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

Kỹ thuật

vLLM Delivers Day-0 Support for MiniMax-M3.1-Flash

Day-0 vLLM support for MiniMax-M3.1-Flash on NVIDIA Blackwell, NVIDIA Rubin, and AMD MI350-series GPUs, with NVFP4 MSA kernels, NVFP4 MegaMoE, DSpark speculative decoding, AgentX results, and day-0 RL.

  • 1-Flash on NVIDIA Blackwell, NVIDIA Rubin, and AMD MI350-series GPUs, with NVFP4 MSA kernels, NVFP4 MegaMoE, DSpark speculative decoding, AgentX results, and day-0 RL.
Nguyên văn nguồn
Chi tiết nguồn
Tác tử

From tasks to goals

LangChain CEO and Co-founder Harrison Chase shares how we're moving from task-specific agents to goal-focused agents. Learn more.

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

Goal Engineering

The future of agents is them taking on open ended goals and driving towards completion of those goals. Learn more.

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

Flint-interactive: A language for creating interactive data visualizations

Câu hỏi This leaves us wondering: (As a user) Can I interact with charts directly in chat to explore the data and ask follow-up questions based on what I select?

At a glance Brings rich interactivity to AI-generated charts: Flint-interactive enables agents to create data visualizations that users can explore directly through clicking, filtering, zooming, highlighting, and cross-chart interactions. Makes complex interactions easier to build: Instead of requiring low-level visualization code, Flint-interactive lets users and agents define behaviors using high-level semantic events and update operators. Improves reliability and reduces development effort: By using Flint’s compiler to handle interaction logic and visual dependencies, the system helps avoid many of the layout, rendering, and debugging challenges common in custom interactive visualizations. Connects charts, text, tables, and applications: Flint-interactive supports interactive data stories, dashboards, and connected experiences where actions in one view can update charts, documents, widgets, or application state in another. AI agents make it easy to analyze data and generate reports with data visualizations. They help users examine patterns more efficiently than by reading verbose text. However, when we’re feeling exploratory, we want to go beyond static visualizations. : We want to interact with charts to zoom in on details, compare data across charts, or even highlight related text in the document from a chart selection, and vice versa. This requires agents to be able to create interactive data visualizations that communicate information while letting readers explore beyond the author’s chosen view, ideally with the quality of interactive stories in sophisticated data journalism. This leaves us wondering: (As a user) Can I interact with charts directly in chat to explore the data and ask follow-up questions based on what I select? (As a developer) Can I enable the AI agents I build for my customers to generate interactive visualizations that help them explore and understand their data? Yes, it’s possible, but there is a significant practical gap. Even with capable frontier models, we still face: High generation cost. Agents rarely produce a fully satisfactory interactive visualization in one turn. Repeated rounds of reasoning, coding, and debugging consume tokens and time. Reliability. Interactive visualizations are challenging enough that expressive interactive charts introduce poor layouts or broken interactions easily, and we risk introducing new problems when trying to fix them. These challenges help explain why consistently generating high-quality interactive visualizations remain difficult in real-world use. Meanwhile, built-in interactions in some visualization libraries are not expressive enough on their own to create bespoke interactive charts. And these challenges are results of the lack of an expressive yet high-level language to specify visual interactions. Flint-interactive Flint-interactive (opens in new tab) is a language for creating interactive data visualizations on top of Flint. It allows users to create dashboards, data stories, and interactive artifacts more easily. The power of Flint-interactive comes from Flint’s compiler, which translates user interactions into semantic events that capture the meaning behind what users select. It then uses high-level operators to update visualizations in response to those events, making sophisticated interactions far easier to create and maintain. This shifts the heavy burden of resolving complex dependencies between interaction and visual logic to the compiler, so the user (or the agent) can easily describe expressive interactions with high-level specifications. Example 1: We can enable the user to brush one visual view, and the chart highlights all groups of country selected by the user with a simple preset interaction linked-brush. How to use 1. Use Flint in your chat Use Flint Interactive directly in your chat or canvas environment through Flint-MCP and MCP Apps, following the MCP setup guide. With the latest release, tell your agent what to chart and which interactions you want; it can generate an interaction_spec alongside the Flint chart spec and render the interactive chart in your conversation. Example 2: Using Flint-interactive in the chat environment, allowing the agent to produce interactive charts more reliably with simple Flint specifications. It also enables the user to select and pass context to the follow-up turns. 2. Build an interactive app with Flint Build interactive charts in a dashboard, data story, or other application with Flint Interactive Canvas. Your app can use event handlers to process semantic events emitted by Flint Canvas, for example to filter a table, update another chart, or change application state. To let your app’s widgets control how the chart updates, send update operators to Flint Canvas. Example 2: Flint-interactive allows developers to define interactions between charts and external elements (e.g., table, paragraph). Here, the user can click the table to highlight points in the chart and verse versa. 3. Create bespoke interactions To create bespoke interactions beyond Flint’s presets, define a trigger and a custom handle function with the programming API that maps semantic events to custom chart update sequences. For example, a handler can highlight a clicked country and label it by name. It combines set-style and set-annotation without manipulating renderer-specific marks. Example 4: The agent can create bespoke interactive visualizations with Flint’s programming API. In this DimpVis example, the user can drag the point along the time path to visualize how all cities’ temperature and precipitation changes. How it works Flint Canvas is a state machine that tracks user interactions on the canvas and manages visual rendering state. Flint makes interaction specification easy by separating visual logic, managed by the compiler, from interaction logic, specified by the user at the semantic level: Visual logic: The compiler automatically resolves raw triggers, such as clicks, drags, and ke

