Async agents are moving into everyday work. For an agent to be trustworthy and useful inside an organization, it needs real enterprise data: CRM records, repositories, inboxes, knowledge bases.
大多数人认为AI助手应该先在受限环境中测试，然后再逐步接入企业敏感数据，但作者认为AI助手应该直接接入企业真实数据才能变得可信和有用，这挑战了传统AI安全部署的渐进式方法。
120B-A12B may be a bit too large for local inference on regular consumer hardware
大多数人认为更大的模型参数量总是带来更好的性能，但作者暗示过度扩展模型规模可能不适合实际应用。这一务实观点挑战了'越大越好'的行业共识，强调了实际部署中的硬件限制。
Everybody wants to be the first to do something and just push things out without careful scrutiny and red-teaming.
大多数人认为企业安全漏洞是技术能力不足的结果，但作者认为这更多是企业文化和管理决策的问题。这个观点挑战了将安全失败简单归因于技术缺陷的主流叙事，指出企业追求'第一'而非'安全'的文化才是根本原因。
Security and utility always have a trade-off
大多数人认为AI安全可以通过技术手段完美解决，但作者认为安全与实用性之间存在根本性权衡。这个观点挑战了技术乐观主义，指出公司在追求AI能力的同时必然会牺牲某些安全措施，暗示AI安全问题的解决不仅仅是技术问题，更是商业决策问题。
Security and utility always have a trade-off
大多数人认为AI安全可以通过技术手段完美解决，但作者指出安全与实用性之间存在根本性权衡。这一观点挑战了行业对'绝对安全'的追求，暗示公司可能为了功能性和竞争力而故意接受某些安全风险，这与安全至上的行业共识相悖。
85–90% of customers using the AI Assistant now completing their claim filing through AI
【令人震惊的企业落地数字】Travelers 保险公司全国部署 AI 报案助手，85-90% 的客户通过 AI 完成完整报案流程——这不是「试点」，而是全国规模的生产部署。更惊人的背景：该系统在 8 个州上线后仅 2 个月就扩展至全国。去年 Travelers 处理了 150 万件索赔、赔付超 $230 亿——这意味着数百万真实事故受害者的第一个「对话对象」已经是 AI。
To get started with the cuOpt verified skill, for example, follow these steps: 1. Pull the cuOpt verified skill from the catalog: git clone github.com/nvidia/skills && cd skills/skills/cuopt 2. Verify the signature: model_signing verify certificate. --signature skill.oms.sig --certificate-chain nv-agent-root-cert.pem --ignore-unsigned-files 3. Open SKILLCARD.yaml to see ownership, dependencies, license, and verification status.
行动建议：按照文中提供的具体步骤，克隆并验证NVIDIA的cuOpt技能，查看技能卡片以了解所有权、依赖关系、许可证和验证状态。这种实践可以确保您使用的技能是经过验证的，并且可以安全地集成到您的AI代理工作流中。
PwC will roll out Claude Code and Cowork starting with U.S. teams and expanding toward a global workforce of hundreds of thousands of professionals
PwC计划将其全球数十万专业人员的 workforce 纳入Claude的使用范围。这是一个大规模部署计划，表明了企业级AI应用的规模化趋势。'数十万'是一个模糊的表述，缺乏精确数字，但足以显示合作规模之大。
pluralism is most decisively made or unmade at the deployment-governance layer: interfaces, preference-data pipelines, and audit infrastructure.
This argument shifts the locus of the problem from the model's architecture to the socio-technical systems that surround it. It's a provocative claim that the core issue isn't 'how to build a better model' but 'how to build a better system for deploying and governing models,' placing the onus on developers and regulators, not just AI researchers.
When inference is expensive, teams limit usage, reduce context, or avoid certain applications altogether.
文章指出推理成本高昂会导致团队限制使用、减少上下文或避免某些应用。这个数据点虽然没有具体数字，但反映了当前AI部署的经济瓶颈，是SubQ试图解决的核心问题之一。
help large enterprises deploy AI responsibly across their core business operations
【令人震惊】「负责任地在核心业务流程部署 AI」——这句话意味着 Anthropic 正在承接以前由麦肯锡、埃森哲做的企业变革咨询工作。纯模型 API 商业模式的顶峰可能已过：Claude 的护城河从「技术优势」升级为「有金融资本背书的企业实施能力」，中间层 AI 集成商和咨询公司的生存空间被直接压缩。
Without any extra setup, agents have everything they need to deploy a new production application in one shot.
