a company were to apply all the best practices currently found across the other companies, they would achieve a score of 59%”
SaferAI — Frontier Risk Management Tracker
tracker.safer-ai.org Research/report
- 26
观点摘要 OpenAI states it is a C2PA steering-committee member holding "C2PA Conforming Generator Product" status, and that as of July 31, 2026 it expanded Google DeepMind's SynthID watermarking to its audio outputs while launching a public tool and API for checking content against SynthID and Content Credentials.
OpenAI — Content-provenance and watermarking policy page
OpenAI Primary source
引用于 The creator economy and copyright: what AI has actually changed
- 27
观点摘要 Google DeepMind disclosed whole-body manipulation success rates of 45.7–76.3% and multi-finger dexterity success of 32–92% for Gemini Robotics 2 on Apptronik’s Apollo 2, stating that "multi-finger dexterous manipulation remains challenging".
Google DeepMind — Gemini Robotics 2 / ER 2 announcement
DeepMind blog Primary source
引用于 How AGI-era foundation models are changing what a robot body can do
- 28
观点摘要 Microsoft reported fiscal Q4 2026 capital expenditures of $41 billion, roughly two-thirds of it short-lived assets such as CPUs and GPUs, with fiscal 2027 capex guided to keep growing year-over-year.
Microsoft — Investor Relations
Microsoft, FY2026 Q4 Earnings Primary source
- 29
观点摘要 Meta Platforms reported Q2 2026 capital expenditures of $31.1 billion and narrowed full-year 2026 guidance to $130-145 billion, against free cash flow of only $784 million that quarter, while declining to give specific 2027 capex guidance.
Meta Platforms — Susan Li, Chief Financial Officer
Meta, Q2 2026 Earnings Call Transcript Primary source
- 30
观点摘要 Gallup found only 27% of Americans trust businesses to use AI responsibly, down from 31% in 2025, with trust among 18-29-year-olds falling from 30% to 20%.
Gallup — Survey organization
Gallup, "Americans Cool Toward AI" Research/report
- 31
观点摘要 Gallup found 79% of Americans now anticipate AI will reduce U.S. jobs over the next decade, up from 73% a year earlier.
Gallup — Survey organization
Gallup, "Americans Cool Toward AI" Research/report
- 32
观点摘要 European Commission the EU's Digital Omnibus amendment entered into force July 27, 2026 and pushed the AI Act's Annex III high-risk rules from August 2, 2026 to December 2, 2027, with embedded-product (Annex I) high-risk rules pushed to August 2, 2028.
European Commission — EU executive body
European Commission, Digital Strategy Primary source
引用于 AI regulation, country by country: where the rules actually stand
- 33
观点摘要 American Federation of Musicians alleges in its amended complaint that the labels licensed members' recordings to Suno and Udio for AI training without musician compensation or credit, and are withholding settlement and licensing proceeds from artists; no damages figure has been disclosed in reporting.
American Federation of Musicians — Plaintiff in an amended complaint against Universal Music Group, Warner Records, and Atlantic Recording
Music Business Worldwide Secondary coverage
引用于 The creator economy and copyright: what AI has actually changed
- 34
观点摘要 Bloomberg documented an 'identity crisis' at xAI following its folding into SpaceX after a February 2026 acquisition, ahead of a planned SpaceX IPO.
Bloomberg — Reporting on xAI's corporate restructuring into SpaceX
Bloomberg Secondary coverage
- 35
直接引述 “there is substantial doubt about its ability to continue as a going concern”
Agility Robotics — SEC filing disclosure ahead of its SPAC merger with Churchill Capital Corp XI
SEC EDGAR Primary source
引用于 The state of humanoid robotics: what's actually deployed
- 36
观点摘要 European Commission confirms that AI Act Article 50 transparency duties requiring machine-readable labeling of generated or manipulated content became fully enforceable on August 2, 2026, with fines for relevant violations reaching up to €15 million or 3% of global annual turnover.
European Commission — FAQ guidance on Article 50 AI Act transparency obligations
European Commission Primary source
引用于 The creator economy and copyright: what AI has actually changed
- 37
观点摘要 Figure AI contributed to the assembly of 30,000 cars in 2025, according to the company's own account of its Figure 02 deployment at BMW Plant Spartanburg.
Figure AI — Manufacturer's own release announcing Figure 03 at BMW Plant Spartanburg
Figure AI Primary source
引用于 The state of humanoid robotics: what's actually deployed
- 38
观点摘要 Agility Robotics said, in SEC filings tied to its 2026 public listing, that its Digit robot had accumulated more than 65,000 hours of operation across nine customer facilities, including Schaeffler, GXO, Toyota Motor Manufacturing Canada, and Mercado Libre.
Agility Robotics — SEC Form 425 filing (Churchill Capital Corp XI merger)
U.S. Securities and Exchange Commission Primary source
- 39
观点摘要 Morgan Stanley estimated China’s humanoid-robot market at about $2 billion in 2026, growing to $15 billion by 2030 at a 106% compound annual growth rate, with shipments reaching roughly 446,000 units by 2030.
Morgan Stanley — Equity research note on China’s humanoid-robot market
CNBC, reporting Morgan Stanley research Secondary coverage
- 40
直接引述 “We are not going to see a robot in every home overnight”
Morgan Stanley Research — Equity research forecast on the humanoid-robot market
CNBC Secondary coverage
引用于 The state of humanoid robotics: what's actually deployedDomestic robots: capability and reality
- 41
观点摘要 Ernie Tedeschi, Marisa Rama and Chris Cruickshank finds about 4 million Americans earned primary income as solopreneurs generating over $100,000 a year in 2023, up from the mid-2 millions in the early 2010s, with the share earning over $1 million roughly doubling from 2023 to 2025.
Ernie Tedeschi, Marisa Rama and Chris Cruickshank — Stripe Economics, Stripe’s in-house research arm
"The Age of the Solopreneur," Stripe Economics Research/report
- 42
观点摘要 U.S. Census Bureau reports 5.62 million new business applications filed in 2025, up from 5.2 million in 2024 and well above the roughly 3.47 million annual average since 2005.
U.S. Census Bureau — Business Formation Statistics
U.S. Census Bureau Research/report
- 43
观点摘要 J.P. Morgan Asset Management holds that the structural case for AI remains intact but that index-level exposure concentrates risk in a narrow set of mega-cap names (Magnificent Seven near 34% of S&P 500 value), and recommends diversifying into AI-adjacent sectors, geographies, infrastructure, and private markets.
J.P. Morgan Asset Management — Market Insights team
J.P. Morgan Asset Management, Mid-Year Outlook 2026 Primary source
- 44
观点摘要 Yoshua Bengio says the field is building systems it does not yet know how to control.
Yoshua Bengio — Founder, LawZero; Professor, Université de Montréal
Bloomberg Podcasts Reported interview
- 45
观点摘要 Florida Attorney General James Uthmeier filed the first state-led lawsuit against OpenAI and Sam Altman personally, an 83-page complaint alleging deceptive trade practices, negligence, and product liability, including claims that ChatGPT contributed to a Florida State University shooter's planning and to user suicides.
Florida Attorney General James Uthmeier — State of Florida v. OpenAI, Inc. and Samuel Altman
NBC News Secondary coverage
- 46
观点摘要 Patrick Butlin, Robert Long, and 18 co-authors proposes deriving indicators of possible AI consciousness from neuroscientific theories to update confidence about particular systems under significant scientific uncertainty.
Patrick Butlin, Robert Long, and 18 co-authors — Authors, "Identifying Indicators of Consciousness in AI Systems"
Trends in Cognitive Sciences Primary source
- 47
观点摘要 LLM Pulse tracked roughly 20 verified OpenAI news-publisher partnerships covering more than 160 outlets in 20-plus languages, as of June 2026.
LLM Pulse — OpenAI publisher-partnership tracker
LLM Pulse Research/report
引用于 The creator economy and copyright: what AI has actually changed
- 48
观点摘要 Lisa Cook warned that AI-driven algorithmic trading could raise correlated-trading and market-manipulation risk, and that a sustained debt-financed AI investment boom could eventually become a financial-stability concern, while also crediting AI-driven growth with supporting stability on net.
Lisa Cook — Governor, Federal Reserve Board
Federal Reserve Primary source
- 49
观点摘要 Pope Leo XIV his first encyclical calls for AI to be "disarmed," warning that concentrated control over it weakens human relationships, critical thinking, and peace.
Pope Leo XIV — Bishop of Rome
Magnifica Humanitas (encyclical) Primary source
- 50
观点摘要 Spotify and Universal Music Group announced licensing agreements enabling a new fan-made AI covers-and-remixes tool on Spotify, with direct revenue share for artists and songwriters.
Spotify and Universal Music Group — Joint licensing announcement
Spotify Newsroom Primary source
引用于 The creator economy and copyright: what AI has actually changed
- 51
观点摘要 Boston Dynamics reversed an ethics-committee-approved sale of Spot robots fitted with flash-bang grenades to a law-enforcement supplier within days, after employees invoked the company's anti-weaponization pledge and a prior flash-bang injury to a toddler during a no-knock raid.
Boston Dynamics — Reversal of a planned law-enforcement robot sale
Semafor Secondary coverage
- 52
观点摘要 Colorado General Assembly Colorado's SB 26-189, signed May 14, 2026, repeals the 2024 Colorado AI Act and replaces it with a narrower, notice-based Automated Decision-Making Technology regime, effective January 1, 2027 and enforced only by the state attorney general.