  • Improves reliability and reduces development effort: By using Flint’s compiler to handle interaction logic and visual dependencies, the system helps avoid many of the layout, rendering, and debugging challenges common in custom interactive visualizations.
  • And these challenges are results of the lack of an expressive yet high-level language to specify visual interactions.
Microsoft Research · Chenglong Wang, Yuwei Xiao, Alper Sarikaya, Michel Galley, Chris Quirk, Jianfeng Gao Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Introducing ChatGPT for Sales

Explore the new Sales plugin and 30+ new and expanded partner plugins in ChatGPT to research prospects, prepare pitches, update your CRM, and move deals forward.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Quoting Go Proverbs

<blockquote cite="https://go-proverbs.github.io"><p>A little copying is better than a little dependency.</p></blockquote> <p class="cite">&mdash; <a href="https://go-proverbs.github.io">Go Proverbs</a>, <a href="https://news.ycombinator.com/item?id=50516689#50526408">via</a></p> <p>Tags: <a href="https://simonwillison.net/tags/go">go</a>, <a href="https://simonwillison.net/tags/programming">programming</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Kiến trúc

Learning the language of machine learning

<p>When deploying deep learning models in high-stakes scenarios, researchers would like to understand how they work. Today, this requires unraveling an enormously complex network of internal processing nodes to understand what the model knows and how it will likely behave in practice.</p><p>Jacob Andreas would rather have the models explain themselves.&nbsp;</p><p>The newly tenured associate professor develops techniques that enable artificial intelligence models to communicate accurately about what they learned from their training data, in language humans can understand.</p><p>“Rather than needing to use all these external tools to try and understand what is going on under the hood, maybe we can train models to talk about themselves in a way that is accurate,” says Andreas, an associate professor in the Department of Electrical Engineering and Computer Science (EECS) and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL).</p><p>By studying how machines come to understand language, Andreas and his group work toward building intelligent systems that can learn about the world from human guidance and communicate effectively with human collaborators. They develop methods to make the training process more natural and efficient, so AI models learn in a way that is more like human learning.</p><p>Ultimately, he and his group want to help people design more accurate and robust deep-learning models that are better suited for their intended application.&nbsp;</p><p>“We used to debug systems just by looking at their parameters and figuring out whether they make sense or not. Now, we need ‘mind-reading tools’ that can take this really complicated neural computation and translate it for us because it is just not in a language that we, as humans, know how to speak,” he says.</p><p><strong>A love of language</strong></p><p>Andreas’ fascination with language blossomed at a young age. As a child growing up in the San Francisco Bay Area, he was a voracious reader of any book he could get his hands on.&nbsp;</p><p>One such book, passed along to him by a family friend, contained a pivotal scene where the main character hacked into a computer system to save the day.</p><p>That piqued his interest in computers. A few years later, his father brought home a simple BASIC Engine that Andreas could run on their home computer, and he started immersing himself in the world of computer programming.