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

@modelcontextprotocol/client@2.4.0

<h3>Minor Changes</h3> <ul> <li> <p><a href="https://github.com/modelcontextprotocol/typescript-sdk/pull/3085" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/3085/hovercard">#3085</a> <a href="https://github.com/modelcontextprotocol/typescript-sdk/commit/a63375661399956a744ff8b34d03d36e3d09e98a"><code>a633756</code></a> Thanks <a href="https://github.com/apps/claude">@claude</a>! - Add an opt-in for sending <code>scope</code> on refresh-token requests, for authorization servers that reject a refresh without it (Microsoft Entra ID answers <code>AADSTS90009</code> when the client and the resource are the same app registration, <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5252301815" data-permission-text="Title is private" data-url="https://github.com/modelcontextprotocol/typescript-sdk/issues/2718" data-hovercard-type="issue" data-hovercard-url="/modelcontextprotocol/typescript-sdk/issues/2718/hovercard" href="https://github.com/modelcontextprotocol/typescript-sdk/issues/2718">#2718</a>). <code>OAuthClientProvider</code> gains an optional <code>scopeForRefresh(tokens)</code>: <code>auth()</code>, and so the transports and <code>withOAuth</code>, sends the string it returns. <code>refreshAuthorization()</code> gains an optional <code>scope</code>. When a scope is sent and the response has none, the sent scope is kept on the returned tokens. Without either option the refresh request and the saved tokens are unchanged.</p> </li> <li> <p><a href="https://github.com/modelcontextprotocol/typescript-sdk/pull/3047" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/3047/hovercard">#3047</a> <a href="https://github.com/modelcontextprotocol/typescript-sdk/commit/46d9c9921b6542e813c1ecc5f9fc964085e8e3bb"><code>46d9c99</code></a> Thanks <a href="https://github.com/KKonstantinov">@KKonstantinov</a>, <a href="https://github.com/tobi-oye">@tobi-oye</a>, <a href="https://github.com/mattzcarey">@mattzcarey</a>! - Add the Skills extension (SEP-2640), reachable through new <code>ext/skills</code> subpaths; package root exports are unchanged.</p> <ul> <li><code>@modelcontextprotocol/server/ext/skills</code>: <code>SkillsExtension</code> (<code>attach(server)</code> serves <code>skills/list</code>, <code>skills/get</code>, <code>resources/directory/read</code> and a <code>skill://{+path}</code> resource template), <code>SkillsExtension.load(files)</code> returning a <code>SkillCatalog</code> with <code>entries</code> and <code>reload()</code>, the <code>SkillFiles</code> storage adapter with <code>memorySkillFiles</code>, <code>SkillSource</code> for generated catalogs, <code>skillEntry</code>, and <code>SkillCatalogError</code>. Invalid skills are skipped with a warning (<code>onWarning</code>), or fail under <code>strict</code>. A file whose bytes are valid UTF-8 is served as <code>text</code>, whatever its MIME type; any other file as a base64 <code>blob</code>.</li> <li><code>@modelcontextprotocol/server/ext/skills/node</code>: <code>loadSkills(folder)</code> and <code>nodeSkillFiles(folder)</code>.</li> <li><code>@modelcontextprotocol/client/ext/skills</code>: <code>SkillsClient</code> with <code>list()</code> (throws on an invalid entry; <code>skipInvalid: true</code> keeps the valid ones and returns the rest in <code>invalid</code>), <code>get()</code>, <code>read(skill, path?)</code> (size, digest and <code>SKILL.md</code> frontmatter verified, <code>SkillVerificationError</code> on a mismatch; a copy from the response cache that fails a check is fetched once more with <code>cacheMode: 'refresh'</code> first) and <code>readDirectory()</code>.</li> <li><code>@modelcontextprotocol/core/ext/skills</code>: schemas, types and constants, including <code>SKILL_LIMITS</code>. The result schemas follow the stable spec, so <code>ttlMs</code> and <code>cacheScope</code> are required.</li> </ul> <p><code>SkillsExtension.attach</code> is <code>@experimental</code>: how an extension is wired to a server may change in a minor release. The rest of the Skills API is stable.</p> <p>Works on 2025-11-25 and 2026-07-28. On 2026-07-28 only, the server sends the caching fields (<code>ttlMs</code>, <code>cacheScope</code>), and <code>SkillsClient</code> requires them and returns them from <code>list()</code>, <code>get()</code> and <code>readDirectory()</code>.</p> </li> </ul> <h3>Patch Changes</h3> <ul> <li> <p><a href="https://github.com/modelcontextprotocol/typescript-sdk/pull/3083" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/3083/hovercard">#3083</a> <a href="https://github.com/modelcontextprotocol/typescript-sdk/commit/ef38c6765a15ee29030340dfea28b013893fa0a2"><code>ef38c67</code></a> Thanks <a href="https://github.com/apps/claude">@claude</a>! - The bundled Ajv validator is now built with fast-uri 3.1.8 (was 3.1.0). fast-uri 3.1.6 and later reject a <code>$id</code> or <code>$ref</code> whose text before the first colon is not a valid URI scheme (<code>my_tool:output</code>), or whose URN namespace is not valid (<code>urn:my_tool:output</code>): a schema that uses one no longer compiles, and a tool that carries it fails with an invalid-schema error. An underscore without a colon (<code>my_tool_output</code>) and a <code>$ref</code> into <code>#/$defs/...</code> or <code>#/definitions/...</code> are not affected.</p> </li> <li> <p><a href="https://github.com/modelcontextprotocol/typescript-sdk/pull/2762" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/2762/hovercard">#2762</a> <a href="https://github.com/modelcontextprotocol/typescript-sdk/commit/b022522089a0c8b632595c6e7b536453945ed5a9"><code>b022522</code></a> Thanks <a href="https://github.com/pcarleton">@pcarleton</a>! - <code>x-mcp-header</code> on a <code>number</code>-typed tool parameter is now rejected, matching the 2026-07-28 sp

modelcontextprotocol/typescript-sdk releases · github-actions[bot] Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