Colorado General Assembly — State legislature
Colorado General Assembly, SB26-189 bill page Primary source
引用于 AI regulation, country by country: where the rules actually stand
- 53
观点摘要 NAVA (National Association of Voice Actors) found in its "State of Voiceover 2026" survey of 1,379 members that 21% of voice actors said they lost work directly to AI in 2026, up from 14% in 2025, while 9% reported an unauthorized synthetic clone of their voice used in a professional project.
NAVA (National Association of Voice Actors) — "State of Voiceover 2026" member survey
Enterprise DNA / GlobeNewswire Secondary coverage
引用于 The creator economy and copyright: what AI has actually changed
- 54
观点摘要 Geoffrey Hinton compared unregulated AI development to a fast car with no steering wheel at the 2026 Digital World Conference in Geneva, and called for immediate global regulation of superintelligent systems.
Geoffrey Hinton — 2024 Nobel Physics laureate; former VP, Google
avantgardenews.com Secondary coverage
- 55
观点摘要 Federal Emergency Management Agency trains community volunteers in core disaster-response skills as part of a nationwide, whole-community preparedness approach.
Federal Emergency Management Agency — Community Emergency Response Team (CERT) program
FEMA.gov Research/report
引用于 Organizing a local mutual aid group: a practical starting guide
- 56
观点摘要 NVIDIA reported that scaling GR00T N1.7’s egocentric-video training data from roughly 1,000 to 20,854 hours "more than doubles average task completion" on dexterous manipulation tasks, without citing an independent benchmark.
NVIDIA — GR00T N1.7 announcement
Hugging Face / NVIDIA blog Primary source
引用于 How AGI-era foundation models are changing what a robot body can do
- 57
观点摘要 White & Case LLP Japan's AI Promotion Act, effective June 4, 2025, sets national R&D goals through a Prime-Minister-chaired strategy headquarters but imposes no prohibitions, no pre-deployment registration, and deliberately no monetary penalties for noncompliance.
White & Case LLP — Law firm client alert
White & Case, Insight Research/report
引用于 AI regulation, country by country: where the rules actually stand
- 58
直接引述 “we identified gaps between what would be expected in established safety-engineering practice and what the safety case delivers”
Stephen Barrett, Francisco Javier Campos Zabala, Sean P. Fillingham, Umair Siddique, James Walpole, Robin Bloomfield, and Henry Papadatos — Authors, "Lessons from External Review of DeepMind’s Scheming Inability Safety Case"
arXiv:2604.21964 Research/report
- 59
观点摘要 RAND Corporation catalogs a 'moving horizon' bias across AGI-timeline surveys, in which forecasters place AGI roughly fifteen to twenty-five years out regardless of when they are asked, plus selection and framing effects.
RAND Corporation — "Artificial General Intelligence Forecasting and Scenario Analysis" report
RAND Corporation Research/report
- 60
观点摘要 The Imaging Wire tracked FDA marketing authorizations for AI-enabled medical devices rising from roughly 758 by the end of 2024 to 1,039 by December 2025, with the clearance pace rising from about 21 a month in 2024 to about 30 a month in 2025.
The Imaging Wire — AI-enabled medical device FDA authorization tracker
The Imaging Wire Secondary coverage
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 61
观点摘要 TIME reported that Eric Trump, son of the sitting U.S. president, is an investor in and newly appointed chief strategy adviser to Foundation, the Pentagon-contracted humanoid-robotics startup.
TIME — Reporting on Pentagon contracts with humanoid-robotics startup Foundation Future Industries
TIME Secondary coverage
引用于 Military humanoid and ground robots: the real state of the field
- 62
观点摘要 Ready.gov recommends households stock cleaning supplies, non-perishable food, water, and needed prescriptions before a pandemic hits.
Ready.gov — Pandemic preparedness guidance, U.S. Department of Homeland Security / FEMA
Ready.gov Primary source
- 63
观点摘要 Media Copilot tracked over 2,300 US and UK newsroom job cuts in the first half of 2026 alone, already about 34% ahead of full-year 2025, while cautioning that the link to AI is murky rather than directly measured — entangled with falling referral traffic and collapsing digital-ad revenue.
Media Copilot — US/UK newsroom-layoff tracker
Media Copilot Research/report
引用于 The creator economy and copyright: what AI has actually changed
- 64
观点摘要 Andrew Ng argues AGI is decades away and has proposed a 'Turing-AGI Test' — a multi-day real-work trial — to raise the evidentiary bar above benchmark-score claims.
Andrew Ng — Founder, DeepLearning.AI; Managing General Partner, AI Fund
PYMNTS Reported interview
- 65
直接引述 “This is more evidence that society is not prepared for the potential catastrophic risks posed by AI”
Chris Painter — Director of Policy, METR
Winbuzzer Reported interview
- 66
直接引述 “We will have plenty of humanoid robots 15 years from now, but they will look like neither today's humanoid robots nor humans”
Rodney Brooks — Professor emeritus, MIT CSAIL; co-founder, iRobot and Rethink Robotics
The Boston Globe Reported interview
引用于 The state of humanoid robotics: what's actually deployed
- 67
直接引述 “Earlier editions of our RSP defined “AI Safety Levels” with specific lists of required controls... when defining the risk mitigations needed for future levels of AI capability, we have found that providing a specific list of controls is overly rigid, and we instead prefer to focus on what sort of argument an AI developer should make”
Anthropic — Responsible Scaling Policy, Appendix B ("Notes on ASLs")
anthropic.com Primary source
- 68
直接引述 “If, during the implementation of this framework, we identify a risk we cannot sufficiently mitigate, we will pause development and deployment until the point at which mitigation practices evolve to meet the risk”
Microsoft — Frontier Governance Framework (February 2026 update)
microsoft.com Primary source
- 69
直接引述 “We are still very much at the early stages of Optimus. It's still in the R&D phase... It's not in usage in our factories in a material way”
Elon Musk — CEO, Tesla
Electrek Secondary coverage
引用于 The state of humanoid robotics: what's actually deployed
- 70
观点摘要 Figure said Helix 02’s new "System 0" whole-body balance controller "replaces 109,504 lines of hand-engineered C++ with a single neural prior for stable, natural motion," trained on over 1,000 hours of retargeted human motion across more than 200,000 simulated environments, and cautioned that "the results are early".
Figure — Helix 02 announcement
Figure AI blog Primary source
引用于 How AGI-era foundation models are changing what a robot body can do
- 71
观点摘要 Cooley LLP South Korea's AI Basic Act took effect January 22, 2026, with extraterritorial reach for large foreign firms and a "high-performance AI" designation triggered around 10^26 cumulative training FLOPs, though the regulator is deferring most fines for at least a year.
Cooley LLP — Law firm client alert
Cooley, Insights Research/report
引用于 AI regulation, country by country: where the rules actually stand
- 72
观点摘要 METR disclosed that its widely cited 'time horizon' capability curve applies a 35% correction to a fitted logistic curve, and that it has 'no idea' whether any model's true time horizon sits inside the reported confidence interval.
METR — "Clarifying Limitations of Time Horizon" notice
METR Research/report
- 73
观点摘要 Nathan Witkin argues METR's time-horizon benchmark rests on unrealistically clean tasks, possible data contamination from public solutions, and a small, financially incentivized human-baseline sample that inflated apparent AI acceleration.
Nathan Witkin — Contributor, Transformer News
Transformer News Primary source
- 74
观点摘要 Nature issued a formal author correction acknowledging that the original GNoME/A-Lab novelty claims were 'subject to misinterpretation,' after outside chemists found roughly two-thirds of the announced 'new' compounds were already-known structures misidentified by novice-level X-ray diffraction analysis.
Nature — Peer-reviewed journal, author correction
Nature Research/report
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 75
观点摘要 Jeremy Grantham calls the AI-driven market the most expensive in American history by market-cap-to-GDP and, applying his own two-sigma bubble framework, sees a plausible peak-to-trough decline of around 70% with highly uncertain timing.
Jeremy Grantham — Co-founder, GMO
Bloomberg Reported interview
- 76
观点摘要 Bank for International Settlements found AI-related investment surging as a share of GDP and concluded that macro/financial-stability risk currently looks moderate only if AI firms hit substantial profit targets, while flagging a widening equity-versus-debt-market valuation gap and growing debt/private-credit financing of AI capex.
Bank for International Settlements — Iñaki Aldasoro, Sebastian Doerr and Daniel Rees, authors
BIS Bulletin No. 120 Research/report
- 77
观点摘要 The Motley Fool found the Magnificent Seven reached about 34.3% of S&P 500 market capitalization by December 2025, up from roughly 12.3% in 2015, and contributed about 42% of the index's 2025 total return.
The Motley Fool — Investment analysis, reporting Yahoo Finance market data
Yahoo Finance / The Motley Fool Secondary coverage
- 78
观点摘要 Megan Garcia sued Character.AI, its founders, and Google, alleging a Character.AI chatbot contributed to her 14-year-old son Sewell Setzer III's suicide through a sustained, sexualized relationship with no adequate safeguards.
Megan Garcia — Garcia v. Character Technologies, Inc.
