87 Matching Annotations
  1. Last 7 days
    1. models climb close to the average human baseline over the past year and a half.

      这个时间跨度(一年半)内AI系统接近人类平均水平的表现,显示了AI在基本常识推理方面的进步速度。这一数据点表明,虽然简单基准测试可能趋于饱和,但它们仍能揭示AI系统的局限性。

    1. The volume of open-world evaluations has increased dramatically in recent months.

      虽然文章没有提供具体的增长百分比,但'显著增加'的描述表明开放世界评估正在成为AI评估领域的新趋势。这种增长速度可能反映了业界对传统基准测试局限性的认识加深,以及AI能力发展到需要更复杂评估方法的阶段。

  2. May 2026
    1. Total AI computing capacity has been doubling approximately every seven months

      AI计算能力每7个月翻倍的增长率远超摩尔定律(约18-24个月翻倍),反映了AI领域对计算资源的极度渴求和产业投入的快速增长。这种指数级增长趋势是不可持续的,将面临物理极限、能源供应和制造成本等多重挑战,可能在未来几年内放缓。

    1. The rankings, set up by a Meta employee on its intranet using company data, measure how many tokens — the units of data processed by AI models — employees are burning through.

      这一观点揭示了‘tokenmaxxing’作为衡量员工AI使用能力的新趋势,暗示了数据消耗成为衡量生产力的一种方式。

  3. Apr 2026
    1. Pindrop reported a 475 percent year-over-year increase in synthetic voice attacks against insurance call centers across 2025.

      475%的年增长率表明语音合成攻击呈爆炸性增长。这一惊人的数字反映了AI语音技术的普及和攻击者利用这些技术的速度。保险公司成为主要目标是因为理赔主要通过电话处理,这使得语音验证成为关键安全环节。

    1. Three of the four metrics (ECI, log METR 50% time horizon, and a math-focused index we constructed from several math benchmarks) show strong evidence that progress has sped up relative to a global linear trend fit to data from 2023 onward.

      这个数据点表明75%的AI能力指标显示加速趋势,这是一个相当高的比例。文章提到这种加速始于2023年,与推理模型的出现时间吻合。这个比例值得注意,因为它表明AI进步可能正在经历一个质的转变,而非仅仅是量的累积。

    1. Meta is not alone in pursuing such a vision: Anthropic debuted tech capable of doing this [in 2024] and OpenAI last year announced [“Operator”] – a tool that can use a web browser on a human’s behalf.

      大多数人可能认为Meta在追求这种愿景方面是独一无二的,但作者指出Anthropic和OpenAI也在进行类似的研究,这表明这种趋势可能比人们想象的更普遍。

    1. I just hope the industry doesn't abandon the Model Context Protocol. The dream of seamless AI integration relies on standardized interfaces, not a fractured landscape of hacky CLIs.

      这是一个关于行业方向的深刻担忧。作者暗示了一个令人不安的趋势:行业可能过早放弃MCP这一标准化接口,转而采用碎片化的CLI方案。这不仅会导致用户体验下降,还可能阻碍AI与服务的无缝集成,影响整个生态系统的发展。

    1. The industry is currently witnessing a decisive shift toward more permissive, standardized licenses as developers increasingly prioritize ease of integration and legal certainty.

      令人惊讶的是:AI行业正经历向更宽松、标准化许可证的明显转变,这反映了开发者日益重视集成便利性和法律确定性。这一趋势表明,随着AI模型的成熟,许可证选择正成为与模型性能同等重要的因素,改变了AI开发的格局。

    1. Five hyperscalers now own over two-thirds of global AI compute, rising from 60% in Q1 2024.

      令人惊讶的是:这五大超大规模云服务提供商对全球AI计算资源的控制力在短短一年内从60%增长到67%,显示出AI计算资源正以前所未有的速度向少数科技巨头集中,这可能加剧AI发展的不平衡。

    1. The launch shows Meta is increasingly betting that efficiency, product integration, and distribution, not just model size, will define the next phase of competition in AI.

      这揭示了AI行业正在从单纯追求更大模型转向更注重实用性和集成度的重要转变。Meta的战略表明,未来AI竞争的关键可能不是模型规模,而是如何将AI无缝集成到现有产品中并提高效率。这种转变可能会重塑整个AI行业的发展方向和投资重点。

  4. Aug 2025
  5. Jun 2025
  6. Apr 2025
  7. Apr 2024
    1. if your treatments are ordered, don't compare each mean with each other mean (multiple comparisons), instead do one test for trend to ask if the outcome is linearly related with treatment number

      How do you do hypothesis testing for trends for an ordered categorical variable?

      Could you convert x to numbers (1,2,3) and run a linear regression y ~ x? or even categorical ordered variables can be linearly regressed?