</p><p>Andreas chose to major in computer science as an undergraduate at Columbia University but wasn’t sure if he wanted to pursue a career as a researcher, as a software engineer, or at a startup.</p><p>His friends signed up for research projects in faculty labs, so Andreas figured he would try that, too.</p><p>“I just blast emailed everyone in the computer science department who was working on anything vaguely AI-related, and the natural language processing faculty got back to me first. That is how I wound up in the field where I am now,” he says.</p><p>Working in the lab of Michael Collins, the Vikram S. Pandit Professor of Computer Science at Columbia University, Andreas completed an undergraduate thesis project on machine translation. His project involved adapting existing statistical techniques for representing the meaning of language into more powerful AI systems that could translate passages of text.</p><p>“I always felt like these deeper questions about how thought works or how to model the world were closer to the surface in language processing problems than in the other tasks that AI people were working on at the time,” he recalls.</p><p><strong>A seismic shift</strong></p><p>He wanted to keep exploring deep questions, so Andreas enrolled in graduate school at the University of California at Berkeley to pursue a career as an academic researcher.&nbsp;</p><p>Before starting his PhD studies, he spent a year earning a master’s degree at Cambridge University in England. In that historic environment, he was inspired to contemplate the nature of language learning as he worked on computational translation problems.</p><p>Those studies fueled his research when he joined the lab of Berkeley professor of computer science Dan Klein.&nbsp;</p><p>Andreas began his studies at the time when neural networks were starting to take off. The field of natural language processing was making a seismic shift from statistical methods to this new world of deep learning.&nbsp;</p><p>Statistical methods utilize transparent, mathematical formulas to analyze text, while deep learning techniques use opaque algorithms to automatically learn complex meanings from massive amounts of text data.</p><p>“It was fun to be there as we made this transition to the modern toolkit, and learning and inventing all of that stuff when it was still new,” Andreas says.</p><p>He began collaborating with scientists in computer vision labs to tackle increasingly complex questions about language learning. Together, they designed AI systems that excelled at generating natural language by teaching those systems to “see” and understand the world around them.&nbsp;</p><p>“This was all super-exciting because there really hadn’t been a common set of tools for working with both language and vision data. Then, all of a sudden, neural networks gave us this common language,” he recalls.</p><p>He completed a thesis project on the use of neural module networks, which are small fragments of neural networks that perform specific tasks, like localizing objects or logical reasoning, in a larger AI system. But it was the more speculative projects he tackled alongside his thesis, like interpretability and creating natural language descriptions of an AI model’s internal states, that would drive the focus of his group at MIT.</p><p><strong>Faculty focus</strong></p><p>After spending a year at the startup Semantic Machines (which spun out of his lab at Berkeley), Andreas joined the MIT faculty in 2019.</p><p>“It felt like this was a fantastic opportunity to

Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

Q&A: Why building math muscles remains important in the AI era

Câu hỏi &nbsp;</p><p><strong>Q</strong>: What was it like using AI to solve math problems?