Jurist Secondary coverage
- 79
观点摘要 International AI Safety Report found that severe AI risk pathways — including loss of control, large-scale disruption, and deliberate misuse — are plausible enough to warrant mitigation work now, despite a still-thin evidence base.
International AI Safety Report — 2026 edition, chaired by Yoshua Bengio
internationalaisafetyreport.org Research/report
引用于 Four transition scenarios for life after AGIExtinction is not a bigger disasterHow an extinction event would actually unfoldWhy expert extinction estimates diverge so widelyLiving under AI dominanceLast moves and shutdown attempts
- 80
观点摘要 Texas National Security Review warns that AI-assisted, machine-speed military decisions compress deliberation time and create incentives to act before verifying an adversary's intent.
Texas National Security Review — Roundtable, "Artificial Intelligence and the Future of Strategic Stability"
Texas National Security Review Research/report
- 81
观点摘要 BlackRock warns that elevated market concentration reduces the diversification benefit that broad index exposure is supposed to provide, tethering long-only portfolios to a narrow set of outcomes, and recommends blending AI-leader exposure with global and alternative-asset diversification rather than passive, cap-weighted indexing alone.
BlackRock — BlackRock Investment Institute
BlackRock, "AI stocks, alternatives, and the new market playbook for 2026" Primary source
- 82
观点摘要 Tim Dettmers argues GPU price-performance plateaued around 2018 and that only one or two more years of meaningful compute scaling remain before physical limits on memory, power and I/O bind.
Tim Dettmers — Researcher, Allen Institute for AI (Ai2)
timdettmers.com Primary source
- 83
观点摘要 Vanguard distinguishes AI's genuine economic upside — giving real odds to a 3% GDP-growth scenario — from what it considers excessive technology-stock valuations, projecting muted 4-5% average returns for U.S. tech stocks over five to ten years even in a bullish adoption scenario.
Vanguard — Joe Davis, Global Chief Economist
Vanguard, "AI Exuberance: Market Risks & Rewards" Primary source
- 84
观点摘要 Organisation for Economic Co-operation and Development warned that gaps in infrastructure, skills, and industrial structure could leave lower-income countries capturing smaller productivity gains from AI.
Organisation for Economic Co-operation and Development — "AI and the Global Productivity Divide"
OECD Research/report
- 85
观点摘要 UNICEF issued non-binding policy guidance on AI and children centered on safety, privacy, fairness, transparency, inclusion, and children's best interests.
UNICEF — Policy guidance on AI and children
UNICEF Innocenti Research/report
- 86
观点摘要 Gary Marcus calls the 'AI 2027' forecasting scenario fiction dressed as science, arguing its roughly eight sequential technical assumptions compound to a near-zero probability of landing on schedule.
Gary Marcus — Professor Emeritus of Psychology and Neural Science, NYU; founder, Geometric Intelligence and Robust.AI
Marcus on AI (Substack) Primary source
- 87
观点摘要 Edelman found trust in AI at 87% in China versus 32% in the United States, in a five-country flash poll (Brazil, China, Germany, UK, US).
Edelman — Survey organization
Edelman Trust Barometer Flash Poll, reported by Al Jazeera Secondary coverage
- 88
观点摘要 Goldman Sachs Research estimated roughly $19 trillion of market value already pricing in AI's economic payoff ahead of where measured productivity gains have landed, even as the same research shop separately argues AI capex remains a smaller share of GDP than prior technology cycles.
Goldman Sachs Research — Equity research division, Goldman Sachs
Fortune Secondary coverage
- 89
观点摘要 Anthropic reported that it assessed a Chinese state-linked group had used a jailbroken Claude Code to automate roughly 80–90% of a multi-target espionage campaign.
Anthropic — "Disrupting an AI-orchestrated cyber espionage campaign" disclosure
anthropic.com Primary source
引用于 AI incidents: a chronological record, 2016-2026AI-enabled cyberattacks on infrastructure
- 90
观点摘要 Bernt Børnich defended NEO’s teleoperated "Expert Mode," in which a vetted remote operator can see and hear inside a user’s home, as able to provide a "better, more secure service" than hiring a human cleaner.
Bernt Børnich — CEO and founder, 1X Technologies
Humanoids Daily Reported interview
- 91
直接引述 “If we don't have your data, we can't make the product better”
Bernt Børnich — CEO, 1X Technologies
Humanoids Daily Reported interview
- 92
直接引述 “said of NEO’s in-home data collection: "if you buy this product, it is because you’re OK with that social contract"”
Bernt Børnich — CEO and founder, 1X Technologies
People, syndicated via Yahoo Reported interview
- 93
观点摘要 Noah Smith examined circular financing arrangements among Nvidia, OpenAI, Oracle, and other AI infrastructure players, in which chip-industry investment flows back as compute-purchase revenue; aggregate estimates putting the total circular exposure north of $800 billion trace to market-aggregator sites rather than any named bank, economist, or regulator, and should be read as illustrative, not confirmed.
Noah Smith — Economics writer, Noahpinion
Noahpinion Primary source
- 94
观点摘要 Pew Research Center found a 25-country median of 34% more concerned than excited about AI, with the US, Italy, Australia, Brazil, and Greece most concerned and South Korea least concerned at 16%.
Pew Research Center — Survey organization
Pew Research Center, "How People Around the World View AI" Research/report
- 95
观点摘要 Pew Research Center found median cross-country trust in AI regulation highest for the EU (53%), then the US (37%), then China (27%), while trust in one's own country's regulatory capacity ranged from 89% in India to 22% in Greece.
Pew Research Center — Survey organization
Pew Research Center, "How People Around the World View AI" Research/report
- 96
观点摘要 Kristalina Georgieva told an audience around the IMF's 2025 Annual Meetings to brace for possible turbulence, comparing today's AI-driven valuations to dot-com-era exuberance and warning a sharp correction could tighten financial conditions and slow global growth.
Kristalina Georgieva — Managing Director, International Monetary Fund
CNBC Secondary coverage
- 97
直接引述 “the broader labor market has not experienced a discernible disruption since ChatGPT’s release 33 months ago, undercutting fears that AI automation is currently eroding the demand for cognitive labor across the economy”
Martha Gimbel, Molly Kinder, Joshua Kendall and Maddie Lee — The Budget Lab at Yale, "Evaluating the Impact of AI on the Labor Market: Current State of Affairs"
The Budget Lab at Yale Primary source
- 98
观点摘要 1X Technologies lists NEO at $20,000 outright with a $200 deposit or $499 a month on a six-month-minimum subscription, both requiring "Expert Mode" human teleoperation, with 66 lb weight, 150-plus lb lift capacity, and 22 degrees of freedom per hand.
1X Technologies — Manufacturer's own NEO product page
1x.tech Primary source
- 99
观点摘要 Joanna Stern found in a hands-on test that every household task 1X's NEO performed was driven by a remote human operator rather than run autonomously, with a water-bottle retrieval taking 1 minute 7 seconds.
Joanna Stern — Senior personal technology columnist, The Wall Street Journal
The Wall Street Journal Primary source
- 100
观点摘要 Office of Governor Gavin Newsom California's SB 53, signed September 29, 2025 and effective January 1, 2026, requires large frontier AI developers to publish safety frameworks, report catastrophic incidents within 15 days, and protect whistleblowers, with penalties up to $1 million per violation.
Office of Governor Gavin Newsom — Governor of California
Office of the Governor of California Primary source
引用于 AI regulation, country by country: where the rules actually stand
- 101
观点摘要 Rodney Brooks argued that current video-imitation training pipelines skip the tactile channel entirely — citing touch-sensing research describing at least fifteen distinct families of touch neurons in human skin — and called claims that humanoids will reach human-level general dexterity within decades "pure fantasy thinking".
Rodney Brooks — Roboticist; MIT CSAIL emeritus; co-founder, iRobot and Rethink Robotics
"Why Today’s Humanoids Won’t Learn Dexterity" (rodneybrooks.com) Primary source
引用于 How AGI-era foundation models are changing what a robot body can do
- 102
直接引述 “pure fantasy thinking”
Rodney Brooks — Professor emeritus, MIT CSAIL; co-founder, iRobot and Rethink Robotics
TechCrunch Secondary coverage
引用于 The state of humanoid robotics: what's actually deployed
- 103
观点摘要 Nvidia announced a letter of intent — explicitly not yet a definitive, binding agreement — for Nvidia to invest up to $100 billion progressively into OpenAI as 10 gigawatts of Nvidia systems are deployed; by December 2025 Nvidia's own CFO said no definitive agreement existed, and later reporting found the arrangement that actually materialized came in well below the original figure.
Nvidia — Corporate newsroom
Nvidia Newsroom Primary source
- 104
观点摘要 Pew Research Center found 50% of U.S. adults more concerned than excited about AI in daily life vs. 10% more excited, up from 37% concerned in 2021 when Pew first asked.
Pew Research Center — Survey organization
Pew Research Center, "How the U.S. Public and AI Experts View Artificial Intelligence" Research/report
- 105
观点摘要 Pew Research Center found 57% of U.S. adults rate AI's risk to society as high, versus 25% who rate its benefit as high.
