AllyshipContent Radar

Deep dives 738

Paper và bài viết kỹ thuật

Crawl live-1795068900000 · 13:22 19 thg 11, 2026 · Xuất bản 13:30 19 thg 11, 2026

Kiến trúc

Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots

Câu hỏi Hosted by Sonya Huang, Sequoia Capital 0:00 – Introduction 1:46 – What Mecka is: pixels, actions and logic 2:45 – Is robot data economically feasible without the internet?

Robotics never got the internet's gift of free training data, so someone has to manufacture it. Josh Gao, founder and CEO of Mecka AI, runs one of the largest robot data factories, and he explains what that actually involves: custom baseball caps and tactile gloves because nothing off the shelf captures force, 100-gigabit upload lines to move hundreds of petabytes, and custom quality gates. Josh lays out the training recipe the industry is converging on (roughly a million hours of egocentric human data, tens of thousands of hours of robot data, and a few hours of in-domain teleop ) and why tactile and simulation data still have to enter the "data pot of soup." He argues the field is still orders of magnitude behind Tesla's FSD, which spent a decade on two degrees of freedom while a robot hand has dozens. His take: there won't be a single GPT-3 moment for robotics, but a steady unlock of commercial tasks, starting with constrained industrial work and ending, he hopes, with robots mining asteroids. Hosted by Sonya Huang, Sequoia Capital 0:00 – Introduction 1:46 – What Mecka is: pixels, actions and logic 2:45 – Is robot data economically feasible without the internet? 4:15 – Teaching robots about carbon life 5:27 – Cloud kitchens, stablecoins and the path to robotics 7:03 – Entering robotics with a fresh lens 8:45 – What egocentric data is and why Mecka bet on it 10:09 – The first pilot: aunties and uncles with a mobile app 10:42 – Force data and tactile gloves 11:55 – Inside the operations of a data factory 14:41 – Heat-mapping the taxonomy of human work 15:42 – Quality gates, fraud and the McDonald’s POV videos 17:44 – Earning a margin as a data intermediary 19:33 – Millions of hours vs. Tesla’s decade 20:55 – Why it isn’t binary: unlocking commercial tasks along the way 22:02 – Home data is plentiful; home robots are the hardest 23:24 – Useful robots can be deployed today 24:35 – When robotics gets an FSD-style flywheel 26:42 – Deployment diffusion and the modern systems integrator 28:21 – Favorite vertical robot categories; mining an asteroid 30:08 – The hardware slate: Waymo-fying people 30:55 – EgoVerse 32:04 – The data pot of soup: the converged training mix 33:54 – Move 37 for robotics: AlphaGo and Federer 36:02 – World models and the training mix 36:58 – Will scaling the current mix get us to the promised land? 39:05 – Will the AGI labs steamroll robotics? 40:50 – Sci-fi roots and the Kawasaki Corleo 42:44 – How many robots in a decade?

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

Agent SDK

The Agent SDK is available today. It allows you to define an agent as a folder of code, test it with evals, and deploy it to Cursor or your own infrastructure. Run `npx @cursor/agent-sdk init ./my-bot`, then `agent-sdk dev` to start the bot locally with a playground. Pass `--template` to start from a security reviewer, an owners-policy approver, a PR autofixer, or a Linear or Jira triager. Deploy agents wherever you need them When you're ready to put an agent into use, you can deploy it to Cursor by running `agent-sdk deploy` from the project directory. The command builds the agent, gives it a URL, and starts it on our infrastructure. You can manage secrets and integrations from the dashboard, while continuing to use the same logs, evals, and playground available locally. Agents as code An agent is a project. `instructions.md` is the prompt. Each file in `tools/` is one typed action the model can call. `skills/` holds procedures the model loads on demand, `mcp-connections/` adds tools from MCP servers, and `channels/` defines where the bot receives and answers messages. `agent-sdk validate` checks the project and `agent-sdk info` prints what was discovered. Channels Agents wake on pull request, comment, and CI events from GitHub, GitLab, and Bitbucket; on mentions and DMs in Slack; on webhooks from any provider; and on a cron schedule. GitHub events pull through your Cursor account, so no public webhook URL is required. Slack is well-supported out of the box. Each conversation keeps its own session and a complete record of every message and tool call. Follow-ups on a pull request or Slack thread resume the same session. Evals An eval sends a fixed message to the bot and asserts on what it did: which tools ran, with what input, and the shape of the reply. Run `agent-sdk eval` locally or in CI. `agent-sdk github replay <pr-url>` turns a real pull request into signed webhook payloads and sends them to the bot, so GitHub-driven bots can be tested without waiting for a push. Add `--dry-run --fixtures-dir fixtures/` to save the payloads as eval fixtures. Get started The Agent SDK is generally available today. Read the docs and get started.

