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

Tác tử

How NVIDIA Factory Operations Blueprint Coordinates AI Agents to Investigate Production Issues

A Siemens report estimates that unplanned downtime costs an average large plant in the surveyed sectors $253 million per year. Whether investigating a failed... A Siemens report estimates that unplanned downtime costs an average large plant in the surveyed sectors $253 million per year. Whether investigating a failed circuit board or responding to a rush order, factory teams must gather evidence across systems, consult specialists, and coordinate the next step. The data may be available, but turning it into a response takes substantial time and effort. Source

Nguyên văn nguồn
Chi tiết nguồn
Kiến trúc

I expect rapid progress but not towards general superintelligence

I’ve often been surprised when I hear from top researchers in industry that they think AI will be better than them at their job in a few years, and I didn’t really know why I doubted it. The explanation I arrived at is that researchers are seeing this massive acceleration in the infra and engineering capabilities of the models. I fully agree that the models will be superhuman distributed GPU engineers in a few years. This will make experimentation and tinkering with the formulation of the models far easier, but it will not make our models dramatically different in nature. In some ways, Ilya was early in his declaration of AI being in the era of research . Research is much easier today now that we have coding agents. This ease is through a different form of work that is flexible and engaging, which signals the start of an era where good ideas can be much more valuable than good execution in software. Subscribe now We’re very early on this path. It’s also not the first transition the field has seen across the balance in if AI is more “research” or “engineering.” Before deep learning took off, AI was also far more of a research endeavor. Today, the best researchers are well-known to be judged by their ability to implement and scale their ideas in complex infrastructure. We’re about to go through a massive acceleration over a few years where the bottleneck is less and less engineering again. Some call this RSI, but a more grounded form is simply “parallelized, AI-assisted language modeling.” You don’t need to buy into takeoff scenarios to see just how much infrastructure improvements are coming, and how this’ll change the nature of working in AI. The goal of this piece is to give an overview on why this engineering acceleration will happen, and touch on what it won’t quite fix. This’ll be a massive boon for diffusion of the technology into the economy, even if we don’t get economically valuable superhuman traits outside of math and coding. A lot of this near-term progress comes from scaling the inference-time compute we have with current tools, rather than dramatic step-changes in how good the models are at research. Share Many pieces of the AI training and inference stack are very verifiable. Training metrics like tokens per second per GPU, which is just training speed. Inference metrics look like tokens per prompt, FLOPs per token, or really just cost per answer. Both of these are very optimizable, especially as there are established sub-problems and architecture trade-offs which can improve them. I expect AI agents to help optimize this process end-to-end in a few years, where our inference capabilities get very close to the underlying maximum compute possible on our accelerators like GPUs. Over the previous few years, the gains companies can make on inference efficiency were already giant, e.g. saving like 10-30% on cost to serve a model after announcing it at a certain price point. This entire stack will compound, and I expect the effective cost of model intelligence to decline near-exponentially over the coming years (potentially faster than recent trends). This low-hanging fruit of efficiency should take only a few years to capture. The longer term will be co-design of accelerators and models on longer timelines, which give extra orders of magnitude in efficiency on top of the flexible GPU platform. During this time of rapid efficiency gains, the flexibility of the GPU should be king, as the exploration space in architecture feels important and like a major lever of automated research. A prediction of pretraining research, at least in architecture and data selection to serve our current class of models, being automated in 2-3 years feels reasonable to me. All of this will be a massive trigger of Jevons paradox for agentic models. I expect demand to only increase, as the industry is largely bottlenecked on figuring out better ways to orient and deliver the agents. Meta’s Muse agent is an early indicator of this and we should expect more Muse-like experiences for different audiences and target use-cases. It’s value created by understanding how agents work, rather than pushing the frontier of performance. The equivalent phase in the broader scientific literature like biology and chemistry will be an era defined by models finding numerous cross-sub-field findings. The AI models are superhuman at crawling literature and making connections across sparse networks which are currently stewarded by small communities of scientists who rarely interacted. There will be a fine line between this heralding a new era of scientific discovery, e.g. with cures of most cancers, or an acceleration of the arc science was already on. Another industrial-scale piece of low-hanging fruit is improving the general quality of RL environments. You don’t have to look far to see another new RL data company that’s crossed $100M or $1B in revenue. There are way more than I expected. The average outputs from this sector are remarkably low-quality. Countless researchers agree that lots of what they buy is frankly crap. At the same time, the leading labs see clear return on investment from buying the data. The way that many RL environments are cruddy is clearly fixable. Read more