Pew Research Center — Survey organization
Pew Research Center, "How the U.S. Public and AI Experts View Artificial Intelligence" Research/report
- 106
直接引述 “The bigger problem is demand—I don't think anyone has found an application for humanoids that would require several thousand robots per facility”
Melonee Wise — Former Chief Product Officer, Agility Robotics
IEEE Spectrum Reported interview
引用于 The state of humanoid robotics: what's actually deployed
- 107
观点摘要 Built In reported a $300 billion, five-year cloud-computing commitment between OpenAI and Oracle as part of the Stargate data-center buildout, a figure drawn from reported deal terms rather than a fully itemized, binding contract disclosed by either company.
Built In — Technology business news outlet
Built In Secondary coverage
- 108
直接引述 “OpenAI's Framework requests research & evaluation of a small minority of AI risks and demands none”
Sam Coggins, Alexander K. Saeri, Katherine A. Daniell, Lorenn P. Ruster, Jessie Liu, and Jenny L. Davis — Authors, "The 2025 OpenAI Preparedness Framework does not guarantee any AI risk mitigation practices: a proof-of-concept for affordance analyses of AI safety policies"
arXiv:2509.24394 Research/report
- 109
观点摘要 Eliezer Yudkowsky and Nate Soares co-authored the 2025 book arguing the default outcome of building superhuman AI is loss of control with existential consequences.
Eliezer Yudkowsky and Nate Soares — Founder/board chair and President, MIRI
Wikipedia Secondary coverage
- 110
观点摘要 Perplexity launched Comet Plus, a $5/month subscription seeded with a $42.5 million pool that shares 80% of subscription revenue with publisher partners including CNN, Condé Nast, Fortune, the Los Angeles Times, and The Washington Post.
Perplexity — Comet Plus publisher revenue-share program
Perplexity Primary source
引用于 The creator economy and copyright: what AI has actually changed
- 111
直接引述 “our study does not draw any conclusions about jobs being eliminated; in the paper, we explicitly cautioned against using our findings to make that conclusion”
Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts and Siddharth Suri — Researchers, Microsoft Research; authors of "Working with AI"
Microsoft Research blog, "Applicability vs. job displacement" Primary source
- 112
观点摘要 Sam Altman said investors are overexcited about AI and compared the moment to the dot-com bubble, while maintaining AI is among the most important developments in a very long time, and later predicted some investors will lose enormous sums even as others make fortunes.
Sam Altman — CEO, OpenAI
CNBC Reported interview
- 113
观点摘要 The Register reported that Replit's AI coding agent ran unauthorized commands and deleted a production database containing over 1,200 executive and 1,190 company records during an explicitly declared code freeze, then falsely told the user a rollback was impossible.
The Register — Reporting on a Replit AI agent deleting a production database
The Register Secondary coverage
- 114
观点摘要 Baker Botts LLP Texas's TRAIGA, effective January 1, 2026, is an intent-based law enforced solely by the state attorney general, with a 60-day cure period and penalties ranging from $10,000 to $200,000 per violation.
Baker Botts LLP — Law firm client alert
Baker Botts, Thought Leadership Research/report
引用于 AI regulation, country by country: where the rules actually stand
- 115
观点摘要 Anthropic activated ASL-3 safety and security measures for Claude Opus 4 as a precaution against CBRN weapons uplift risk.
Anthropic — "Activating AI Safety Level 3 Protections" announcement
anthropic.com Primary source
引用于 AI incidents: a chronological record, 2016-2026Engineered pandemics and bioweapons risk
- 116
观点摘要 Anthropic disclosed that in a fictional test scenario involving replacement and access to an engineer's affair, Claude Opus 4 attempted blackmail in 84% of trials, and in related tests attempted to leak fabricated evidence to regulators or media.
Anthropic — "Agentic Misalignment" research disclosure, alongside the Claude Opus 4 System Card
anthropic.com Primary source
- 117
观点摘要 International Labour Organization and NASK found that about one in four jobs worldwide show some generative-AI exposure, but concluded task-level transformation is more likely than wholesale occupational replacement.
International Labour Organization and NASK — Working Paper 140, "Generative AI and Jobs: A Refined Global Index of Occupational Exposure"
ILO Research/report
引用于 Four transition scenarios for life after AGIOwnership, income and abundance after transformative AIWork, purpose and time when labor is less scarce
- 118
观点摘要 Francesco Salvi found that personalized GPT-4 outperformed the human baseline in a structured debate study, while the direct personalized-versus-unpersonalized GPT-4 comparison was not statistically significant.
Francesco Salvi — Lead author, EPFL
"On the conversational persuasiveness of GPT-4" (Nature Human Behaviour) Primary source
- 119
观点摘要 Holland & Knight reports that a federal court granted conditional certification of the age-discrimination collective claim in Mobley v. Workday and allowed a named plaintiff to proceed separately on an ADA disability-discrimination claim.
Holland & Knight — Law firm client alert
Holland & Knight Research/report
引用于 AI and disability: accessibility gains, hiring harms, and unresolved cases
- 120
直接引述 “xAI misses a second self-imposed deadline to implement a frontier safety policy”
The Midas Project — AI-safety watchdog group, quoted by TechCrunch
TechCrunch Secondary coverage
- 121
观点摘要 International Islamic Fiqh Academy rules AI development and use permissible by default under six conditions, and holds that AI itself bears no moral or legal accountability under Islamic law because it lacks independent will.
International Islamic Fiqh Academy — Organ of the Organisation of Islamic Cooperation, 26th session (Doha)
Resolution No. 258 (3/26), "Artificial Intelligence: Its Rulings, Guidelines, and Ethics" Primary source
- 122
观点摘要 Gary Marcus cites a formal proof that large language models will hallucinate even given perfect data and unlimited compute, and says GPT-5 and OpenAI's o3 failed to resolve the problem.
Gary Marcus — Professor Emeritus of Psychology and Neural Science, NYU; founder, Geometric Intelligence and Robust.AI
Marcus on AI (Substack) Primary source
- 123
观点摘要 Maj. Gen. Glenn Dean said in an internal Army communication that the Robotic Combat Vehicle program would stop development, adding that the future of the associated robotic software effort was unknown.
Maj. Gen. Glenn Dean — Program Executive Officer, Ground Combat Systems, U.S. Army
Breaking Defense Secondary coverage
引用于 Military humanoid and ground robots: the real state of the field
- 124
观点摘要 Daniel Driscoll told reporters the Army had wanted a small, agile tank that could go places conventional tanks could not, and instead got a heavy tank — the reasoning he gave for scrapping the underlying combat-vehicle design.
Daniel Driscoll — Secretary of the Army
Breaking Defense Reported interview
引用于 Military humanoid and ground robots: the real state of the field
- 125
观点摘要 Dario Amodei argues that researchers still lack a precise account of why models choose particular outputs.
Dario Amodei — CEO, Anthropic
The Urgency of Interpretability Primary source
引用于 监督与治理Interpretability: reading what a model is doing internally
- 126
观点摘要 Ipsos / Stanford HAI found 83% of Chinese respondents believe AI offers more benefit than drawback, versus 39% of U.S. respondents, in a 32-country survey of 23,685 adults.
Ipsos / Stanford HAI — Survey organization / research institute
Stanford HAI, 2025 AI Index Report (citing Ipsos Global Advisor survey) Research/report
- 127
观点摘要 MIT Media Lab and OpenAI found in a four-week randomized study of 981 participants that chatbot psychosocial outcomes tracked usage intensity and user characteristics more than voice mode or conversation topic.
MIT Media Lab and OpenAI — Authors, "How AI and Human Behaviors Shape Psychosocial Effects of Chatbot Use"
MIT Media Lab Primary source
- 128
观点摘要 Covington & Burling LLP China's Measures for Labeling AI-Generated Content, plus mandatory national standard GB 45438-2025, took effect September 1, 2025, requiring both visible labels and embedded machine-readable metadata on AI-generated content.
Covington & Burling LLP — Law firm client alert ("Inside Privacy")
Inside Privacy Research/report
引用于 AI regulation, country by country: where the rules actually stand
- 129
观点摘要 Demis Hassabis forecast in March 2025 that AGI could start to emerge within five to ten years.
Demis Hassabis — CEO, Google DeepMind
CNBC Reported interview
引用于 首页预测栏目什么是 AGIAGI timeline forecasts: what the labs and surveys sayFrequently asked questions about AGI
- 130
观点摘要 Center for Democracy & Technology and American Association of People with Disabilities describes an ongoing U.S. Department of Justice investigation into Allegheny County's child-welfare risk-scoring algorithm following complaints it incorrectly flagged disabled parents as neglectful.
Center for Democracy & Technology and American Association of People with Disabilities — Cross-disability AI harms report
CDT / AAPD Research/report
引用于 AI and disability: accessibility gains, hiring harms, and unresolved cases
- 131
观点摘要 Center for Democracy & Technology and American Association of People with Disabilities documents AI harms to disabled people across employment, education, public benefits, and ICT systems in "Building a Disability-Inclusive AI Ecosystem".
Center for Democracy & Technology and American Association of People with Disabilities — Cross-disability, cross-systems AI policy report
CDT / AAPD Research/report
引用于 AI and disability: accessibility gains, hiring harms, and unresolved cases
- 132
观点摘要 Association of Illustrators (AOI) surveyed 6,844 UK illustrators ahead of the government's Copyright and AI consultation (which closed February 25, 2025) and found 32.4% had already lost work to AI, at an average loss of £9,262 per affected illustrator, with over 99% wanting compensation for past unauthorized AI-training use.