最佳实践建议是简化部署过程，避免手动步骤，使自动化部署更加高效。
Coding agents are great at building software. But to deploy to production they need three things from the cloud they want to host their app —an account, a way to pay, and an API token.
This highlights a common pitfall for beginners: understanding the infrastructure requirements for deploying software, especially the need for accounts and payment methods.
Coding agents are great at building software. But to deploy to production they need three things from the cloud they want to host their app —an account, a way to pay, and an API token.
初学者常见陷阱：错误地认为部署到生产环境只需要代码，而忽略了账户、支付和API令牌等必要条件。
The company also says the Pentagon has the opportunity to test models before deployment
可能带有偏见的表述：Anthropic 声称五角大楼有机会在部署前测试模型，这种表述可能暗示了 Anthropic 对五角大楼决策过程的看法。
This dynamic forces a brutal new discipline in how enterprises deploy capital and architect their internal workflows.
这种动态迫使企业以全新的方式部署资本和架构内部工作流程，表明了人工智能对企业管理方式的深远影响。
over one million Trainium2 chips to train and serve Claude
使用超过100万颗Trainium2芯片的数据，展示了Anthropic在AI硬件部署上的巨大规模。这一数字不仅反映了计算能力的投入，也显示了与AWS在芯片定制上的深度合作。对于AI模型训练而言，百万级芯片的部署规模是行业顶尖水平，表明Claude可能需要大量计算资源进行训练和推理。
over one million Trainium2 chips to train and serve Claude
使用超过100万个Trainium2芯片，这是一个惊人的硬件部署规模。这一数字不仅显示了Anthropic与Amazon的深度合作，也反映了训练和运行大型语言模型所需的庞大计算资源。相比其他AI公司，这种规模的芯片部署表明Anthropic正在全力投入AI基础设施。
Claude is now being deployed to NEC Group employees around the world
大多数人认为企业会谨慎地小规模试点AI工具，但作者认为NEC正在全球范围内大规模部署Claude，这表明企业对AI技术的信任度远高于预期，挑战了传统的技术采用曲线和变革管理理论。
the surrogate is activated only when its agreement with the LLM exceeds a user-specified threshold α
大多数人认为模型部署应该是全有或全无的，要么完全替代原模型要么完全不使用。但作者提出了一种'部分激活'的激进方法，只在代理模型与原模型达到特定一致性阈值时才使用代理，这种细粒度的控制方式打破了传统的二元部署思维。
Infrastructure Provisioning cd deploy/terraform/aliyun terraform init terraform plan terraform apply Helm Deployment cd deploy/helm helm install aegis-core ./aegis-core \ --namespace aegis \ --create-namespace \ --set image.repository=<acr-registry>/aegis-core \ --set image.tag=lat
使用Terraform和Helm进行云基础设施部署体现了现代DevOps实践在AI安全平台中的应用。这种基础设施即代码(IaC)方法确保了部署的可重复性和一致性，同时支持阿里云等特定云平台，显示了平台对生产环境的适应性。
Quick Start # Clone the repository git clone https://github.com/fxp/aegis-core.git cd aegis-core # Start all services with Docker Compose docker-compose up -d # The API is available at http://localhost:8000 # Health check: http://localhost:8000/health
简化的启动流程展示了容器化部署的优势，使用Docker Compose一键启动所有服务，大大降低了部署复杂度。这种设计反映了现代AI平台开发的一个重要趋势：简化环境配置，使研究人员能够快速开始工作，而不是陷入环境设置的困境。
Because small, cheap, fast models are sufficient for much of the detection work, you don't need to judiciously deploy one expensive model and hope it looks in the right places. You can deploy cheap models broadly, scanning everything, and compensate for lower per-token intelligence with sheer coverage and lower cost-per-token.
这一观点提出了AI安全的经济新模式，通过广泛部署小型廉价模型来弥补单一大模型的不足。这种'广撒网'策略可能比依赖少数昂贵模型更有效，尤其在大规模代码库扫描场景中，为AI安全的经济可行性提供了新思路。
In practice, deployed model implementations are often flexible (e.g., mixing kernel variants, hybrid attention patterns, MoE blocks, and serving-optimized layouts), which can deviate from the assumptions required by a given conversion recipe.