<p><em>John Urschel PhD ’21, the Class of 1956 Career Development&nbsp;Assistant Professor in the MIT Department of Mathematics, is no stranger to building muscle: Before earning his PhD at MIT, Urschel strengthened his offensive muscles as a lineman for the Baltimore Ravens. He remains committed to helping his students exercise, but now his focus is on critical thinking skills. In the classroom, Urschel shares with his students his lifelong love for solving puzzles, while honing their abilities to be strong quantitative thinkers — skills he argues will be essential in an artificial intelligence-enabled world.&nbsp;</em></p><p><em>Urschel recently published two preprints on arXiv, one on the</em><a href="https://math.mit.edu/~urschel/preprints/random_growth.pdf" target="_blank"><em> growth factor of random matrices</em></a><em>, the other on worst-case behavior in&nbsp;</em><a href="https://math.mit.edu/~urschel/preprints/entrygrowth.pdf" target="_blank"><em>Gaussian elimination</em></a><em>, in which he used a combination of human insight and AI tools to solve problems that had stumped mathematicians for decades. That experience provided Urschel with new insights into the future of mathematics and higher education, and the enduring importance of the in-person, residential education offered on college campuses, where students can experience the importance of struggle, problem solving, and human connection. Urschel recently chatted with&nbsp;</em>MIT News<em> about education, AI, and mathematics; excerpts from the conversation follow.&nbsp;</em></p><p><strong>Q</strong>: What is it about studying math that you hope to share with your students?</p><p><strong>A</strong>: When I was very little, I loved solving puzzles, whether that’s Sudoku puzzles or Martin Gardner math puzzles or crossword puzzles. I think there’s something really enjoyable about being faced with some problem that you don’t quite understand, trying to get your bearings, trying to make your way through it, and finally gleaning some sort of understanding from it. That’s a process that I’ve enjoyed for as long as I can remember, and that’s the thing that I’ve really enjoyed most about math.</p><p>When it comes to teaching, first, I want to help our students train themselves and hone their ability to solve problems and their ability to be strong quantitative thinkers.</p><p>The second part, which is more specific to whichever course I’m teaching, is to help them assimilate the big ideas or tools or techniques or themes that help them become better problem solvers in that specific discipline. If they’re taking a course on a specific subject, the idea is for me to try to convey important fundamentals so that they can use that learned expertise to become better problem solvers out in the world.</p><p><strong>Q</strong>: You recently used AI to assist in solving two proofs. Can you talk a bit about the experience?</p><p><strong>A</strong>: These are the first two papers where I’ve ever touched the machine, and it does feel different. It’s not unlike having a collaborator, but it’s a collaborator who has access to unlimited resources, seemingly, and does certain things much, much faster than I ever could.&nbsp;</p><p>In the same way that computers can do calculations much faster than any of us can, I think at this point it’s very clear that artificial intelligence systems can do certain types of mathematics much, much more efficiently than any of us can.&nbsp;</p><p><strong>Q</strong>: What was it like using AI to solve math problems? Was it just as rewarding?</p><p><strong>A</strong>: For me, personally, I would say that the feeling that I talked about of problem solving, of thinking your way through a problem, I find that that’s most joyful for me when I do that by myself, without AI.&nbsp;</p><p>One thing that I think is really important is to have an independent sense of problem-solving ability, a sense of curiosity — this idea of building those mathematics muscles, which I think are important, even in the era of AI.</p><p>Especially when you are talking with a machine, asking the machine for things, I think it’s very important to elevate the idea of understanding. If you’re talking to a machine, let’s say, and it gives you a proof of some mathematical fact, I think the true value in that proof comes from a human understanding and assimilating the way in which the machine combines certain mathematical ideas, and then internalizing that and communicating that knowledge to others.</p><p>Understanding is the big thing that I stress to my students who use AI systems. And then for the students who do not use AI systems, I focus on building their problem-solving abilities, which is what we’ve done for centuries in mathematics.&nbsp;</p><p>I think in the age of modern AI machines, as long as there is a role for humans to think and try to solve problems, which I believe there is, then building these skills remains important.</p><p><strong>Q</strong>: The recent <a href="https://aiandeducation.mit.edu/" target="_blank">report from MIT’s Ad Hoc Committee on AI Use in Teaching, Learning and Research Training</a> stressed the importance of experiencing the necessary and productive struggle of learning, along with the collaborative aspect of learning. I’m curious what you think about the role of AI in math education going forward.&nbsp;</p><p><strong>A</strong>: On one hand, the development of AI technologies should be viewed as a giant win for education, a giant win for leveling playing fields when you have technology that’s easily accessible, that gives you access to vast amounts of knowledge.&nbsp;</p><p>Neither of my parents are professors; I remember when I was a kid, I didn’t know what a mathematician was until I got to college. And the way I learned about advanced mathematical concepts in high school, but also when I got to college, was with a laptop, and with Google. I was able to Google some topic I want to learn about, and it was

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Introducing OpenAI for Chile

OpenAI and the Government of Chile are partnering to support responsible AI adoption, build public-sector skills and help local startups develop AI applications.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Claude solves two open problems in the physics of black holes

Claude has solved two open problems in theoretical physics. Both concern black holes and the search for a theory of quantum gravity.