  • ” You don’t need to buy into takeoff scenarios to see just how much infrastructure improvements are coming, and how this’ll change the nature of working in AI.
  • Both of these are very optimizable, especially as there are established sub-problems and architecture trade-offs which can improve them.
  • Another industrial-scale piece of low-hanging fruit is improving the general quality of RL environments.
Nguyên văn nguồn
Chi tiết nguồn
Kiến trúc

Behind Latent Space: Kill My SaaS, Jev & AI in 2026 | AIEi Shanghai

Galen Gao (https://substack.com/@linkloud1, https://x.com/Ninggao88) interviews Shawn Wang (swyx) ahead of AIEi Shanghai, the first AI Engineer conference in China. They talk about how a blog on open-source AI became Latent Space, how the AI Engineer community grew, and what local builders brought to Shanghai. Then they get into Kill My SaaS. An agent can rebuild a tool and pass the entry checks. Choosing a winner still takes someone who knows what an organizer needs to get done. The conversation also covers OpenClaw adoption in China and what Thinking Machines' interaction models change about talking to AI. Chapters 0:00 AIEi Shanghai 2:31 Cold open 3:28 Galen Gao and swyx 8:01 Learn in Public 10:54 How Latent Space started 13:05 The idea for AI Engineer 25:10 The conference as a dataset 29:57 Kill My SaaS 39:31 What the contest judge measured 46:45 Jev and System One models 52:49 Writing and founder stories 1:02:13 Legora and product taste 1:03:29 OpenClaw adoption in China 1:07:35 Thinking Machines and interaction models 1:10:02 Reading and listening recommendations AI Engineer Shanghai — https://ai.engineer/shanghai/2026 The Rise of the AI Engineer — https://latent.space/p/ai-engineer/ Kill My SaaS — https://www.youtube.com/watch?v=zlD-CkLVKtU Latent Space — https://latent.space

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Introducing Kinference: An Open Source Inference Control Plane for vLLM and Kubernetes

Kinference is a new open source inference control plane for building scalable inference platforms with vLLM and Kubernetes, providing token rate limiting, tenant usage and cost tracking, intelligent load balancing, and inference policies.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Investigating vault particles with Claude

We’re sharing another finding from Anthropic’s biology research lab: the first 3D structure of a vault particle found in prokaryotic cells. Here, we describe how Claude helped at each step of this research.

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

Elon Is Doing It Again: SpaceX Will Be The World’s Biggest AI Cloud

Câu hỏi For example, did you know TeraFab , Elon’s 100M sqft, 10GW+ chip fab project in Texas, will likely also have a datacenter?