Nguyên văn nguồn
Chi tiết nguồn
  • Agent SDK Cursor Changelog · 20 thg 10, 2026
Kỹ thuật

Introducing vLLM TorchTPU

vllm-torchtpu is a new vLLM backend for Google Cloud TPUs built on TorchTPU, with Pallas kernels, multi-host parallelism, KV cache offloading, and disaggregated serving.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

How Claude designed protein binders against scorpion neurotoxins

We asked an internal research Claude model to design minibinders (small proteins that attach tightly to a target) against scorpion neurotoxins.

Anthropic Research (unofficial RSS mirror) · Chưa ghi tác giả Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Patch release v5.20.2

<h1>Patch release: v5.20.2</h1> <p>This patch notably solves an issue where some inputs could loose their original type inside <code>prepare_inputs_for_generation</code> in <code>generate</code>, leading to problems with multimodal models. It also fixes some other issues listed on the below commits.</p> <p>It contains the following commits:</p> <ul> <li>Fix inputs dict casting in generate (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5922118541" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49731" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49731/hovercard" href="https://github.com/huggingface/transformers/pull/49731">#49731</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/Cyrilvallez/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/Cyrilvallez">@Cyrilvallez</a></li> <li>Fix zero expert gradients in bfloat16 with expert parallelism and ep_size &lt; fsdp_size (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5894228582" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49688" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49688/hovercard" href="https://github.com/huggingface/transformers/pull/49688">#49688</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/qgallouedec/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/qgallouedec">@qgallouedec</a></li> <li>Declare the input modalities of the VLMs that report text only (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5887394024" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49677" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49677/hovercard" href="https://github.com/huggingface/transformers/pull/49677">#49677</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/qgallouedec/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/qgallouedec">@qgallouedec</a></li> <li>Avoid a host-to-device copy in eager_mask (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5892857440" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49684" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49684/hovercard" href="https://github.com/huggingface/transformers/pull/49684">#49684</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/hmellor/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/hmellor">@hmellor</a></li> </ul>

Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

ClusterMAX 3.0: The Industry Standard GPU Cloud Rating System Returns

This post has bonus content for paid subscribers. Upgrade to get full access. In 8 months since our last major release of ClusterMAX, slavering investors have just about run out of pockets to stuff checks into. GPU supply has gone to zero. Meanwhile, we have been hard at work putting clusters through the ringer. Weeks ago, we teased this report with some R-rated anecdotes from our experiences probing the security practices of neoclouds, eliciting a PSA from a neocloud customer that you may have heard of. Today, we finally go through the full breadth and depth of our testing, more thorough in ClusterMAX 3.0 than ever before. This includes compute, networking, storage, orchestration, UI, monitoring, support, and just about everything that you could think to check on a GPU cloud. We’ll explain who’s been cutting deals, whose engineers have been hard at work, and whose clusters you should negotiate to come with a bottle of Aspirin. Without further ado, voilà the ClusterMAX 3.0 podium: Source: SemiAnalysis ClusterMAX 3.0, September 2026 Source: SemiAnalysis Neocloud Dashboard, available to our AI Cloud TCO Model subscribers Results (Executive Summary) Podcast summary discussion available on YouTube now! ClusterMAX 3.0 debuts with a comprehensive review of the neocloud industry, covering 77 providers. We increase our market view to cover 323 providers, up from 209 in ClusterMAX 2.0 , 169 in ClusterMAX 1.0 , and 124 in the original AI Neocloud Playbook and Anatomy article. We have now interviewed well over 200 end users of neoclouds as part of this research. We update our itemized list of criteria across 10 categories , and update our direct descriptions of our expectations for Slurm, Kubernetes, Standalone Machines, Monitoring Dashboards, and Health Checks . All of this content is live on our website. We encourage providers to use these lists when developing their offerings. We still consider these lists as an amalgamation of our experience interviewing end users, making them representative of the features that end users expect from their cloud providers. Nebius joins CoreWeave in the Platinum tier. While CoreWeave still sets the technical bar for others to follow, Nebius is now established as a provider that consistently commands a premium pricing over others. Strong business decisions by Nebius have put them in a position to serve an entire class of neolabs at seller’s prices. Google Cloud joins Oracle in the Gold tier. Azure moves to Silver, Fluidstack moves to Unavailable, and Crusoe drops to Bronze. Lambda, Firmus and TensorWave remain in Silver, while GMI moves up to Silver from Bronze. Many companies drop from Silver (or Gold) to Bronze or lower. We raise the bar this round as only 19 neoclouds globally achieve a Medallion rating. We establish a tier between Bronze and Underperforming: the Participation Ribbon tier. 15 providers join this rating, which more accurately describes our opinion that they do the bare minimum to get by. The rest of this article provides analysis of key trends: Financing, Blackwell and Grace-Blackwell deployments, the Move To Vera Rubin, Scale Out Networks, Reliability, Security (of course), and Agentic Coding. We provide an Appendix with individual comments about every provider that once again stretches this article to over 30,000 words. We hope you enjoy. In light of the SemiAnalysis imperative to publish the most rigorous possible testing of… everything… we are hiring MTS for ClusterMAX and adjacent projects. If you are a neocloud sicko, fire off an application here and let’s get to work. We are also hiring across Research, Consulting, and Technical Staff . We have multiple openings in our New York and San Francisco offices. Remote work is also acceptable. ClusterMAX MTS (full time): All levels of experience with Slurm, Kubernetes, and GPUs considered. Tokenomics MTS (full time): All levels of experience with model evals, harnesses, inference endpoints, and RL infrastructure considered. Research Analyst – AI Infrastructure and Economics ( full time or internship ): All levels of experience covering the financials of neoclouds, neolabs, and frontier labs tokenomics considered. Technical Consultant (full time): Lead engagements from technical strategy to technical due diligence. All levels of experience from consulting and technical backgrounds considered. Scope For the folks in the back: we are ranking managed clusters. This excludes a number of popular offerings throughout the industry. We are not testing who can build the best powered shell or run the tidiest datacenter—at least not directly. We are not handcrafting our images or hitting an API endpoint for tokens as a service. Here’s our idealized ClusterMAX reader: You dropped out of your Stanford PhD to pursue your dream of acquiring social status in San Francisco, getting started with the old reliable “Cladue make me pithc deck w technial languge for agetnci ai make no mistakes.” You have 7 or even as many as 10 figures to wager on compute, but you don’t have any opinion on Ubuntu versions or how to handle NAT, and you’ve never managed a fleet at scale. Ideally, all that infrastructure will fade into the background. You just want to focus on your idiosyncratic perf optimizations, model architectures, training strategies, data mixes, applications, or wherever else you seek your edge. To compete, you need bleeding-edge hardware, and you aim to spend 0 engineering time trying to figure out why GPU 7 in node 19 has been stalling your job, or why a quarter of your cluster has disappeared altogether. You need a managed cluster. The economic argument here is straightforward: a neocloud can amortize the cost of standing up all this management infrastructure over its huge fleet. It pays a team of engineers to build bulletproof health checks, keep its machine images up to date, make convenient monitoring systems, and handle all the related chores that we cover in great detail further on. In an ideal world, a lab

  • 0, September 2026 Source: SemiAnalysis Neocloud Dashboard, available to our AI Cloud TCO Model subscribers Results (Executive Summary) Podcast summary discussion available on YouTube now!
  • We raise the bar this round as only 19 neoclouds globally achieve a Medallion rating.
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Datacenters 2027: The Year of Acceleration, +50GW in a Single Year

Câu hỏi So, what are we seeing that others can’t?