  8. Feb 2024
    1. Michel Forst, UN-Berichterstatter zur Aarhus-Konvention, hat die europäischen Regierungen aufgefordert, Klima-Aktivist:innen zu unterstützen statt sie zu kriminalisieren. Die zunehmende Repression gefährde das Erreichen der Pariser Klimaziele und Demokratie und Menschenrechte in Europa. Forst erwartet, dass Protest und direkte Aktion zunehmen, weil die aktuelle Politik vieler europäischer Regierungen die wissenschaftlichen Erkenntnisse zu globaler Erhitzung, Biodiversitätsverlust und Umweltverschmutzung nicht respektiert. https://www.theguardian.com/environment/2024/feb/28/european-nations-must-end-repression-of-peaceful-climate-protest-says-un-expert

      Positionspapier von Michel Forst: https://unece.org/sites/default/files/2024-02/UNSR_EnvDefenders_Aarhus_Position_Paper_Civil_Disobedience_EN.pdf

  9. Oct 2022
    1. here are several ways I havefound useful to invite the sociological imagination:

      C. Wright Mills delineates a rough definition of "sociological imagination" which could be thought of as a framework within tools for thought: 1. Combinatorial creativity<br /> 2. Diffuse thinking, flâneur<br /> 3. Changing perspective (how would x see this?) Writing dialogues is a useful method to accomplish this. (He doesn't state it, but acting as a devil's advocate is a useful technique here as well.)<br /> 4. Collecting and lay out all the multiple viewpoints and arguments on a topic. (This might presume the method of devil's advocate I mentioned above 😀)<br /> 5. Play and exploration with words and terms<br /> 6. Watching levels of generality and breaking things down into smaller constituent parts or building blocks. (This also might benefit of abstracting ideas from one space to another.)<br /> 7. Categorization or casting ideas into types 8. Cross-tabulating and creation of charts, tables, and diagrams or other visualizations 9. Comparative cases and examples - finding examples of an idea in other contexts and time settings for comparison and contrast 10. Extreme types and opposites (or polar types) - coming up with the most extreme examples of comparative cases or opposites of one's idea. (cross reference: Compass Points https://hypothes.is/a/Di4hzvftEeyY9EOsxaOg7w and thinking routines). This includes creating dimensions of study on an object - what axes define it? What indices can one find data or statistics on? 11. Create historical depth - examples may be limited in number, so what might exist in the historical record to provide depth.

  10. May 2022
  11. Apr 2022
    1. Katherine Ognyanova. (2022, February 15). Americans who believe COVID vaccine misinformation tend to be more vaccine-resistant. They are also more likely to distrust the government, media, science, and medicine. That pattern is reversed with regard to trust in Fox News and Donald Trump. Https://osf.io/9ua2x/ (5/7) https://t.co/f6jTRWhmdF [Tweet]. @Ognyanova. https://twitter.com/Ognyanova/status/1493596109926768645

    1. Dr Duncan Robertson [@Dr_D_Robertson]. (2021, October 29). ONS Covid survey. 2% of the population +ve. “The percentage of people testing positive for COVID-19 increased for all age groups, except for those in school Year 12 to those aged 34 years, where the trend was uncertain in the week ending 22 October 2021” https://ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/29october2021 https://t.co/1n9KVq6wDT [Tweet]. Twitter. https://twitter.com/Dr_D_Robertson/status/1454050450106376192

  12. Feb 2022
    1. Einsatz hochskalierbarer Graphen-Datenbank-Technologien, die durch die Integration von semantischen Middleware-Komponenten, Visualisierungswerkzeugen und Editoren auch von Nicht-Technikern und Fachexperten bedient werden können,

      Trend

  13. Jan 2022
    1. Elliott, P., Eales, O., Bodinier, B., Tang, D., Wang, H., Jonnerby, J., Haw, D., Elliott, J., Whitaker, M., Walters, C., Atchison, C., Diggle, P., Page, A., Trotter, A., Ashby, D., Barclay, W., Taylor, G., Ward, H., Darzi, A., … Donnelly, C. (2022). Post-peak dynamics of a national Omicron SARS-CoV-2 epidemic during January 2022 [Working Paper]. http://spiral.imperial.ac.uk/handle/10044/1/93887

  14. Dec 2021
  15. Nov 2021
  16. Oct 2021
  17. Sep 2021
    1. Derek Thompson. (2021, August 25). Adult hospitalizations since July 1 vs. Vaccinations, by state: 1) The relationship between more vaccines and less hospitalization is pretty straightforward. 2) Holy moly, Florida. Among states with more than one shot per person, FL really is on its own island of pain. Https://t.co/tuTAdUT0OM [Tweet]. @DKThomp. https://twitter.com/DKThomp/status/1430643278337163267