这个观点揭示了现有方法在实际部署中的一个重要局限性：它们通常依赖于特定的模型实现假设，而实际部署的模型往往更加灵活和复杂。这强调了Attention Editing框架的优势——它不依赖于精细的结构要求，可以适应各种实际部署场景，为模型转换提供了更大的灵活性。
单张 24GB 4090 直接部署 32B LLM
令人惊讶的是：一张消费级显卡竟然能直接运行320亿参数的大模型，这打破了人们对大模型硬件需求的固有认知。过去需要多张高端显卡或专业服务器才能运行的模型，现在单张RTX 4090就能实现，这标志着大模型普及的门槛大幅降低。
A deployment cascade combining both stages attains 90% accuracy at 71% coverage without any task-specific labels.
令人惊讶的是：SELFDOUBT方法通过两级部署策略，在没有任务特定标签的情况下实现了90%的准确率和71%的覆盖率。这一成果表明，通过简单分析模型输出中的犹豫和验证行为，可以构建出高效的置信度过滤器，大幅提升模型在实际应用中的可靠性，无需额外标注数据。
Unlike methods that require multiple sampled traces or model internals, SELFDOUBT operates on a single observed reasoning trajectory, making it suitable for latency- and cost-constrained deployment over any proprietary API.
令人惊讶的是：SELFDOUBT方法仅需单个推理轨迹就能进行不确定性量化，而传统方法通常需要多次采样或访问模型内部参数。这一突破使得该方法可以在延迟和成本受限的部署环境中使用，特别适用于无法获取模型内部信息的专有API，大大降低了实际应用门槛。
Anthropic says Managed Agents is designed to cut the time it takes to move from prototype to production from months to days, with early adopters like Notion, Rakuten, Asana, Vibecode, and Sentry already using it across coding, productivity, and internal workflow automation.
将AI原型到生产的时间从几个月缩短到几天是一个惊人的加速，这将彻底改变企业采用AI的方式。这种快速部署能力可能加速AI在各行业的普及，但也带来了关于AI系统安全性和治理的紧迫问题，企业需要在快速采用和确保安全之间找到平衡。
We do not plan to make Claude Mythos Preview generally available, but our eventual goal is to enable our users to safely deploy Mythos-class models at scale.
大多数人认为强大的AI模型应该广泛普及以造福更多人。但作者明确表示不会公开发布这个最强大的模型，暗示了AI能力扩散可能带来的风险大于收益，这与技术民主化的主流观点相悖。
谷歌在沉寂了很长时间以后，终于发了一个不错的模型，而且还是开源的 Gamma 4 系列。专门用来在本地设备（比如手机、电脑）上跑
大多数人认为谷歌作为 AI 领域的领导者会持续专注于云端大模型，但其突然转向端侧开源模型的做法令人意外。这种战略转变表明谷歌可能重新评估了 AI 部署的未来方向，从集中式向分布式转变，挑战了'更大模型更好'的行业共识，暗示了端侧 AI 可能成为下一个技术热点。
Using vLLM high-throughput LLM serving on DGX Spark provides a high-performance platform for the largest Gemma 4 models
大多数人认为运行最大的Gemma 4模型需要专门的硬件和复杂的部署流程。但作者声称vLLM可以在DGX Spark上高效运行这些大型模型，暗示推理优化技术可能已经达到了一个临界点，使得复杂模型部署变得更加简单和高效。
The 31B and 26B A4B variants are high-performing reasoning models suitable for both local and data center environments.