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

llm 0.37.3

<p><strong>Release:</strong> <a href="https://github.com/simonw/llm/releases/tag/0.37.3">llm 0.37.3</a></p> <p>A bug fix release for <a href="https://llm.datasette.io">LLM</a>, mainly to cover a problem with the most recent Homebrew package of the software.</p> <p>Homebrew now ships the latest SQLite, <a href="https://www.sqlite.org/releaselog/3_54_1.html">3.54.1</a>, which includes <a href="https://sqlite.org/bugs/info/2026-10-12T13:52:22Z">a fix</a> that <a href="https://github.com/simonw/llm/issues/1879">broke LLM's logging code</a>.</p> <p>Short version: in SQLite 3.54.0 the following SQL:</p> <pre><code>select json_object('a', (select json_array(1))) </code></pre> <p>Returns this:</p> <pre><code>{"a":[1]} </code></pre> <p>But in 3.54.1 the same query returns this:</p> <pre><code>{"a":"[1]"} </code></pre> <p>According <a href="https://sqlite.org/forum/forumpost/dff1a343157e422f">to D. Richard Hipp</a>:</p> <blockquote> <p>That was a bug fix. Per the documentation, (SELECT json_array(...)) returns SQL text, not JSON. Hence the 3.54.1 behavior is correct and 3.54.0 was a bug.</p> </blockquote> <p>The workaround looks like this:</p> <pre><code>select json_object('a', json((select json_array(1)))) </code></pre> <p>Tags: <a href="https://simonwillison.net/tags/llm">llm</a>, <a href="https://simonwillison.net/tags/webassembly">webassembly</a>, <a href="https://simonwillison.net/tags/sqlite">sqlite</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/d-richard-hipp">d-richard-hipp</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

Accelerate Molecular Dynamics Simulations with Accurate, Flexible Primitives Using NVIDIA cuPotential

Molecular dynamics (MD) is an indispensable tool across physics, chemistry, biology, and materials science. Owing to its arithmetic structure, particularly the... Molecular dynamics (MD) is an indispensable tool across physics, chemistry, biology, and materials science. Owing to its arithmetic structure, particularly the parallel evaluation of many atom-pair interactions, MD codes were among the first to be ported to run on GPUs. However, as GPUs evolve, optimizing kernels and ensuring compatibility across heterogeneous hardware have become more difficult… Source

Nguyên văn nguồn
Chi tiết nguồn
Kiến trúc

Building Visual Intelligence — Andreas Blattmann & Dustin Podell, Black Forest Labs

Câu hỏi How much can it learn about that missing view just by watching video?

A camera moves around a person. Halfway around, the model has to draw the back of a head it has never seen. How much can it learn about that missing view just by watching video? That’s the question we put to Andreas Blattmann and Dustin Podell of Black Forest Labs. Their answer runs through FLUX 3. BFL trained one backbone on images, video and audio. The guests explain what each adds, how they decide which parts of denoising deserve the most compute, and why a detail that looks great in a still can flicker from frame to frame. We also get into the case for autoregression and where they think diffusion still has an edge. There’s personal history here. Stable Diffusion is why Dustin left his software job and why swyx started Latent Space. They revisit the first FLUX training run in a French chalet, the photo experiment behind FLUX Kontext, and some of the things people actually use these models for, from restoring old family photos to generating a clip with synchronized sound. This episode is sponsored by Deepgram. Deepgram is the voice AI infrastructure behind natural conversation—bringing speech-to-text, text-to-speech, voice-focused LLMs, and agents to teams building with voice AI. https://deepgram.com/?utm_source=latent-space&utm_medium=sponsored-podcast&utm_campaign=fy26_q4_brand_campaign&utm_term=latent-space&utm_content=host-read-youtube Chapters 0:00 Host introduction 3:39 Cold open 4:23 Black Forest roots 9:23 Dustin’s path 17:04 Visual intelligence 27:38 Inside SDXL 33:59 Learning motion 39:57 Flow matching 46:57 Starting BFL 50:36 Deepgram sponsor break 51:43 Early use cases for FLUX 54:58 Editing with Kontext 1:03:15 Training data 1:07:32 Testing an edit 1:13:12 Autoregression 1:20:50 World action models 1:30:08 Robotics in 2027 1:32:16 Who uses FLUX? Show notes • BFL’s account of why it trained FLUX 3 across images, video and audio — https://bfl.ai/blog/flux-3 • Watch FLUX 3 generate video with synchronized sound — https://bfl.ai/blog/flux-3-video • Try the night-to-day image edit mentioned in the episode — https://bfl.ai/models/flux-kontext • Andreas Blattmann co-created Stable Diffusion and co-invented latent diffusion — https://raais.co/speakers-2025-andreas-blattmann-black-forest-labs • Dustin Podell led FLUX Kontext and was lead author of SDXL — https://dustinpodell.com/about/ • The first Latent Space essay, written after the Stable Diffusion release — https://latent.space/p/multiverse-not-metaverse/ Sponsorship and business inquiries: business@latent.space

Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

How the World’s Most Advanced AI Infrastructure Supply Chain Invents the Future

Câu hỏi Foxconn Q: How would you describe your job to someone outside NVIDIA?

Deb Shoquist leads operations at NVIDIA, overseeing the supply chain and manufacturing behind its AI platforms. That description sums up the job, but it doesn’t begin to encompass what Shoquist has made possible over decades. Shoquist, her team and their global network of manufacturing partners enable NVIDIA to introduce world-leading technologies year-round. Much of the team’s success comes down to her deep relationships with some of the world’s most accomplished manufacturers, including TSMC, Foxconn, Wistron, SPIL, King Yuan Electronics (KYEC), Unimicron, Lumentum, Coherent and Amkor. TSMC has partnered with NVIDIA since 1998. SPIL and KYEC were assembling and testing its chips more than two decades ago. For Shoquist, those relationships grow stronger through each challenge solved together on the factory floor, as every generation of technology demands more from the people and tools behind it. Ask her about the global network she relies on, and she starts with the people. There’s the supplier in Taiwan so eager to get production moving that he calls her at 3 a.m. with a fix to the latest problem. There are the factory workers in Vietnam whose only ask of her are more tools to get more work done. She remembers the partners that supported NVIDIA when it was small, and the work they’ve done to help each other grow. Shoquist meets with key suppliers quarterly and spends six weeks each year on the road visiting partners in Taiwan, China, Korea, Vietnam and the U.S., where conversations move from meeting rooms to factory floors. Recently, she paused her travels to accept the 2026 Dr. Lisa Su Woman of Innovation Award — and to share how she finds the right people and companies to imagine and produce at massive scale technology that has never been built before. Foxconn Q: How would you describe your job to someone outside NVIDIA? A: At the most basic level, our job is to convert designs into the physical world. Our design team creates a blueprint, and our job is to transition it into a reliable, high-performance supply chain. A matchmaker is probably the best analogy. You have to find the partner that matches your ramp style and has the technology that can meet the needs of very precise designs. The supplier has to be financially strong enough to invest in our ramps. They have to be highly technical. And they have to be willing to do something that has never been done before. If you don’t have all three, you can’t create products at the volume and speed that we need. You need those three legs, and in some cases, if you only have two, you have to invest in the company to make them a partner. Q: What’s an example of that in practice ? A: Amphenol’s cable cartridge is the most complex interconnect assembly they have ever built. When I recently visited the Vietnam factory, now in full production, I was struck by the level of precision required to build a product of this complexity. In this large cable cartridge, there are 1,500 small cable legs. Each one requires a very precise assembly to yield a quality level of less than 1 DPPM, or defect part per million. Amphenol built over a hundred automated lines to produce this product at the speed and scale we require. That takes know-how and technology, but it also takes heart, willingness and dedication. Creating anything of that complexity and precision with a supplier forges the partnership. The partners NVIDIA maintains are the ones who supported us when we were small, and yet are willing to scale with us today. Q: What changed as NVIDIA went from selling chips to building systems and platforms? A: When we were primarily a chip manufacturer, we were small, and the semiconductor industry was at scale. In that environment, it was about picking the best technology match for the products we envisioned. As we became a data center systems company, we needed more partners. But we’ve also now entered into new technology areas where the scale of the supply base is still developing. For example, our optical component partners have the technology and the willingness, but they don’t yet have the scale or capital to build fabs for the volume we need. To grow, we needed to support and invest, like we did with Coherent and Lumentum . Beyond that, with products like CPO (co-packaged optics), we’re creating entirely new supply chains with scale partners that are new to the optics world. Leading the integration of talent across the supply chain is the key to driving this incredible technology to high volume quickly. Wistron Q: You clearly enjoy spending time with supply partners, and in particular on the factory floor. What have you learned there over time that can’t be learned anywhere else? A: Over 20 years, you meet many individuals key to the success of your supply. You meet not only the CEOs, but also the people that are running the production floor. These are the people that really know the problems of manufacturing. If you don’t have a tight working relationship with the people deeper in the organization, then it’s hard to have an open dialogue about any challenges. But when you have tight relationships, they want to help you, and they feel comfortable enough telling you, “I need better technical support on the floor. I need more quality engineers. We need a better design. We need to automate this. We need more buffers of material.” That’s what 20 years of working with our suppliers has taught us. It’s year after year of learning from people, getting to know people deeper in the organization and creating the trust where they can tell us what we need to do to be successful with them. Q: Demand for NVIDIA technology is already massive and growing. Your job is to make sure NVIDIA’s volume can keep up. How do you make it happen? A: It goes back to relationships. When you decide that you’re going to push a line hard to increase volume, you have to have the relationships. You have to know their processes in detail. You have to know their people in detail, and they