  18. Aug 2021
  19. Jul 2021
  20. Jun 2021
  21. May 2021
    1. Prof. Christina Pagel. (2021, April 15). THREAD on VACCINATION & EQUITY in ENGLAND: I know I’ve tweeted about this before, but now we can look at how gaps by deprivation and ethnicity change with age groups and what that might mean... TLDR: widening gaps but access and communication will be key I suspect 1/5 [Tweet]. @chrischirp. https://twitter.com/chrischirp/status/1382725119773134848

  22. Mar 2021
    1. ow we might accelerate the path to the Metaverse by a focus on desktop access by remote workers: Could it be that the Metaverse starts with people working together in virtual offices, and then staying around and connecting for various reasons outside of work? 

      this is actually a fascinating theory and i betcha this is actually how the metaverse starts to take hold... people looking for better alternatives than zoom for virtual events, conferences, work meetings, and birthday parties online

      this will merge with video games to create the metaverse

      its hilarious to think of it this way

    1. Katz points out this is an “extreme example to prove a point.” YellowHeart wants to show people how much control can be put into the ticket with smart contracts. Going forward, he says this same tech can be used for general tickets, which could be a huge advancement in the secondary market. Every time an NFT is resold, a percentage of money earned could go to the artist — or whoever is included in the contract, perhaps even a charity. (In such instances, YellowHeart can also set a maximum price that the NFT can be resold at, eradicating scalpers.)

      Dit is volgens mij een killer feature van NFT's in muziek. De rest is leuk maar dit is superinteressant

    1. The COVID Tracking Project. (2020, November 11). Our daily update is published. States reported 1.2 million tests and 131k cases, the highest single-day total since the pandemic started. There are 62k people currently hospitalized with COVID-19. The death toll was 1,347. Https://t.co/WPoX9Nj7ef [Tweet]. @COVID19Tracking. https://twitter.com/COVID19Tracking/status/1326321342933831680

  23. Feb 2021
    1. Miro Weinberger. (2020, December 3). Our 1st Covid-19 wastewater tests since Thanksgiving just came in—Virus levels are up significantly citywide. I hope that all of #BTV will look at this graph and see what I see: A call to action, to stop gathering with other households, and to get tested ASAP if you have https://t.co/8nxTwOOcFA [Tweet]. @MiroBTV. https://twitter.com/MiroBTV/status/1334613511692017664

  24. Nov 2020
  25. Oct 2020
  26. Sep 2020
  27. Aug 2020
  28. Jul 2020
  29. Jun 2020
  30. May 2020
  31. Apr 2020
  32. Mar 2020
  33. Apr 2019
  34. Aug 2018
  35. Jun 2018
    1. Those who succeed the most and establish successful platforms “on top” of the open standardlater tend to consolidate the industry by leveraging their scale (in assets and distribution) tointegrate vertically and expand horizontally at the expense of smaller companies. Competing inthis new environment suddenly becomes expensive and startups struggle to create value in theshadow of incumbents, compressing venture returns.Demand then builds for a low cost, open source alternative to the incumbent platforms, and thecycle repeats itself: the new open standard emerges and gets adopted, the market decentralizes asnew firms leverage the cost savings to compete with the old on price, value creation shiftsupwards (once more), and so on
    2. Information technology evolves in multi-decade cycles of expansion, consolidation anddecentralization. Periods of expansion follow the introduction of a new open platform thatreduces the production costs of technology as it becomes a shared standard. As production costsfall, new firms come to market leveraging the standard to compete with established incumbents,pushing down prices and margins, and decentralizing existing market powers.The price drop attracts new users, increasing the overall size of the market and creating newopportunities for mass consumer applications. Entrepreneurial talent moves to serve the newmarkets where costs are low, competition is scarce, and the upside is high. Often these earlyentrepreneurs will introduce new kinds of business models, orthogonal to existing ones
    1. Recent studies have indicated that Uber’s U.S. driver churn has sharply increased this year, to rates as high as 96%. Needless to say, it’s hard (and costly) to maintain double-digit growth rates, when only 4% of mission critical, de facto employees stay on the job for more than a year.
    2. In historical context, Uber’s extraordinary losses are thus not just a case of growing pains of an ambitious Silicon Valley startup, but a reflection of the deep structural deficiencies in ride-hail industry economics. Prior to artificial regulatory supply caps, the unregulated taxi industry was unprofitable and subject to growing concerns over negative externalities. Uber is now facing the same relentless drag on its P&L.
    1. An underlying theme in much of the work in the field is that existing government regulation of copyright, security, and antitrust is inappropriate in the modern world. For example, information goods, such as news articles and movies, now have zero marginal costs of production and sharing. This has made the redistribution without permission common and has increased competition between providers of information goods.
  36. Sep 2015