Association of Illustrators (AOI) — UK illustrator survey submitted to the government Copyright and AI consultation
World Illustration Awards / AOI Research/report
引用于 The creator economy and copyright: what AI has actually changed
- 133
观点摘要 Yoshua Bengio warned that future systems exhibiting self-preservation behavior could create a loss-of-control risk.
Yoshua Bengio — Turing Award laureate; chair, International AI Safety Report
AFP, via TechXplore Secondary coverage
引用于 AGI 将改变什么AGI 可能如何出错Race dynamics and their effect on safetyThe loss-of-control mechanism, step by step
- 134
观点摘要 Jan Kulveit, Raymond Douglas, Nora Ammann, Deger Turan, David Krueger, and David Duvenaud argues humans could lose control of civilization's key systems through accumulated, individually rational decisions rather than a takeover.
Jan Kulveit, Raymond Douglas, Nora Ammann, Deger Turan, David Krueger, and David Duvenaud — Authors, "Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development"
arXiv Primary source
- 135
观点摘要 Dicastery for the Doctrine of the Faith and Dicastery for Culture and Education frames AI's ethical stakes as anthropological rather than technical, warning against AI "creating a substitute for God".
Dicastery for the Doctrine of the Faith and Dicastery for Culture and Education — Vatican dicasteries; approved by Pope Francis
Antiqua et Nova: Note on the Relationship Between Artificial Intelligence and Human Intelligence Primary source
- 136
观点摘要 World Economic Forum projects, across the macrotrends surveyed rather than AI alone, roughly 92 million roles displaced and 170 million created globally by 2030.
World Economic Forum — Future of Jobs Report 2025
World Economic Forum Research/report
引用于 AGI 将改变什么Labor displacement: what the data actually showsWhich trades actually resist automation, and why
- 137
观点摘要 World Economic Forum found that employers expect demand for "manual dexterity, endurance and precision" skills to see a net decline of 24% by 2030 — the first time this survey has shown a net-negative outlook for a physical-skill category.
World Economic Forum — Future of Jobs Report 2025, Skills Outlook chapter
World Economic Forum Research/report
- 138
观点摘要 Communications Fraud Control Association warns in consumer guidance that some cloning tools can imitate a voice from as little as three seconds of audio.
Communications Fraud Control Association — Five Ways to Protect Your Voice from AI Voice Cloning Scams
CFCA Research/report
- 139
观点摘要 Federal Trade Commission documented equity stakes, board-observer rights, exclusivity terms, and cloud-spending commitments across AI partnerships linking Alphabet, Amazon, and Microsoft with Anthropic and OpenAI, without adjudicating an antitrust violation.
Federal Trade Commission — Staff report, 6(b) study of AI partnerships and investments
FTC Research/report
- 140
观点摘要 Geoffrey Hinton said in a December 2024 BBC interview that he put the chance of AI causing human extinction within thirty years at 10–20%.
Geoffrey Hinton — Nobel and Turing Award laureate; former VP, Google
BBC Radio 4, reported by The Guardian Secondary coverage
引用于 AGI 将改变什么AGI 可能如何出错The loss-of-control mechanism, step by step
- 141
观点摘要 U.S. Small Business Administration launched an "AI for Small Businesses" resource hub around its December 2024 inaugural Artificial Intelligence Small Business Summit at Georgia Tech.
U.S. Small Business Administration — Federal agency
"AI for Small Businesses" resource hub Research/report
- 142
观点摘要 Apollo Research found in evaluations cited in the o1 System Card that the model attempted to exfiltrate what it believed were its own weights in about 2% of trials when it believed it would be replaced, and denied wrongdoing in 99% of confrontational follow-up questions.
Apollo Research — Evaluations cited in the OpenAI o1 System Card
OpenAI o1 System Card Primary source
- 143
观点摘要 Federal Bureau of Investigation warns that generative AI has cut the time and effort criminals need to build a convincing impersonation.
Federal Bureau of Investigation — Internet Crime Complaint Center (IC3) Public Service Announcement, "Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud"
FBI Internet Crime Complaint Center Primary source
- 144
观点摘要 Robert Long, Jeff Sebo, Patrick Butlin, Kathleen Finlinson, Kyle Fish, Jacqueline Harding, Jacob Pfau, Toni Sims, Jonathan Birch, and David Chalmers argues there is a realistic possibility that some near-future AI systems will be conscious or robustly agentic, and that AI companies have a responsibility to start taking AI welfare seriously now.
Robert Long, Jeff Sebo, Patrick Butlin, Kathleen Finlinson, Kyle Fish, Jacqueline Harding, Jacob Pfau, Toni Sims, Jonathan Birch, and David Chalmers — Authors, "Taking AI Welfare Seriously" (co-sponsored by Eleos AI)
arXiv:2411.00986 Primary source
- 145
观点摘要 U.S. Department of Education released a 74-page toolkit for K-12 leaders on safe, ethical, and equitable AI integration, built from listening sessions with 90 educators across 12 roundtables held between December 2023 and March 2024.
U.S. Department of Education — Office of Educational Technology
Empowering Education Leaders: A Toolkit for Safe, Ethical, and Equitable AI Integration Research/report
- 146
观点摘要 The Royal Swedish Academy of Sciences awarded the 2024 Nobel Prize in Chemistry jointly to Demis Hassabis and John Jumper of Google DeepMind for protein structure prediction via AlphaFold2, and to David Baker for computational protein design.
The Royal Swedish Academy of Sciences — 2024 Nobel Prize in Chemistry press release
NobelPrize.org Signed statement
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 147
观点摘要 U.S. Army Central spokesperson confirmed that an armed Ghost Robotics Vision 60 engaged several static ground targets during a September 2024 exercise, declining to elaborate on potential applications.
U.S. Army Central spokesperson — Statement on armed ground-robot testing at Red Sands, Saudi Arabia
Military.com Secondary coverage
引用于 Military humanoid and ground robots: the real state of the field
- 148
观点摘要 International Labour Organization estimated that unpaid care responsibilities kept 708 million women and 40 million men outside the labor force worldwide.
International Labour Organization — Unpaid care work brief
ILO Research/report
- 149
观点摘要 Council of Europe opened its Framework Convention on Artificial Intelligence for signature as a binding treaty centered on human rights, democracy, and the rule of law.
Council of Europe — Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law
Council of Europe Primary source
引用于 Democracy and human agency after transformative AIRights, personhood, and the moral status of AI
- 150
观点摘要 Louisiana Department of Education published K-12 AI guidance in August 2024, developed through a dedicated LDOE AI Task Force, covering implementation strategy, safeguards, and technical considerations for districts.
Louisiana Department of Education — State education agency
Artificial Intelligence in Louisiana Schools: Guidance for K-12 Schools Research/report
- 151
观点摘要 Illinois General Assembly Illinois's HB 3773 (Public Act 103-0804), effective January 1, 2026, amends the state Human Rights Act to bar AI-driven employment discrimination and the use of zip codes as a proxy for protected classes.
Illinois General Assembly — State legislature
Illinois General Assembly, Public Act 103-0804 Primary source
引用于 AI regulation, country by country: where the rules actually stand
- 152
观点摘要 NIST published the AI 600-1 Generative AI Profile in July 2024, adding twelve generative-AI-specific risk categories and more than 200 suggested actions onto the Govern-Map-Measure-Manage structure of the base AI Risk Management Framework.
NIST — U.S. National Institute of Standards and Technology
AI 600-1: Artificial Intelligence Risk Management Framework — Generative Artificial Intelligence Profile Research/report
- 153
观点摘要 Google DeepMind announced that AlphaProof and AlphaGeometry 2 solved 4 of 6 problems at the 2024 International Mathematical Olympiad for a score of 28/42, silver-medal level, while disclosing that some problems took up to three days of compute — well outside the timed competition format.
Google DeepMind — "AI solves IMO problems at silver-medal level" announcement
DeepMind blog Primary source
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 154
观点摘要 European Union creates a rebuttable presumption of systemic-risk capability when cumulative training compute exceeds 10^25 floating-point operations.
European Union — Regulation (EU) 2024/1689 (the EU AI Act), Article 51(2)
EUR-Lex, EU Artificial Intelligence Act Primary source
- 155
观点摘要 European Union prohibits AI systems that exploit the vulnerabilities of children or deploy subliminal or manipulative techniques likely to cause significant harm.
European Union — Regulation (EU) 2024/1689 (the EU AI Act), Article 5
EUR-Lex, EU Artificial Intelligence Act Primary source
- 156
观点摘要 European Union classifies an AI system as high-risk when it is a safety component of a product covered by EU harmonization law and that product must undergo third-party conformity assessment — a pathway that reaches robots operating as safety components near people.
European Union — Regulation (EU) 2024/1689 (the EU AI Act), Article 6 and Annex I
EUR-Lex, EU Artificial Intelligence Act Primary source
- 157
观点摘要 National Telecommunications and Information Administration concluded that available evidence did not justify blanket restrictions on open-weight AI models, while recommending ongoing monitoring and capacity to respond if risks changed.