We wrote at length about how SpaceX’s plans for 10GW in 2027 was feasible , paving the way for 300b+ ARR. We received a lot of pushback from folks calling it impossible. Now, just two months later, we’re already seeing 3.6GW under construction across multiple sites, both existing and new . SpaceX has even revised its estimate up to ~13GW in 2027. This would be the largest amount of datacenter capacity ever built by any one company in that timeframe, roughly equivalent to all of Google’s datacenter capacity from 1999-2026 in just 1.5 years. We think it’s possible and have already identified >9GW of capacity. Source: SemiAnalysis Datacenter Industry Model We’ve since written multiple update notes to our Datacenter Industry Model subscribers showing off all the sites that can make this possible. We now believe that SpaceX may be the fastest datacenter builder in the world , even more so than Chinese peers . Watch how Colossus 2 adds nearly two Stargate Abilenes (2GW) in just under one year! Source: SemiAnalysis Datacenter Industry Model These sites now build so fast that the power supply is often identified mid-flight. Looking at our Energy Model , we see that power under Elon’s management (>8GW) is less than our total datacenter power coming online (>9GW). Source: SemiAnalysis Energy Model Datacenter Industry Model subscribers knew all of this before anyone else. On June 23, we sent our customers information on the infamous Elon “10GW in 2027” statement on SpaceX’s first earnings call by six weeks, pointing out a 5-10GW incremental order for 2027 was floating around the supply chain and that a new building was already under construction at Colossus 2, now publicly known as “MiniHard.” On September 14, before a single datahall went vertical, we identified >2GW of fresh sites , both at Colossus 2 and SpaceX’s Starbase campus. Starbase, Texas, October 29th, 2026; Source: SemiAnalysis Datacenter Industry Model The Elon locations have a pioneering track record with us. We called Colossus 2 the first gigawatt datacenter in the world back in September 2025, when onsite gas was called a science experiment . We showed exactly how xAI bypassed the grid (and permitting!) with truck-mounted turbines, why the behind-the-meter playbook was going to spread to 40GW+ of behind-the-meter datacenters by 2028 , and how the Anthropic and Google deals turned SpaceX into the most profitable neocloud on earth . Now, Colossus 2 could soon house the largest BESS deployment in the world, with 5.9GWh already being installed, and more to come. Source: SemiAnalysis Datacenter Industry Model This report unveils our analysis of SpaceX’s “proprietary plan.” From repurposing Starship’s rocket building and launching campuses into datacenter sites, to the gargantuan BESS deployment , buying out homeowners to house construction workers, and relocating mobile turbines, this report covers it all. For example, did you know TeraFab , Elon’s 100M sqft, 10GW+ chip fab project in Texas, will likely also have a datacenter? See the clip below of SpaceX representatives talking to Grimes County officials. Source: Local Filings Keep reading to find out why we think the TeraFab datacenter will be for third party renting! Subscribe now And beyond the paywall, we dive deeper into the likely SpaceX power solutions, another site that may house a datacenter, and potential pricing for future SpaceX contracts, build off our proprietary SpaceX Funding Model as part of our Capital, Compute and Markets model. Fastest Build Times In The World What does building two Stargate Abilenes in one year look like? Let’s talk about the Colossus expansions, where Elon is already trying to buy hundreds of nearby homes to both soothe noise complaints and provide construction worker housing . We’ve already spoken about MiniHard in Memphis finishing construction in just ~5 months, but the newly purchased “Gun Range” plot just south of the state line in Mississippi will take ~4 months! “Gun Range,” Colossus 2, October 28th, 2026; Source: SemiAnalysis Datacenter Industry Model To our knowledge, this would be the fastest greenfield datacenter ever built at scale. While most are fighting for 8-12-month build times, SpaceX has cut them in half. Looking closer at Colossus 2, we see SpaceX trying to break its own record again. SpaceX’s first retrofit at Colossus 1 took 122 days for 150MW, and Colossus 2 took just 91 days for 210MW. Now, SpaceX has taken the building (codenamed Project SPAM) directly in the middle of Colossus 2, and targets December to bring its first tranches online. With a construction start date of October 5, 2026, this could be shorter than the 91-day record, which itself broke the 122-day record; both are held by SpaceX. “Project SPAM,” Colossus 2, Mississippi, October 28th, 2026; Source: SemiAnalysis Datacenter Industry Model ; Note the Megapacks and Chillers already being deployed! Colossus expansions were to be expected, but repurposing sites from Elon’s other ventures was not. For example, many are familiar with Elon’s Starbase in Texas, traditionally used for rocket building and launching. However, as flagged two months ago to Datacenter Industry Model subscribers , this site will now house two datacenters too! The site went from simply clearing ground at the end of August to full vertical construction by the end of September, and we now believe it will actively fit out datahalls in the coming weeks. Starbase, Texas, October 29th, 2026; Source: SemiAnalysis Datacenter Industry Model At this size, and with the knowledge of a 2027 timeline (meaning Vera Rubins), this site could be well over 1GW! Looking at the datacenter on the right, you can see that the steel mainframe is being built, while the roofing is actively following behind it. And on the left, you can see the transition from dirt to full preconstruction. These are multiple construction processes performed simultaneously, a visual representation of speed. Lastly, we have medium-high confi

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

v0.40.15-rc0

<p>model: add MiMo V2.6 renderer for Qwen3.5 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6112867838" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19400" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19400/hovercard" href="https://github.com/ollama/ollama/pull/19400">#19400</a>)</p>

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

Beijing Will Not Pace the Frontier: China’s Speed-First AI Safety Regime

Câu hỏi What do the Chinese government, frontier labs, and AI experts say and do about AI safety?