Not a week goes without another piece of datacenter FUD. We’ve already debunked in depth why moratoriums aren’t as impactful as one may think . Despite that, many still misunderstand the pace of the global AI datacenter buildout. Just the other week, Brad Gerstner from Altimeter showed our data on the All-In Podcast and stated that “SemiAnalysis’ forecast of +43GW in 2027 is too aggressive” and that he thought only about 25GW would get built, close to half of our number, based on his “hunch”. But let’s not operate on hunches. It’s very tempting to cede to the FUD given all the constraints on the datacenter buildout: power, labor, supply chains, capital, and many more. But at SemiAnalysis, we care about the truth, not about narratives. And here’s the truth for you: one picture, >24GW of datacenters in advanced stage of construction, most of it to be delivered in 2027. Most of them are more than halfway done, with exterior cooling and electrical equipment already installed. That’s just 40 campuses, a tiny share of the multiple hundred sites in advanced stage of construction. We track them all, and for the full granularity, use our Datacenter Industry Model . Source: SemiAnalysis Datacenter Industry Model Of course, there are plenty of challenges and shortages. Behind these pictures, there are labor issues, supply chain delays, fundraising challenges, among many others. We’ve written in-depth about Modular Datacenters and Labor , Datacenter Electrical systems, Cooling systems, Behind-The-Meter power , the US Grid Constraints , and Nvidia’s ecosystem financial support . But the fact is simple: gigantic sites are simply getting built, and many are on time. For example, just go look at what Crusoe built in Abilene, what AWS built in Indiana, or QTS/Microsoft’s giant site in Georgia. These are the sites that have produced SOTA technology like GPT-6 and Fable 5.5. Looking ahead, despite all the constraints, an accelerating number of giant sites keep getting built, as shown by our Top 40 picture Source: SemiAnalysis Datacenter Industry Model If you ask any datacenter builder, they’ll tell you how hard it is, how some workers aren’t showing up, how vendors are delayed, and many other issues. All of this is accurate. And that’s why we’ve successfully called out countless datacenter delays and keep calling out developer timelines that appear overly aggressive in our Datacenter model, including the Oracle New Mexico Force Majure site delay in April . To be fair, +50GW is essentially the global cumulative datacenter market up through 2024 being built in one year! But our Datacenter Industry Model adjusts for all of that and still points to a material acceleration in datacenter capacity in 2027. So, what are we seeing that others can’t? Source: SemiAnalysis Datacenter Industry Model SemiAnalysis is a reader-supported publication. To receive new posts and support our work, consider becoming a subscriber. Let’s dig in. We’ll explore the following topics in this report: SemiAnalysis proprietary leading indicators: we’ll show a sneak peak of our proprietary data supporting our forecast, such as self-build construction starts and datacenter leasing (in MWs and USD). Year-to-date, slightly under USD 1T of binding / non-cancellable datacenter leases have been signed – that’s key to supporting our aggressive 2027-28 forecast. The shape of the buildout: how can we support +50GW in a single year? Where, what design, what power solutions? The “real” question: are we going to face a datacenter oversupply? As we explained in The Great Silicon Shortage , datacenters & power may be the tip of the iceberg, with memory and logic possibly becoming the new #1 constraint in 2027-28. Overall, we are very excited about 2027. The natural implication of the acceleration will be significantly more compute supply. That means cheaper and smarter AI, faster pace of AI model progress, and likely material revenue acceleration for leading AI labs and other token vendors. That also means unthinkable capex numbers. Reflexively rates will have to shoot up to enable the buildouts from a capital demand perspective. Let’s dive in! SemiAnalysis Datacenter Leading Indicators: how we see the buildout before everyone else Today we want to dive into two crucial indicators that give us strong confidence on our 2027 forecast. The first one is quarterly self-build construction starts. To accurately forecast US hyperscalers, we track the exact construction evolution of every single building. And the conclusion is simple: US hyperscalers have never launched more construction of datacenters. Each quarter, we track how many MWs start construction, and that number is going parabolic, driven in particular by two hyperscalers. If you want to know who or track the granularity of our data, subscribe to our Datacenter Industry Model. Source: SemiAnalysis Datacenter Industry Model The second indicator is Datacenter Leasing. We’ve talked about this a lot in prior reports, but as a quick recap: given the capital-intensity of building multi hundred MW datacenters, developers typically only spend 5-10% of project capital without a customer. They need a “pre-lease” to fund and develop the whole project. That’s typically a 15-20 year contract, with a TCV of $30-50B per GW. Check out our HPC Lease Tracker as part of the Datacenter Model to dive into actual contract economics. Subscribe now As such, tracking these lease contracts is a key leading indicator, as it leads debt raises, which itself leads the actual buildout. A big reason why 2026 datacenter growth isn’t “that” high relative to 2025, is that quarterly datacenter leasing remained relatively stable between H2 2023 and H1 2025. Of course, the key reason was the Microsoft Datacenter Freeze, which we called out before anyone else in December 2024 for Datacenter model subscribers : the Redmond giant single-handedly drove the leasing market in 2023 and 2024, then paused, and others could only pick up what was

Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

Together Post-Training: frontier-grade model shaping, the way that fits you

Post-train open LLMs like Qwen, GLM and Kimi into custom models you own, with SFT or RL. You write the training loop. Together runs the hard infrastructure.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Investing in the public benefit

We’re committing a portion of Anthropic’s shares to the Anthropic Public Benefit Fund, a new entity with a structure that increases the stake as the company grows.