大多数人认为大型语言模型(31B参数)只能在数据中心环境中运行，但作者声称这些模型可以在本地环境中高效运行。这一观点与行业共识相悖，暗示边缘计算能力可能比我们想象的更强大，可能会改变AI部署的格局。
Absolute size thresholds for degenerative aneurysms: [1][3][5]
Ascending aorta/aortic root: ≥5.5 cm
Descending thoracic aorta: ≥6.0 cm (or 5.5 cm if favorable anatomy for TEVAR)
Thoracoabdominal aorta: ≥6.0 cm
Lower thresholds apply for: [3]
Marfan syndrome or genetic conditions: 4.0-5.0 cm depending on condition
Bicuspid aortic valve: 5.0-5.5 cm
Rapid growth: >0.5 cm/year
Concomitant cardiac surgery: >4.5 cm if undergoing aortic valve surgery
Immediate surgical evaluation: [5]
Any symptomatic aneurysm regardless of size (chest/back pain, dysphagia, hoarseness, hemoptysis)
Acute complications (dissection, rupture, malperfusion)
Post-Repair Surveillance
After TEVAR: CT at 1 month, 12 months, then annually if stable. [1]
After open repair: CT or MRI within 1 year, then every 5 years if no residual aortopathy. Annual imaging if residual disease or abnormal findings.
The potential for cuts in 2030 is 31 gigatons of CO2 equivalent – which isaround 52 per cent of global greenhouse gas emissions in 2023 – and 41gigatons in 2035.· Increased deployment of solar photovoltaic technologies and wind energy coulddeliver 27 per cent of this total emission reduction potential in 2030 and 38 percent in 2035.· Action on forests could deliver around 20 per cent of the potential in both years.• Other strong options include efficiency measures, electrification and fuelswitching in the buildings, transport and industry sectors.
for - stats - 27% of the gap can be reduced by wind and solar deployment and 20% by action on forests, while efficiency, electrification, fuel switching in buildings, transport and industry sectors can also contribute - UN Emissions Gap Report 2024 - Key Messages
A TRUSTworthy repository needs to focus on serving its target user community. Each user community likely has differing expectations from their community repositories, depending in part on the community’s maturity regarding data management and sharing. A TRUSTworthy repository is embedded in its target user community’s data practices, and so can respond to evolving community requirements
TRSP Desirable Characteristics
TRUSTworthy repositories take responsibility for the stewardship of their data holdings and for serving their user community.
TRSP Desirable Characteristics
TRSP Desirable Characteristics Data governance should take into account the potential future use and future harm based on ethical frameworks grounded in the values and principles of the relevant Indigenous community. Metadata should acknowledge the provenance and purpose and any limitations or obligations in secondary use inclusive of issues of consent.
TRSP Desirable Characteristics Ethical processes address imbalances in power, resources, and how these affect the expression of Indigenous rights and human rights. Ethical processes must include representation from relevant Indigenous communities
TRSP Desirable Characteristics Ethical data are data that do not stigmatise or portray Indigenous Peoples, cultures, or knowledges in terms of deficit. Ethical data are collected and used in ways that align with Indigenous ethical frameworks and with rights affirmed in UNDRIP. Assessing ethical benefits and harms should be done from the perspective of the Indigenous Peoples, nations, or communities to whom the data relate
TRSP Desirable Characteristics Resources must be provided to generate data grounded in the languages, worldviews, and lived experiences (including values and principles) of Indigenous Peoples.
TRSP Desirable Characteristics Use of Indigenous data invokes a reciprocal responsibility to enhance data literacy within Indigenous communities and to support the development of an Indigenous data workforce and digital infrastructure to enable the creation, collection, management, security, governance, and application of data
TRSP Desirable Characteristics Indigenous Peoples have the right to develop cultural governance protocols for Indigenous data and be active leaders in the stewardship of, and access to, Indigenous data especially in the context of Indigenous Knowledge
TRSP Desirable Characteristics Indigenous Peoples have the right to data that are relevant to their world views and empower self-determination and effective self-governance. Indigenous data must be made available and accessible to Indigenous nations and communities in order to support Indigenous governance.
TRSP Desirable Characteristics Indigenous Peoples have rights and interests in both Indigenous Knowledge and Indigenous data. Indigenous Peoples have collective and individual rights to free, prior, and informed consent in the collection and use of such data, including the development of data policies and protocols for collection.
TRSP Desirable Characteristics Indigenous data are grounded in community values, which extend to society at large. Any value created from Indigenous data should benefit Indigenous communities in an equitable manner and contribute to Indigenous aspirations for wellbeing.
TRSP Desirable Characteristics Data enrich the planning, implementation, and evaluation processes that support the service and policy needs of Indigenous communities. Data also enable better engagement between citizens, institutions, and governments to improve decision-making. Ethical use of open data has the capacity to improve transparency and decision-making by providing Indigenous nations and communities with a better understanding of their peoples, territories, and resources. It similarly can provide greater insight into third-party policies and programs affecting Indigenous Peoples.