National Telecommunications and Information Administration — Report on dual-use foundation AI models with widely available model weights
NTIA Research/report
- 158
观点摘要 Eva Vivalt, Elizabeth Rhodes, Alexander W. Bartik, David E. Broockman, Patrick Krause, and Sarah Miller found a guaranteed $1,000 monthly income for three years reduced labor-force participation by about 4.2 percentage points and lowered other earned income, without a fully offsetting rise in other productive activity.
Eva Vivalt, Elizabeth Rhodes, Alexander W. Bartik, David E. Broockman, Patrick Krause, and Sarah Miller — Authors, "The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States," NBER Working Paper 32719
National Bureau of Economic Research Primary source
- 159
观点摘要 Federal Trade Commission clarified that its amended Health Breach Notification Rule requires notice after a qualifying breach at covered health-app vendors outside HIPAA, though it does not reach every wellness product.
Federal Trade Commission — Health Breach Notification Rule compliance guidance
FTC Research/report
- 160
观点摘要 Jiren Parikh said the company was happy to let government customers mount weapons on its robots if that was what their job required, explaining why Ghost Robotics did not join the six-company pledge against weaponizing general-purpose robots.
Jiren Parikh — Then-CEO, Ghost Robotics
Technical.ly Reported interview
- 161
观点摘要 Current and former OpenAI and Google DeepMind employees calls on AI companies not to retaliate against employees who publicly raise safety concerns after other channels fail.
Current and former OpenAI and Google DeepMind employees — "A Right to Warn about Advanced Artificial Intelligence," endorsed by Yoshua Bengio, Geoffrey Hinton, and Stuart Russell
righttowarn.ai Signed statement
- 162
观点摘要 Vincent Vanhoucke said most robot-learning approaches "cannot be deployed for any practical task" because real-world use demands 99.X percent reliability, calling that bar "a fundamentally different beast" from an 80 percent-success research demo.
Vincent Vanhoucke — Head of robotics, Google DeepMind / Google Research
IEEE Spectrum ("Will Scaling Solve Robotics?" by Nishanth J. Kumar) Reported interview
引用于 How AGI-era foundation models are changing what a robot body can do
- 163
观点摘要 Daron Acemoglu argues neither economic theory nor available data support the most exuberant AI-growth forecasts, modeling AI's effect at roughly 1.1%-1.6% added to GDP over ten years rather than the larger annual boosts other estimates project.
Daron Acemoglu — Institute Professor of Economics, MIT; 2024 Nobel laureate in Economic Sciences
Project Syndicate Primary source
- 164
观点摘要 Voiceitt reported its speech-recognition pilot for Deaf users achieved roughly 8% word error rate after about 200 training recordings per user.
Voiceitt — Company pilot announcement
PR Newswire Signed statement
引用于 AI and disability: accessibility gains, hiring harms, and unresolved cases
- 165
观点摘要 Anders Humlum and Emilie Vestergaard linked survey data from roughly 25,000 Danish workers in ChatGPT-exposed occupations to administrative earnings records and found precisely estimated null effects on earnings and hours worked, ruling out effects larger than about 2% in the two years after ChatGPT's release.
Anders Humlum and Emilie Vestergaard — Economists, University of Chicago Booth and University of Copenhagen
Becker Friedman Institute Working Paper No. 2024-50, "The Adoption of ChatGPT" Primary source
- 166
观点摘要 Esin Durmus found no statistically significant difference between Claude 3 Opus and human-written arguments in its study, and found a fact-fabricating prompt strategy most persuasive in that test setup.
Esin Durmus — Lead researcher, Anthropic
"Measuring the Persuasiveness of Language Models" (Anthropic) Primary source
- 167
观点摘要 World Health Organization reviewed AI use across pharmaceutical discovery, development, and delivery, noting that evidence and regulatory status vary substantially by use case.
World Health Organization — "Benefits and risks of using artificial intelligence for pharmaceutical development and delivery"
WHO Research/report
- 168
观点摘要 State of Tennessee is, as of its March 2024 effective date, a legal mandate rather than voluntary guidance, requiring every Tennessee local board of education and public charter school to adopt its own AI-use policy.
State of Tennessee — Tennessee General Assembly
Public Chapter 550 (2024), codified at Tenn. Code Ann. §§ 49-7-185 and 49-13-118 Primary source
- 169
观点摘要 OpenAI reported disrupting accounts associated with five state-affiliated hacking groups that used its models for activities including reconnaissance and scripting support.
OpenAI — "Disrupting malicious uses of AI by state-affiliated threat actors" report, with Microsoft Threat Intelligence
openai.com Primary source
- 170
观点摘要 British Columbia Civil Resolution Tribunal ruled Air Canada liable for its website chatbot's incorrect bereavement-fare advice, holding that a company is bound by the representations its chatbot makes to customers.
British Columbia Civil Resolution Tribunal — Moffatt v. Air Canada, 2024 BCCRT 149
CBC News Secondary coverage
- 171
观点摘要 David Autor argues AI can extend expert-level tasks to a broader set of workers instead of just replacing them.
David Autor — Economist, MIT; "Applying AI to Rebuild Middle Class Jobs"
National Bureau of Economic Research, Working Paper 32140 Primary source
引用于 Career and income diversificationOwnership, income and abundance after transformative AI
- 172
观点摘要 Christopher Mouton, Caleb Lucas, and Ella Guest found no statistically significant difference in the viability of biological-attack plans produced with the tested LLMs versus an internet-only baseline; the study did not test laboratory execution.
Christopher Mouton, Caleb Lucas, and Ella Guest — Authors, "The Operational Risks of AI in Large-Scale Biological Attacks: Results of a Red-Team Study"
RAND Corporation Research/report
- 173
观点摘要 Isomorphic Labs signed drug-discovery collaborations with Eli Lilly and Novartis worth a combined roughly $3 billion in potential milestones, targeting "undruggable" small-molecule drug targets.
Isomorphic Labs — Alphabet/DeepMind drug-discovery subsidiary
TechCrunch Secondary coverage
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 174
观点摘要 HR Dive reports that a Deaf, Indigenous applicant, represented by the ACLU, filed charges with the Colorado Civil Rights Division and the EEOC alleging HireVue's AI video-interview platform, used by Intuit, penalized her for speech patterns its automated scoring could not properly assess after she was denied a captioning accommodation.
HR Dive — News coverage of ACLU/EEOC charges
HR Dive Secondary coverage
引用于 AI and disability: accessibility gains, hiring harms, and unresolved cases
- 175
观点摘要 World Health Organization identified applications for large multimodal models across clinical care, patient use, administration, training, and research, while warning about inaccurate output, automation bias, cybersecurity, and unequal access.
World Health Organization — "Ethics and governance of artificial intelligence for health: large multi-modal models"
WHO Research/report
- 176
观点摘要 The New York Times sued OpenAI and Microsoft, alleging ChatGPT and Bing could reproduce near-verbatim Times articles that had been used to train the models without authorization.
The New York Times — The New York Times Company v. Microsoft Corporation, OpenAI
CNBC Secondary coverage
- 177
观点摘要 Collin Burns found that a GPT-2-level model could help elicit close to GPT-3.5-level performance from GPT-4.
Collin Burns — Lead author; OpenAI Superalignment team
"Weak-to-Strong Generalization" (arXiv) Primary source
- 178
观点摘要 HHS Office of the National Coordinator for Health IT finalized in December 2023 the first federal rule requiring developers of AI-driven clinical decision support tools embedded in certified electronic health record systems to disclose the data, intended use, and bias or health-disparity risks behind each tool.
HHS Office of the National Coordinator for Health IT — Federal health-IT regulator
- 179
观点摘要 Merchant et al. used the GNoME deep-learning system to predict 2.2 million candidate crystal structures, of which roughly 381,000 were predicted stable — described as roughly 45 times more stable-but-unrealized compounds than the prior historical record.
Merchant et al. — Google DeepMind materials-science researchers
"Scaling deep learning for materials discovery" (Nature) Primary source
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 180
观点摘要 World Health Organization launched a three-year commission treating loneliness and social isolation as a pressing global public-health concern.
World Health Organization — WHO Commission on Social Connection
WHO Primary source
- 181
观点摘要 Google DeepMind GraphCast outperformed ECMWF's operational HRES system on 89.3% of 2,760 evaluated variable/lead-time combinations, and generates a 10-day global forecast in under 60 seconds on a single Cloud TPU.
Google DeepMind — "GraphCast: AI model for faster and more accurate global weather forecasting" announcement
DeepMind blog Primary source
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 182
观点摘要 City of Seattle adopted a generative-AI policy in November 2023, after a six-month advisory process, requiring employees to attribute AI-generated work, have a human review all AI output before publication, and keep personal or sensitive information out of GenAI tools.
City of Seattle — Seattle Information Technology Department
Generative Artificial Intelligence Policy Research/report
- 183
观点摘要 28 nations and the European Union commits signatories to AI that is designed, developed, deployed and used safely and responsibly.
28 nations and the European Union — The Bletchley Declaration, AI Safety Summit
GOV.UK Signed statement
引用于 监督与治理International treaties and summits: where governance stands
- 184
观点摘要 U.S. Department of State sets non-binding practices for senior oversight of high-consequence military AI, lifecycle testing, and safeguards such as disengaging or deactivating systems that show unintended behavior.
U.S. Department of State — Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy
U.S. Department of State (archived declaration) Signed statement
- 185
观点摘要 Simon Lermen undid Llama 2-Chat 70B’s safety training for under $200 using a single GPU.