Nguyên văn nguồn
Chi tiết nguồn
Kiến trúc

Computation and Data Movement for Inference

Mixture of Experts, now widely used in frontier models, has changed both the structure of serving and the economics of useful inference. It did more than increase parameter count. It changed which tensors are active for each token, what must remain close together, which transfers need strong local bandwidth, which can tolerate a weaker network link, and how memory movement, storage, and scheduling contribute to useful throughput. The best place to begin is the service as a whole. Inference runs inside a cluster coordinated by an orchestration layer such as NVIDIA Dynamo, Mooncake, or a custom scheduler. These work closely with inference servers like vLLM or SQlang, and those also have their own orchestration features. This article will not go into the details of how you work with the orchestration software or which you should choose. The article aims to provide an overview of the process and reasons for various features you may use. A user (or their agent) starts a conversation with a request (query), and the conversation may continue after answers with more requests. The request- answer is a “turn”. In modern AI systems the user is often running a in a client application (whether GUI or command-line oriented) and some of the answers coming back from the AI are intercepted by that application as instructions for it to run, things like editing your code or searching your company guidelines on an HR question. The AI system can also do some of those things itself from its data center, for example it may search the web. The results from these intercept tool actions are also returned to the AI as requests, creating more turns. The server keeps a context (often referred to as KV cache or just cache) which is the distillation of the session, allowing each new request, from user or from tool, to be properly interpreted for overall forward progress. There can be thousands of turns per hour when the user launches an agent on a long-running task. The conversation can also pause and resume hours or even days later. When a request arrives at the inference servers the orchestration puts it into a queue, while will feed into an input-fill worker, along with any context from system prompts and prior turns in the conversation. The query is then converted into new context appended to the conversation. That new context state is then moved to a decode worker. Decode repeatedly reads the accumulated state, generates answer tokens, and those answer tokens are also appended to the conversation state. The process may loop through tools, agents, users, and further input-fill work. A model worker running on GPUs is only one part of this token factory; storage, networking, and orchestration connect its stages. Inside the workers, four operating regimes are worth distinguishing from the beginning: Prefill , where the initial block of new tokens is processed together. Midfill , where a continuation request is appended to an existing cached context. Decode attention , where the context is used to generate the basics of newly generated tokens. Decode experts , where each basic new token is refined by a selection of experts a large total set of possible experts. These regimes place different demands on compute, memory, and networking. Prefill and midfill are closely related, with Prefill being a Midfill with zero prior context. However, Prefill is a common special case, corresponding to one-shot queries like classic “”chatbot” use, and can be optimized a bit differently than Midfill. Prefill reaches high arithmetic intensity (ratio of computation to data movement) and does not need to wait for context to be located and read. Midfill begins from an existing KV cache for prior state and appends a new request input sequence. This generally has a more moderate arithmetic intensity than prefill because there are fewer new tokens and more existing data to move per token. Decode attention and decode expert work are generally low arithmetic intensity since there are few new tokens compared to the prior context or expert weights, which at decode state are data that needs to be moved. Meanwhile, across all of them, the tensors for attention and for experts are the same regardless of token value, so there is benefit to sharing a single reading of a tensor with as many tokens as can be queued to use it, even tokens coming from unrelated requests by other users which simply happen to be running at the same time and which are networked closely enough to share. Treating all four mode as the same workload gives away much of the structural advantages that MoE models offer in terms of different intensity and different sharing patterns. In this document we will treat the 4 stages separately, and sometimes assume they are disaggregated. There are tradeoffs between aggregated (where the models stay on one server, which reconfigures as the request progresses between the stages) and disaggregated where the orchestrator may find a free slot on another machine already configured appropriately. Generally we will talk about the stages as disaggregated, with aggregation being a special case of always making the current node available when a previous stage finishes. There will be a section on the trade-offs at the end, after the work to be done has been more completely introduced. A transformer model is a deep stack of repeating layer groups. A group may contain one kind of layer, or it may contain a full-attention layer together with several linear, local, selective, or otherwise optimized-attention layers (this is a hybrid group of layers). Within each layer, attention, shared transformations, routing, selected experts, and recombination occur in sequence. Those steps are known in advance and repeat continuously, allowing the same hardware resources to serve different parts of the flow at different moments. The machines can also be described equally simple recurring units. A node is a group of accelerators, typically a tray in a rack, each accel

  • The results from these intercept tool actions are also returned to the AI as requests, creating more turns.
Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

Import AI 473: The US's superintelligence strategy; human brain in a mouse skull; and machine hermeneutics

Câu hỏi The five main uncertainties: Danger proximity : At what point does AI become so dangerous the risks aren’t worth the benefits?

Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv, cappuccinos, and feedback from readers. If you’d like to support this, please subscribe. Subscribe now RAND thinks the best AI strategy for the US is to keep all options open: …It’s not clear exactly what is optimal, so the US should spend money now to preserve options… RAND has published a lengthy paper about what the US should do to “secure geopolitical advantage on an uncertain path to superintelligence”. The main takeaway is that because so much of how the next phase of the AI takeoff will occur is unknown, the best thing is for the US to maintain a “Freedom of Action” strategy. The purpose of this strategy is “to build and preserve the ability to secure U.S. geopolitical advantage and ensure humanity’s survival with agency through the transition to Superintelligence”. The four key ingredients of the Freedom of Action Strategy: Build a human-AI ecosystem: Invest in AI safety; build tools to preserve human agency; invest in sharing the benefits of the technology; prepare society for AI-driven disruption; shape the incentives of the AI ecosystem. Build an AI-security architecture: Develop the tools to understand the frontier of AI systems both at home and abroad, including visibility into underlying compute; build technology to verify potential agreements; create regulatory expertise to govern the technology. Adapt legacy national security enterprises for the AI era : Overhaul all the security institutions of the country so that they’re ready for the changes AI brings on and able to capitalize on AI capabilities. Invest in the capacity of citizens, firms, and governments to respond: Give decisionmakers information to help them think about AI; develop and diffuse positive uses of AI to deal with its changes; build break-glass plans for responding to crises; raise AI literacy. Seven archetypal strategies in three families: Coexistence: Dominance ; the US leads development of AI and suppresses others. Co-Development ; the US leads a consortium (including China) to co-develop safe superintelligence with shared governance, pooled compute, and verification and monitoring. Preparedness ; the US mobilizes a coexistence effort via informal coordination, “including voluntary commitments, phased deployment, independent evaluation, incident reporting, and the domestic capacity to absorb and recover from AI incidents.” Denial: Moratorium ; the US pursues “a verifiable global halt to AI development above specified thresholds”. Deterrence ; the US “halts its own frontier development above defined thresholds and uses the full toolkit of national power to prevent and, if necessary, destroy foreign programs”. Continuity of Society ; “When every preventive strategy has failed or been judged infeasible, and coexistence is deemed impossible, the remaining objective is survival with some human populations retaining agency: The United States creates geographically distributed, biologically self-sustaining settlements hardened against AI-enabled threats”. Acceleration: The strategy implicitly believes that constraining AI development is more dangerous than AI development itself, so the US “relies on markets, competition, and rapid-iteration to produce safety as a byproduct of useful AI”. The five main uncertainties: Danger proximity : At what point does AI become so dangerous the risks aren’t worth the benefits? “If danger is close, strategies that buy time, build defenses, and invest in resiliency should be favored. If danger is not close or if humans are in greater danger without progress on advanced A, the case for accelerated development strengthens.” Coexistence feasibility: Is it possible for humans and AI systems to find a stable equilibrium? “If coexistence is feasible, strategies that build and deploy advanced AI become more desirable. If coexistence is infeasible, the case shifts toward restraining or suppressing development or other means of protecting human civilization.” Restraint feasibility: Can humans cooperate to restrain AI development? “If restraint is feasible, cooperative strategies become available. If it is not, restraint-based approaches are reachable only through more-coercive enforcement.” Decisive strategic advantage: Can a single actor build a decisive strategic lead over its competitors? “If decisive advantage is achievable and the United States assesses that it can maintain dominance, a unilateral strategy may be sensible. If a decisive advantage is impossible, because of proliferation or other causes, no single actor can leverage AI to dictate global terms.” Suppression feasibility: Can domestic or rival AI programs be suppressed by technology controls? “If suppression is feasible, Deterrence becomes possible. If it is not, actors can press ahead with dangerous development even when others restrain, without fear of being restrained themselves.” Why this matters - taking AI seriously requires spending money: The big takeaway from this RAND piece is that the US needs to spend a lot of money if it wants to pre-position itself to take advantage of continued progress in artificial intelligence - even if the main pre-positioning is about retaining optionality. Right now, it feels like the US strategy can mostly be described as the “acceleration” one described here by RAND, which is analogous to me to sitting in a car and spending all your resources on making the car go faster and upgrading the engine, and nothing on proactive safety measures like seatbelts or headlights or brakes. Read more : A U.S. Strategy to Secure Geopolitical Advantage on an Uncertain Path to Superintelligence (RAND) . *** Human researchers grow mice with partially human brain matter: …When is a mouse not a mouse? When a chunk of its brain is human brain tissue… A group of researchers have figured out how to grow chunks of human-like brains inside baby mice with deliberately depleted mouse brains, then use these animals as potential ex