TRSP Desirable Characteristics Governments and institutions must actively support the use and reuse of data by Indigenous nations and communities by facilitating the establishment of the foundations for Indigenous innovation, value generation, and the promotion of local self-determined development processes
TRSP Desirable Characteristics Indigenous Peoples’ rights and wellbeing should be the primary concern.
TRSP Desirable Characteristics Data ecosystems shall be designed and function in ways that enable Indigenous Peoples to derive benefit from the data.
TRSP Desirable Characteristics Indigenous Peoples’ rights and interests in Indigenous data must be recognised and their authority to control such data be empowered. Indigenous data governance enables Indigenous Peoples and governing bodies to determine how Indigenous Peoples, as well as Indigenous lands, territories, resources, knowledges and geographical indicators, are represented and identified within data.
TRSP Desirable Characteristics Those working with Indigenous data have a responsibility to share how those data are used to support Indigenous Peoples’ self determination and collective benefit. Accountability requires meaningful and openly available evidence of these efforts and the benefits accruing to Indigenous Peoples.
No mention of how to authenticate the deployment, which is disappointing. See the repository for more.
Generalist repositorie
it’s important to understand some of the fundamental concepts around what a “container” is and how it compares to a Virtual Machine (VM)
Odoo standard (no cutomizations)
Odoo is a multi-tenant system: a single Odoo system may run and serve a number of database instances. It is also highly customizable, with customizations (starting from the modules being loaded) depending on the "current database"
one should not upgrade a production environment without extensive testing. I prefer to not upgrade prod at all. Instead, I create a new instance with latest everything, host my apps there, test everything out, and then make it production.
Deployment: Significance of Branching for Continuous Delivery
The deployment of software in DevOps is based on Continuous Delivery. Continuous Delivery enables all kinds of changes, including new features, configuration changes, bug fixes and experiments, to be put into production safely and quickly in a sustainable manner. A Branching strategy and Base Truncs play an important role in this.
Model deployment in Azure ML
On the technical front, the number of components affected is much smaller; on the business front, it's usually a much easier conversation to persuade the team to roll back one small feature than twenty big features
Optimising covid-19 vaccine uptake—Time for a more nuanced approach. (2021, April 30). The BMJ. https://blogs.bmj.com/bmj/2021/04/30/optimising-covid-19-vaccine-uptake-time-for-a-more-nuanced-approach/
Guidance on developing a national deployment and vaccination plan for COVID-19 vaccines. (n.d.). Retrieved January 21, 2021, from https://www.who.int/publications-detail-redirect/WHO-2019-nCoV-Vaccine_deployment-2020.1
Europe, W. H. O. R. O. for. (2020). Strategic considerations in preparing for deployment of COVID-19 vaccine and vaccination in the WHO European Region, 9 October 2020. https://apps.who.int/iris/handle/10665/335940
This is a major development—given the combined efforts of Google and the mobile networks, RCS will be the fastest deployed messaging technology of all time.
Put a load balancer or another nginx on the host for providing HTTPS.
darklang.com
Record a deployment with POST
Newrelic deployment markers can also be generated via a post request although it's generally simpler to use the agent specific method.
config_file: Path to the config file name
Since Hypothesis uses a server side config I'm not sure what to do here as there is no config file to provide. 🤔
If it turns out this isn't feasible (although I'm sure there's a solution for this config_file issue) the alternative approach would be to use the post method.
newrelic-admin record-deploy config_file description [revision changelog user]
To enable recording of deploys on the python agent via New Relic, you can simply call the newrelic-admin record-deploy command and pass it the necessary revision information. This will place a deployment marker on any graph you view in newrelic as a vertical line-indicating that a new revision of the code was released at that point in time.
Repositorio NPM privado grátis com Verdaccio e AWS
Excelente para você entender, na prática, sobre Cloud Deployment (um de nossos importantes subtópicos!). Além disso, vai sair da palestra com mais ferramentas para seu cinto de utilidades!
You can push an alternative branch to Heroku using Git. git push heroku-dev test:master This pushes your local test branch to the remote's master branch (on Heroku).
Push a local non-master branch to heroku master