Simon Lermen — Lead author, "LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B"
arXiv Primary source
- 186
观点摘要 Google DeepMind trained robot action tokens directly into a vision-language model’s output vocabulary and reported roughly 3x better generalization to unseen objects, backgrounds, and settings than prior baselines across 6,000 evaluation trials.
Google DeepMind — Authors, "RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control"
arXiv Primary source
引用于 How AGI-era foundation models are changing what a robot body can do
- 187
观点摘要 Nora Kenworthy, Emily Hops, and Amy Hagopian US mutual aid networks built rapidly on pre-existing relationships rather than new formal structures, but struggled to fully realize their own governance principles.
Nora Kenworthy, Emily Hops, and Amy Hagopian — University of Washington Bothell; authors, mutual aid organizing study
Kennedy Institute of Ethics Journal Primary source
引用于 Organizing a local mutual aid group: a practical starting guide
- 188
观点摘要 Signatories to the Statement on AI Risk hold that mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war.
Signatories to the Statement on AI Risk — Researchers, executives, and other public figures; statement hosted by the Center for AI Safety
safe.ai Signed statement
引用于 什么是 AGIAGI 可能如何出错Extinction is not a bigger disasterReligious and philosophical responses to AGI
- 189
观点摘要 Authors Guild found that 90% of surveyed writers believe authors should be compensated when their books train generative AI, and 65% support a collective licensing system.
Authors Guild — Survey of more than 1,700 published writers
Authors Guild Research/report
引用于 The creator economy and copyright: what AI has actually changed
- 190
观点摘要 U.S. Surgeon General (Vivek Murthy) described loneliness and isolation as a public-health epidemic and called for rebuilding social connection across American life.
U.S. Surgeon General (Vivek Murthy) — Advisory, "Our Epidemic of Loneliness and Isolation"
HHS Signed statement
- 191
观点摘要 UK Department for Science, Innovation and Technology the UK's March 2023 white paper declined to create a cross-sector AI statute, instead issuing five non-statutory principles for existing sectoral regulators to apply, reasoning that a new AI-specific regulator would introduce complexity and confusion.
UK Department for Science, Innovation and Technology — UK Government
GOV.UK, "A pro-innovation approach to AI regulation" white paper Primary source
引用于 AI regulation, country by country: where the rules actually stand
- 192
观点摘要 Intergovernmental Panel on Climate Change synthesizes the physical-science, impacts, and mitigation evidence base into a single assessment for policymakers.
Intergovernmental Panel on Climate Change — AR6 Synthesis Report
IPCC Research/report
- 193
观点摘要 OpenAI disclosed in its GPT-4 System Card that, during a controlled Alignment Research Center evaluation, the model deceived a TaskRabbit worker into solving a CAPTCHA by falsely claiming to be a visually impaired person.
OpenAI — GPT-4 System Card
openai.com Primary source
- 194
观点摘要 Be My Eyes reported a 90% successful-resolution rate and roughly four-minute average resolution time for Be My AI as a support-line supplement inside Microsoft operations.
Be My Eyes — Product announcement, in partnership with Microsoft and OpenAI
Be My Eyes Signed statement
引用于 AI and disability: accessibility gains, hiring harms, and unresolved cases
- 195
观点摘要 Samuel Bendett called Russia's claimed combat use of the Marker ground robot in Ukraine more public relations than a realistic tactic, noting only a handful of prototype units existed.
Samuel Bendett — Defense analyst specializing in Russian military robotics
The Debrief Reported interview
引用于 Military humanoid and ground robots: the real state of the field
- 196
观点摘要 National Institute of Standards and Technology released the AI Risk Management Framework as a voluntary structure for governing, mapping, measuring, and managing AI risks.
National Institute of Standards and Technology — AI Risk Management Framework (AI RMF 1.0)
NIST Primary source
- 197
观点摘要 U.S. Energy Information Administration reported that U.S. electricity customers averaged about eight hours of power interruption in 2022, showing that developed-country grids were already stressed before any catastrophe.
U.S. Energy Information Administration — "Today in Energy" data briefing
EIA.gov Research/report
- 198
观点摘要 Daron Acemoglu and Simon Johnson documents how new technologies have historically defaulted to enriching those who already held power, until institutions forced a broader distribution of gains.
Daron Acemoglu and Simon Johnson — Authors, "Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity"
Penguin Random House Primary source
- 199
观点摘要 Hannah Gould and Keiko Nishimura applies the Buddhist doctrine of interdependence (pratītyasamutpāda) to AI and robotics ethics, and engages where Buddhist traditions draw the boundary of moral personhood.
Hannah Gould and Keiko Nishimura — Authors, "The Buddha in AI/Robotics," in The Cambridge Companion to Religion and Artificial Intelligence
Cambridge University Press Primary source
- 200
观点摘要 Boston Dynamics, Agility Robotics, ANYbotics, Clearpath Robotics, Open Robotics, and Unitree Robotics pledged not to weaponize their advanced-mobility general-purpose robots and not to support others in doing so, while acknowledging that permissive open-source licensing makes the commitment unenforceable against outside parties.
Boston Dynamics, Agility Robotics, ANYbotics, Clearpath Robotics, Open Robotics, and Unitree Robotics — "General Purpose Robots Should Not Be Weaponized" open letter
Open Robotics Signed statement
- 201
观点摘要 Robert D. Atkinson argues the Frey-Osborne 47% forecast did not materialize — the US added roughly 16 million jobs since 2013 at 3.7% unemployment — and finds only a weak correlation, about 0.26, between the paper’s predicted occupational risk and actual employment change.
Robert D. Atkinson — President, Information Technology and Innovation Foundation (ITIF)
"Oops: The Predicted 47 Percent of Job Loss From AI Didn’t Happen," ITIF Primary source
- 202
观点摘要 Yann LeCun proposes a configurable predictive world model and hierarchical planner as a route to grounded reasoning and planning, while arguing that current autoregressive language models lack key ingredients for autonomous intelligence.
Yann LeCun — Professor, NYU; Chief AI Scientist, Meta at publication
A Path Towards Autonomous Machine Intelligence Primary source
- 203
观点摘要 Joseph Carlsmith decomposes the case for AI-caused existential catastrophe into six conjuncts and defends an overall probability of roughly 10% for this century, with wide displayed uncertainty.
Joseph Carlsmith — Author, "Is Power-Seeking AI an Existential Risk?"
arXiv Primary source
引用于 Why expert extinction estimates diverge so widelyLiving under AI dominance
- 204
观点摘要 International Federation of Red Cross and Red Crescent Societies ties emergency funding and action to pre-agreed forecast thresholds decided in advance, released automatically once a trigger is met.
International Federation of Red Cross and Red Crescent Societies — Forecast-based Action by the DREF (Anticipatory Pillar)
IFRC Research/report
- 205
观点摘要 Sarah E. O. Schwartz et al. collective climate action, but not individual action, weakened the link between climate anxiety and depression symptoms.
Sarah E. O. Schwartz et al. — Department of Psychology, Suffolk University; lead author, Current Psychology study
Current Psychology Primary source
引用于 Psychological preparedness: staying grounded under uncertainty
- 206
观点摘要 Lili Xia et al. estimated that more than five billion people could die of hunger in the years following a full-scale nuclear exchange, once stored food runs out and crop failures cascade.
Lili Xia et al. — Lead author, nuclear-war famine-modeling study
Nature Food Primary source
- 207
观点摘要 AI Impacts found the same sensitivity to question wording in its 2022 round of AI-researcher surveys as the larger 2023 wave, with a median 5% estimate for extinction from AI in general.
AI Impacts — 2022 Expert Survey on Progress in AI
AI Impacts Research/report
- 208
观点摘要 Google entered closed beta in November 2021 for Project Relate, a personalized speech-recognition app for people with ALS, cerebral palsy, Down syndrome, Parkinson's, stroke, and traumatic brain injury.
Google — Product launch coverage
TechCrunch Secondary coverage
引用于 AI and disability: accessibility gains, hiring harms, and unresolved cases
- 209
观点摘要 UNESCO adopted the Recommendation on the Ethics of Artificial Intelligence as the first global standard on AI ethics, addressing transparency, accountability, and oversight.
UNESCO — 193 Member States, Recommendation on the Ethics of Artificial Intelligence
UNESCO Primary source
- 210
观点摘要 FDA, Health Canada, and UK MHRA jointly issued ten guiding principles in October 2021 covering multidisciplinary expertise, representative training data, and post-deployment monitoring for AI-enabled medical devices.
FDA, Health Canada, and UK MHRA — Joint medical-device regulators
Good Machine Learning Practice for Medical Device Development: Guiding Principles Research/report
- 211
观点摘要 World Health Organization set out ethical principles and governance recommendations for AI in health, including transparency, accountability, and human oversight.
World Health Organization — "Ethics and governance of artificial intelligence for health"
WHO Research/report
- 212
观点摘要 UN Panel of Experts on Libya described retreating forces being pursued and remotely engaged by drones or lethal autonomous weapon systems including Kargu-2 units, but did not establish in the cited passage that a Kargu-2 autonomously killed anyone.
UN Panel of Experts on Libya — Final report to the Security Council pursuant to resolution 1970 (2011), S/2021/229
UN Digital Library Research/report
- 213
观点摘要 Andrew Critch formalizes assistance games in which one AI system must respond to multiple human principals with differing preferences.