  • “If decisive advantage is achievable and the United States assesses that it can maintain dominance, a unilateral strategy may be sensible.
Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

v0.31.0

<h2>What's Changed</h2> <h3>Notable Changes</h3> <ul> <li>The default API endpoint is now CoreWeave Forge at <code>https://api.forge.coreweave.com</code>, and run links point to <code>https://forge.coreweave.com/wandb</code>. Existing API keys stored for <code>api.wandb.ai</code> keep working without logging in again.</li> </ul> <h3>Added</h3> <ul> <li>Python 3.15 is now supported (<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/dmitryduev/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/dmitryduev">@dmitryduev</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5857700092" data-permission-text="Title is private" data-url="https://github.com/wandb/wandb/issues/13167" data-hovercard-type="pull_request" data-hovercard-url="/wandb/wandb/pull/13167/hovercard" href="https://github.com/wandb/wandb/pull/13167">#13167</a>)</li> <li><code>wandb.init(resume="must")</code> now works for offline runs. <code>wandb sync</code> fails if there is no run to resume (<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/geoffhardy/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/geoffhardy">@geoffhardy</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5318645297" data-permission-text="Title is private" data-url="https://github.com/wandb/wandb/issues/12666" data-hovercard-type="pull_request" data-hovercard-url="/wandb/wandb/pull/12666/hovercard" href="https://github.com/wandb/wandb/pull/12666">#12666</a>)</li> <li><code>wandb sweep-scheduler [entity/project/]sweep_id</code> runs a sweep's search locally with Optuna or Ax as the optimizer, configured in a <code>scheduler</code> block of the sweep config. <code>wandb sweep</code> prints the command for sweeps that set one (<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/kmikowicz-wandb/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/kmikowicz-wandb">@kmikowicz-wandb</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5494766731" data-permission-text="Title is private" data-url="https://github.com/wandb/wandb/issues/12900" data-hovercard-type="pull_request" data-hovercard-url="/wandb/wandb/pull/12900/hovercard" href="https://github.com/wandb/wandb/pull/12900">#12900</a>)</li> <li><code>wandb agent --max-consecutive-failed-runs N</code> shuts the agent down once N runs in a row have failed (<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/nathancy-wandb/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/nathancy-wandb">@nathancy-wandb</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5454864954" data-permission-text="Title is private" data-url="https://github.com/wandb/wandb/issues/12821" data-hovercard-type="pull_request" data-hovercard-url="/wandb/wandb/pull/12821/hovercard" href="https://github.com/wandb/wandb/pull/12821">#12821</a>)</li> <li><code>wandb sync</code> warns when a run was logged with a newer version of <code>wandb</code> than the one syncing it (<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/timoffex/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/timoffex">@timoffex</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5427982003" data-permission-text="Title is private" data-url="https://github.com/wandb/wandb/issues/12808" data-hovercard-type="pull_request" data-hovercard-url="/wandb/wandb/pull/12808/hovercard" href="https://github.com/wandb/wandb/pull/12808">#12808</a>)</li> <li>W&amp;B LEET TUI: <ul> <li><code>wandb leet inspect --summary</code> prints a run's state, latest metric values, config and last console lines from its local <code>.wandb</code> file, a quick way for a script or a coding agent to check on a run. <code>--json</code> prints the records, or the summary, as JSON, and <code>--follow</code> keeps printing records as a running run writes them (<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/dmitryduev/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/dmitryduev">@dmitryduev</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5560577252" data-permission-text="Title is private" data-url="https://github.com/wandb/wandb/issues/12951" data-hovercard-type="pull_request" data-hovercard-url="/wandb/wandb/pull/12951/hovercard" href="https://github.com/wandb/wandb/pull/12951">#12951</a>, <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5560578372" data-permission-text="Title is private" data-url="https://github.com/wandb/wandb/issues/12952" data-hovercard-type="pull_request" data-hovercard-url="/wandb/wandb/pull/12952/hovercard" href="https://github.com/wandb/wandb/pull/12952">#12952</a>)</li> <li><code>l</code> links the console logs to the charts in the single-run view: with the logs pane focused, the charts show a crosshair at the time the selected line was logged, and inspecting a chart with the mouse selects the console line logged nearest to that point. <code>l</code> again unlinks (<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/dmitryduev/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/dmitryduev">@dmitryduev</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5353177306" data-permission-text="Title

Nguyên văn nguồn
Chi tiết nguồn
  • v0.31.0 Weights & Biases — SDK releases · 17 thg 10, 2026
Kỹ thuật