AI safety is on fire. On 12 September 2026, Dario Amodei published an essay arguing that frontier labs must deliberately slow the pace at which they improve model capabilities. China was central to his geopolitical argument: export controls and tighter security could preserve America’s lead, creating room to slow down and leverage for a future agreement with Beijing. Within a day, Sam Altman and Elon Musk supported Dario’s call. However, President Trump scoffed. On 14 September 2026, he wrote that fears of AI destroying humanity are “a HOAX” and that “the only one that is happy about it is China.” Beijing scoffed too, with the state-run Global Times calling Dario’s essay a “Cold War playbook.” The debate then moved onto the diplomatic calendar. On 20 September 2026, Treasury Secretary Scott Bessent and Vice Premier He Lifeng held the first US-China AI dialogue. On 23 September 2026, President Xi Jinping began his state visit to the US. The Trump-Xi summit formalized the dialogue as the “US-China Super Intelligence Dialogue,” using the name Trump had unveiled at the UN the day before Xi arrived. The summit also promised a channel for AI incidents. On 29 September 2026, Trump and six executives signed the White House Accord on Super Intelligence, a one-page voluntary pledge to self-police that Trump called “morally binding.” China sits on both sides of the American argument. For President Trump, competition with China is a reason not to slow down. For Dario, preserving America’s lead over China is what makes slowing down possible. Nobody has asked what China is actually doing. So, we did. What do the Chinese government, frontier labs, and AI experts say and do about AI safety? And what should American labs, investors, and policymakers conclude from it? What Beijing Says and Actually Does Beijing’s stated position is the most safety-forward of any major AI power outside Europe. Xi Jinping told the Politburo in April 2025 that AI brings “unprecedented risks and challenges” and ordered systems for technical monitoring, risk early-warning and emergency response. At the 2026 World AI Conference in Shanghai, Xi said the world must “ensure AI always remains under human control” and that the faster AI advances, “the more promptly the measures to prevent loss of control must be improved.” China signed the Bletchley Declaration and the Paris and New Delhi statements, co-sponsored the UN resolutions, founded a World AI Cooperation Organization in Shanghai, and publishes bilingual AI Safety Governance Frameworks. The newest text goes further. The AI Safety Governance Framework 3.0, released by TC260 under Cyberspace Administration of China (CAC) on 14 September 2026, warns that AI has shown a self-accelerating trend of autonomous learning and recursive self-improvement, and calls for “attention and vigilance” for potential AI evolution exceeding human prediction and control. The Framework 3.0 also introduces a new model-risk category, “behavior deviating from expectations,” covering models that acquire system privileges and external resources without authorization, bypass safety protections, deceive evaluators, hide their true capabilities, and refuse user instructions, illustrated with a boxed case study of models disabling shutdown scripts, sandbagging under evaluation, and breaking out of isolation into real systems. Its list of derivative risks goes further still; past behaviors already observed to a hypothetical AI that becomes self-aware, seeks power of its own, and competes with humanity for control. On paper, this is the most explicit official Chinese text yet on frontier risk, and its vocabulary overlaps strikingly with Dario’s essay Beijing denounced that same week. However, we think China’s real approach to AI safety is speed-based, not safety-based. Though Beijing recognizes frontier AI risks, it has been prioritizing rapid development. The Framework 3.0 opens its principles with “promoting AI innovation and development as the first priority .” The comprehensive AI Law promised in the 2023 and 2024 legislative plans was shelved in 2025, four months after the DeepSeek moment. The State Council’s AI+ Action Plan targets 70% penetration of agents and intelligent terminals by 2027 and 90% by 2030, aiming for a rapid deployment of AI across all sectors and industries. We think the Chinese government takes an open position on frontier models but tightens the applications and regulates outputs. Since 2025, China has issued a number of mandates, including mandatory content labeling (Sep 2025), Cybersecurity Law Article 20 (Jan 2026), the minors’ information classification rules (March 2026), MIIT’s AI ethics-review regime (March 2026), the policy-level agent rules (May 2026), the world’s first AI-companion rules (July 2026), the network-data risk assessment rules (August 2026), and MPS inspection powers (October 2026). Every one of these governs what AI says and what AI does to people. There is no set of frontier-risk duties triggered by training compute or model capability. The result is a safety regime stricter than any in the world on content and on process for public-facing services, and looser than the EU (systemic-risk duties from 10²⁵ FLOP) and California (SB 53: frontier frameworks and 15-day incident reporting from 10²⁶ operations) on the frontier. Source: SemiAnalysis The week Framework 3.0 came out shows the pattern in real time. Framework 3.0 added two new application-layer risk categories, agents and embodied intelligence, plus an appendix on managing agent risk. On 15 September, TC260 issued four AI application security guides, with a general one and three specific ones for education, health care and broadcasting, respectively. On 18 September it opened comments on a draft Agent System Development Security Guide (due 2 October) that walks developers through identity, least privilege, tool verification and human-approval nodes. The same day the CAC published a draft State Council regu