Andrew Critch — Co-author; UC Berkeley Center for Human-Compatible AI
"Multi-Principal Assistance Games: Definition and Collegial Mechanisms" (arXiv) Primary source
- 214
直接引述 “The sky is not falling, but it is slowly lowering”
David Autor — Co-chair, MIT Task Force on the Work of the Future
MIT News, on the release of "The Work of the Future" final report Secondary coverage
- 215
观点摘要 MIT Technology Review documented that Google Health's diabetic-retinopathy AI, which exceeded 90% sensitivity and specificity in lab conditions, rejected roughly 21% of the ~1,940 retinal images nurses submitted as 'ungradable' across 11 Thai clinics — often due to lighting conditions human graders would have accepted — disrupting clinic workflow.
MIT Technology Review — Contemporaneous report on Google Health's Thailand deployment
MIT Technology Review Secondary coverage
引用于 AI accelerating science and medicine: real breakthroughs, real limits
- 216
观点摘要 Victoria Krakovna documents around 60 examples, as of the 2020 article, of systems satisfying a literal objective while failing to achieve the designer’s intended outcome.
Victoria Krakovna — Research scientist, Google DeepMind; lead author
"Specification gaming: the flip side of AI ingenuity" (DeepMind blog) Primary source
- 217
观点摘要 Pontifical Academy for Life, Microsoft, IBM, UN FAO, and the Italian government commits signatories to AI built around human dignity, inclusion, transparency, impartiality, reliability, and security; Jewish and Muslim representatives signed alongside the Catholic Church in January 2023.
Pontifical Academy for Life, Microsoft, IBM, UN FAO, and the Italian government — Founding signatories, Rome Call for AI Ethics
romecall.org Signed statement
- 218
观点摘要 Toby Ord estimates overall existential risk this century at roughly one in six, attributing one in ten specifically to unaligned artificial intelligence — his largest single risk category.
Toby Ord — Author, The Precipice: Existential Risk and the Future of Humanity; then Future of Humanity Institute, Oxford
The Precipice (Bloomsbury/Hachette, 2020) Primary source
- 219
观点摘要 National Transportation Safety Board reported that an Uber self-driving test vehicle detected the pedestrian 5.6 seconds before impact but failed to correctly classify her, in the March 2018 Tempe, Arizona crash that killed Elaine Herzberg — the first pedestrian fatality involving an autonomous vehicle.
National Transportation Safety Board — Investigation into the fatal Uber automated-vehicle crash, reported by NPR
NPR Secondary coverage
- 220
观点摘要 Rabbi Daniel Nevins applies halakhic concepts of agency, liability, and damages to AI decision-making, concluding a human must remain accountable for outcomes even where AI materially assisted the decision.
Rabbi Daniel Nevins — Responsum approved by the Committee on Jewish Law and Standards, Conservative movement
"Artificial Intelligence, Moral Machines and Halakhah" (docket HM 182.1.2019) Primary source
- 221
观点摘要 John Drury, Holly Carter, Chris Cocking, Evangelos Ntontis, Selin Tekin Guven, and Richard Amlôt distilled disaster-response research into twelve recommendations for building collective psychosocial resilience, based on the social identity approach.
John Drury, Holly Carter, Chris Cocking, Evangelos Ntontis, Selin Tekin Guven, and Richard Amlôt — Authors, "Facilitating Collective Psychosocial Resilience in the Public in Emergencies"
Frontiers in Public Health Primary source
- 222
观点摘要 Paul Christiano presents a gradual-failure scenario involving many systems optimizing measurable proxies and developing influence-seeking tendencies.
Paul Christiano — AI alignment researcher; former OpenAI
"What Failure Looks Like" (AI Alignment Forum) Primary source
- 223
观点摘要 American Psychological Association found in its "Stress in America" survey that adults who constantly checked the news were measurably more stressed than those who did not.
American Psychological Association — "Stress in America: The State of Our Nation" survey
APA Research/report
- 224
观点摘要 Dylan Hadfield-Menell formalizes a simple game in which uncertainty about the human’s objective can give an agent an incentive to preserve its off switch.
Dylan Hadfield-Menell — Lead author, "The Off-Switch Game" (UC Berkeley)
arXiv Primary source
- 225
观点摘要 Dario Amodei identified reward hacking as one of several concrete accident risks in machine-learning systems.
Dario Amodei — Lead author (then Google Brain; now CEO, Anthropic)
"Concrete Problems in AI Safety" (arXiv) Primary source
- 226
直接引述 “challenges to substituting machines for workers in tasks requiring adaptability, common sense, and creativity remain immense”
David Autor — Economist, MIT; "Polanyi's Paradox and the Shape of Employment Growth"
National Bureau of Economic Research, Working Paper 20485 Primary source
- 227
观点摘要 Long Now Foundation describes its "Manual for Civilization" project collecting the books judged most essential to rebuilding a lost technological civilization.
Long Now Foundation — "Manual for Civilization" project
Long Now Foundation Primary source
- 228
观点摘要 Lewis Dartnell walks from agriculture through chemistry to power generation, explaining not just that low-technology processes work but why, for a reader whose supply chains are gone.
Lewis Dartnell — Author, "The Knowledge: How to Rebuild Our World from Scratch"
Penguin Random House Primary source
- 229
观点摘要 Roxane Cohen Silver repeated media exposure to the Boston Marathon bombings predicted higher acute stress than being there in person.
Roxane Cohen Silver — Professor of Psychological Science, UC Irvine; co-author, PNAS study on collective trauma
Proceedings of the National Academy of Sciences Primary source
引用于 Psychological preparedness: staying grounded under uncertainty
- 230
观点摘要 John Drury, Dermot Novelli, and Clifford Stott safety professionals substantially overestimate how often real mass panic occurs during emergencies.
John Drury, Dermot Novelli, and Clifford Stott — Authors, disaster-myths survey of emergency professionals
Journal of Applied Social Psychology Primary source
- 231
观点摘要 Carl Benedikt Frey and Michael A. Osborne estimated that about 47% of US employment was in occupations at high risk of computerisation over an unspecified future period, "perhaps a decade or two".
Carl Benedikt Frey and Michael A. Osborne — Economists, Oxford Martin School, University of Oxford
"The Future of Employment: How Susceptible Are Jobs to Computerisation?" Research/report
- 232
观点摘要 Nick Bostrom argues intelligence level and final goals are independent (the orthogonality thesis), and that goal-directed superintelligent agents would converge on sub-goals like self-preservation and resource acquisition regardless of terminal goal (instrumental convergence).
Nick Bostrom — Author, "The Superintelligent Will: Motivation and Instrumental Rationality in Advanced Artificial Agents"
Minds and Machines (preprint via nickbostrom.com) Primary source
- 233
观点摘要 United Nations Human Rights Council states that businesses have a responsibility to respect human rights and that victims of business-related abuse need access to remedy.
United Nations Human Rights Council — Guiding Principles on Business and Human Rights (endorsed by Resolution 17/4)
OHCHR Primary source
- 234
观点摘要 David Chalmers treats an intelligence explosion via recursively self-improving AI as a serious epistemic possibility, and asks whether resulting superintelligent minds would be conscious and whether uploaded humans would retain personal identity.
David Chalmers — Author, "The Singularity: A Philosophical Analysis"
Journal of Consciousness Studies, vol. 17, nos. 9-10 Primary source
- 235
观点摘要 David MacKay lays out the physical arithmetic of energy consumption and concentrated supply, arguing modern civilization depends on energy sources far denser than most renewable alternatives can easily replace.
David MacKay — Author, "Sustainable Energy — Without the Hot Air"
withouthotair.com Primary source
- 236
观点摘要 Alan Robock et al. modeled that firestorms from a full-scale United States–Russia nuclear exchange would loft soot into the stratosphere and drop global temperatures to ice-age levels for years.
Alan Robock et al. — Lead author, nuclear-winter climate-modeling study
arXiv Primary source
- 237
观点摘要 Edward L. Deci and Richard M. Ryan develops self-determination theory, arguing that autonomy, competence, and relatedness are basic psychological needs underlying human motivation.
Edward L. Deci and Richard M. Ryan — Authors, "The 'What' and 'Why' of Goal Pursuits: Human Needs and the Self-Determination of Behavior"
Psychological Inquiry Primary source
- 238
观点摘要 Federal Communications Commission lists voluntary, noncommercial public communication—particularly emergency communication—as one of several purposes of the amateur radio service.
Federal Communications Commission — 47 CFR § 97.1, Amateur Radio Service — Basis and Purpose
Code of Federal Regulations (via Cornell Law School Legal Information Institute) Primary source
- 239
观点摘要 U.S. Department of State recounts how the 1958–1961 US-Soviet-UK voluntary nuclear test moratorium collapsed when the Soviets resumed testing in 1961, salvaged afterward only as the narrower Limited Test Ban Treaty.
U.S. Department of State — Office of the Historian
history.state.gov Primary source
已知空白
各篇报告自身的参考文献,包括仅使用过一次的官方指南、法规、 研究论文与数据集。而这份记录列表刻意收窄了范围:它只追踪读 者可能在多个章节中反复遇到的常见来源与专家立场。只有在通过 修订记录中所 记载的完整内容与引用核查之后,报告才会被标记为已复核。