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

v0.40.17

<h2>What's Changed</h2> <ul> <li>Improve ChatGPT CLI restore and graceful restart by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/hoyyeva/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/hoyyeva">@hoyyeva</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6251854019" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19583" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19583/hovercard" href="https://github.com/ollama/ollama/pull/19583">#19583</a></li> <li>cmd/launch: use thinking metadata for OpenCode settings by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/hoyyeva/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/hoyyeva">@hoyyeva</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6122368894" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19409" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19409/hovercard" href="https://github.com/ollama/ollama/pull/19409">#19409</a></li> <li>llm: use mmap for Vulkan integrated GPUs by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/dhiltgen/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/dhiltgen">@dhiltgen</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6309814784" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19640" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19640/hovercard" href="https://github.com/ollama/ollama/pull/19640">#19640</a></li> <li>rpreslar4765/Ollama by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/rpreslar4765/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/rpreslar4765">@rpreslar4765</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="4877598143" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/17163" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/17163/hovercard" href="https://github.com/ollama/ollama/pull/17163">#17163</a></li> <li>mlx: support concurrent generation by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/dhiltgen/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/dhiltgen">@dhiltgen</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5970886020" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19197" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19197/hovercard" href="https://github.com/ollama/ollama/pull/19197">#19197</a></li> <li>llama.cpp: version bump b12257 by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/dhiltgen/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/dhiltgen">@dhiltgen</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6315369343" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19646" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19646/hovercard" href="https://github.com/ollama/ollama/pull/19646">#19646</a></li> <li>MLX, MLX-C: version bump by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/dhiltgen/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/dhiltgen">@dhiltgen</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6316056226" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/19647" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/19647/hovercard" href="https://github.com/ollama/ollama/pull/19647">#19647</a></li> </ul> <h2>New Contributors</h2> <ul> <li><a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/rpreslar4765/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/rpreslar4765">@rpreslar4765</a> made their first contribution in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="4877598143" data-permission-text="Title is private" data-url="https://github.com/ollama/ollama/issues/17163" data-hovercard-type="pull_request" data-hovercard-url="/ollama/ollama/pull/17163/hovercard" href="https://github.com/ollama/ollama/pull/17163">#17163</a></li> </ul> <p><strong>Full Changelog</strong>: <a class="commit-link" href="https://github.com/ollama/ollama/compare/v0.40.16...v0.40.17-rc0"><tt>v0.40.16...v0.40.17-rc0</tt></a></p>

Nguyên văn nguồn
Chi tiết nguồn
  • v0.40.17 ollama/ollama releases · 17 thg 11, 2026
281–300 / 562
Trang 15 / 29