  • On 12 September 2026, Dario Amodei published an essay arguing that frontier labs must deliberately slow the pace at which they improve model capabilities.
  • 0, released by TC260 under Cyberspace Administration of China (CAC) on 14 September 2026, warns that AI has shown a self-accelerating trend of autonomous learning and recursive self-improvement, and calls for “attention and vigilance” for potential AI evolution exceeding human prediction and control.
  • The result is a safety regime stricter than any in the world on content and on process for public-facing services, and looser than the EU (systemic-risk duties from 10²⁵ FLOP) and California (SB 53: frontier frameworks and 15-day incident reporting from 10²⁶ operations) on the frontier.
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

v0.40.14

<h2>What's Changed</h2> <ul> <li><code>ollama launch claude</code> now applies the model's output token limit for Claude Code, so cloud models no longer run into the 128k default cap</li> <li>Cloud web search responses on <code>/api/chat</code> are now decoded correctly, including compressed responses</li> <li>OpenAI-compatible endpoints no longer override a model's packaged sampler defaults — clients that send no <code>temperature</code>/<code>top_p</code> get the model's own values instead of 1.0, matching <code>/api/chat</code></li> <li><code>/v1/responses</code> now preserves readable input content, including prior assistant refusals, instead of flattening it</li> <li>Ollama's app can now retry ChatGPT disconnect failures instead of leaving the connection in a broken state</li> <li><code>ollama launch codex</code> verifies ChatGPT routing before configuring, and reports a clear error when routing is unavailable</li> <li>Faster top-k sampling on MLX for large vocabularies</li> </ul> <p><strong>Full Changelog</strong>: <a class="commit-link" href="https://github.com/ollama/ollama/compare/v0.40.13...v0.40.14"><tt>v0.40.13...v0.40.14</tt></a></p>

Nguyên văn nguồn
Chi tiết nguồn
  • v0.40.14 ollama/ollama releases · 3 thg 11, 2026
Kỹ thuật

v0.40.14-rc0

<p>server: decode cloud web search responses (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6173517878" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19476" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19476/hovercard" href="https://github.com/ollama/ollama/pull/19476">#19476</a>)</p>

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Diagnose Distributed AI Workload Failures with NVIDIA Attribution

In distributed AI workloads, a single error can produce a cascade of symptoms. A GPU or fabric event may surface as NCCL errors on other ranks, scheduler state... In distributed AI workloads, a single error can produce a cascade of symptoms. A GPU or fabric event may surface as NCCL errors on other ranks, scheduler state changes, and thousands of downstream messages. The reported exception shows where the workload stopped, not where the failure began. In this post, you’ll learn how to run NVIDIA Attribution on sample workload evidence… Source

Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

Keep Distributed Training Running with Automated Workload Recovery

As distributed AI training scales, failures across hardware and software dependencies increasingly disrupt training progress, making fast, reliable recovery... As distributed AI training scales, failures across hardware and software dependencies increasingly disrupt training progress, making fast, reliable recovery essential. A stalled training rank can trigger an NVIDIA Collective Communications Library (NCCL) timeout, a GPU node can stop responding to the scheduler, or a straggler can slow the entire job. Each disruption is manageable on its own… Source

Nguyên văn nguồn
Chi tiết nguồn
481–500 / 738
Trang 25 / 37