Release v5.20.1

<h1>Release v5.20.1</h1> <p>Special surprise including the new model Beam from Reflection AI 🥳</p> <a target="_blank" rel="noopener noreferrer" href="https://private-user-images.githubusercontent.com/73884904/675834992-bfc7ff16-4340-490c-925c-b1512e4ceb32.jpeg?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3OTUwNjk2MTgsIm5iZiI6MTc5NTA2OTMxOCwicGF0aCI6Ii83Mzg4NDkwNC82NzU4MzQ5OTItYmZjN2ZmMTYtNDM0MC00OTBjLTkyNWMtYjE1MTJlNGNlYjMyLmpwZWc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYxMTE5JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MTExOVQwNjIxNThaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hMzBlZjk1MzEyNTRkOTgwYTVkODJlZTZkNzRjNjQxZjAxMzYyYzk3NjI5YTRlOWIwMGNhZjkwMTNmY2VkNzM5JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZyZXNwb25zZS1jb250ZW50LXR5cGU9aW1hZ2UlMkZqcGVnIn0.2QITVuv77xWO3iuPrxSfWIQ-sHM2vYO_NbPb1b6HkpU"><img width="2000" height="989" alt="Gemini_Generated_Image_ox4hjlox4hjlox4h" src="https://private-user-images.githubusercontent.com/73884904/675834992-bfc7ff16-4340-490c-925c-b1512e4ceb32.jpeg?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3OTUwNjk2MTgsIm5iZiI6MTc5NTA2OTMxOCwicGF0aCI6Ii83Mzg4NDkwNC82NzU4MzQ5OTItYmZjN2ZmMTYtNDM0MC00OTBjLTkyNWMtYjE1MTJlNGNlYjMyLmpwZWc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYxMTE5JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MTExOVQwNjIxNThaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hMzBlZjk1MzEyNTRkOTgwYTVkODJlZTZkNzRjNjQxZjAxMzYyYzk3NjI5YTRlOWIwMGNhZjkwMTNmY2VkNzM5JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZyZXNwb25zZS1jb250ZW50LXR5cGU9aW1hZ2UlMkZqcGVnIn0.2QITVuv77xWO3iuPrxSfWIQ-sHM2vYO_NbPb1b6HkpU" content-type-secured-asset="image/jpeg" style="max-width: 100%; height: auto; max-height: 989px;"></a> <p>Beam is a token-efficient, frontier-scale model focused on coding and agentic use cases. It is built for teams that want to run a powerful model on their own hardware, infrastructure, or cloud, and customize it to their specific needs.</p> <p><strong>Links:</strong> <a href="https://huggingface.co/docs/transformers/main/en/model_doc/beam" rel="nofollow">Documentation</a></p> <ul> <li>Fix small bug in auto mappings (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5883900140" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49664" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49664/hovercard" href="https://github.com/huggingface/transformers/pull/49664">#49664</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/yonigozlan/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/yonigozlan">@yonigozlan</a> in <a href="https://github.com/huggingface/transformers/pull/49664" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49664/hovercard">#49664</a></li> </ul> <h2>Some real patch fixes</h2> <p>We broke loading models on older torch versions 😢 should be fixed now. But here is the full list of fixes</p> <ul> <li>Fix HunYuan-VL image assignment during batched generation (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5825498496" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49557" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49557/hovercard" href="https://github.com/huggingface/transformers/pull/49557">#49557</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/Arnav66692/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/Arnav66692">@Arnav66692</a></li> <li>Fix DeepGEMM Mega MoE under expert parallelism(<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5762276340" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49428" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49428/hovercard" href="https://github.com/huggingface/transformers/pull/49428">#49428</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/IlyasMoutawwakil/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/IlyasMoutawwakil">@IlyasMoutawwakil</a></li> <li>Fix from_pretrained on torch 2.7: require torch&gt;=2.8 for distributed checkpointing (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5876370444" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49652" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49652/hovercard" href="https://github.com/huggingface/transformers/pull/49652">#49652</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/naveecode/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/naveecode">@naveecode</a></li> <li>Make DtensorShardOperation work on torch 2.6 and 2.7 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5878143473" data-permission-text="Title is private" data-url="https://github.com/huggingface/transformers/issues/49657" data-hovercard-type="pull_request" data-hovercard-url="/huggingface/transformers/pull/49657/hovercard" href="https://github.com/huggingface/transformers/pull/49657">#49657</a>) by <a class="user-mention notranslate" data-hovercard-type="use

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

Cloud Agent Canvases and Environment Improvements

Canvases from cloud agents Cloud agents can now build interactive canvases as part of their work. Share a finished canvas with your team by sending its link. <figure><video src="https://cursor.com/marketing-static/changelog/cloud-agent-canvases-and-environments.mp4" controls preload="metadata"></video></figure> Learn more about sharing canvases . Environments API The Environments API is now available in beta. Use it to programmatically create and manage the cloud environments your agents run in. Larger, faster environments Cloud agent environments now support up to 300 repositories. Large environments also load 5x faster.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

The AI expectations gap

Câu hỏi How will AI reshape the economy in the years ahead?

How will AI reshape the economy in the years ahead? What is the core source of disagreement about AI's future impact on the economy? This note presents new survey evidence on these questions.

Anthropic Research (unofficial RSS mirror) · Chưa ghi tác giả Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
581–600 / 738
Trang 30 / 37