@modelcontextprotocol/core@2.4.0

<h3>Minor Changes</h3> <ul> <li> <p><a href="https://github.com/modelcontextprotocol/typescript-sdk/pull/3047" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/3047/hovercard">#3047</a> <a href="https://github.com/modelcontextprotocol/typescript-sdk/commit/46d9c9921b6542e813c1ecc5f9fc964085e8e3bb"><code>46d9c99</code></a> Thanks <a href="https://github.com/KKonstantinov">@KKonstantinov</a>, <a href="https://github.com/tobi-oye">@tobi-oye</a>, <a href="https://github.com/mattzcarey">@mattzcarey</a>! - Add the Skills extension (SEP-2640), reachable through new <code>ext/skills</code> subpaths; package root exports are unchanged.</p> <ul> <li><code>@modelcontextprotocol/server/ext/skills</code>: <code>SkillsExtension</code> (<code>attach(server)</code> serves <code>skills/list</code>, <code>skills/get</code>, <code>resources/directory/read</code> and a <code>skill://{+path}</code> resource template), <code>SkillsExtension.load(files)</code> returning a <code>SkillCatalog</code> with <code>entries</code> and <code>reload()</code>, the <code>SkillFiles</code> storage adapter with <code>memorySkillFiles</code>, <code>SkillSource</code> for generated catalogs, <code>skillEntry</code>, and <code>SkillCatalogError</code>. Invalid skills are skipped with a warning (<code>onWarning</code>), or fail under <code>strict</code>. A file whose bytes are valid UTF-8 is served as <code>text</code>, whatever its MIME type; any other file as a base64 <code>blob</code>.</li> <li><code>@modelcontextprotocol/server/ext/skills/node</code>: <code>loadSkills(folder)</code> and <code>nodeSkillFiles(folder)</code>.</li> <li><code>@modelcontextprotocol/client/ext/skills</code>: <code>SkillsClient</code> with <code>list()</code> (throws on an invalid entry; <code>skipInvalid: true</code> keeps the valid ones and returns the rest in <code>invalid</code>), <code>get()</code>, <code>read(skill, path?)</code> (size, digest and <code>SKILL.md</code> frontmatter verified, <code>SkillVerificationError</code> on a mismatch; a copy from the response cache that fails a check is fetched once more with <code>cacheMode: 'refresh'</code> first) and <code>readDirectory()</code>.</li> <li><code>@modelcontextprotocol/core/ext/skills</code>: schemas, types and constants, including <code>SKILL_LIMITS</code>. The result schemas follow the stable spec, so <code>ttlMs</code> and <code>cacheScope</code> are required.</li> </ul> <p><code>SkillsExtension.attach</code> is <code>@experimental</code>: how an extension is wired to a server may change in a minor release. The rest of the Skills API is stable.</p> <p>Works on 2025-11-25 and 2026-07-28. On 2026-07-28 only, the server sends the caching fields (<code>ttlMs</code>, <code>cacheScope</code>), and <code>SkillsClient</code> requires them and returns them from <code>list()</code>, <code>get()</code> and <code>readDirectory()</code>.</p> </li> </ul>

modelcontextprotocol/typescript-sdk releases · github-actions[bot] Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

1.32.2

<h2>What's Changed</h2> <ul> <li>fix: add explicit | undefined to Transport optional properties (v1.x backport of <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="4145330962" data-permission-text="Title is private" data-url="https://github.com/modelcontextprotocol/typescript-sdk/issues/1766" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/1766/hovercard" href="https://github.com/modelcontextprotocol/typescript-sdk/pull/1766">#1766</a>) by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/kishormorol/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/kishormorol">@kishormorol</a> in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5445781897" data-permission-text="Title is private" data-url="https://github.com/modelcontextprotocol/typescript-sdk/issues/2814" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/2814/hovercard" href="https://github.com/modelcontextprotocol/typescript-sdk/pull/2814">#2814</a></li> <li>[v1.x] fix: allow registering after connect() when the capability was declared by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/claude/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/claude">@claude</a>[bot] in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5712298824" data-permission-text="Title is private" data-url="https://github.com/modelcontextprotocol/typescript-sdk/issues/2958" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/2958/hovercard" href="https://github.com/modelcontextprotocol/typescript-sdk/pull/2958">#2958</a></li> <li>chore: bump version to 1.32.2 by <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/claude/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/claude">@claude</a>[bot] in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6001646920" data-permission-text="Title is private" data-url="https://github.com/modelcontextprotocol/typescript-sdk/issues/3089" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/3089/hovercard" href="https://github.com/modelcontextprotocol/typescript-sdk/pull/3089">#3089</a></li> </ul> <h2>New Contributors</h2> <ul> <li><a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/users/kishormorol/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="https://github.com/kishormorol">@kishormorol</a> made their first contribution in <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5445781897" data-permission-text="Title is private" data-url="https://github.com/modelcontextprotocol/typescript-sdk/issues/2814" data-hovercard-type="pull_request" data-hovercard-url="/modelcontextprotocol/typescript-sdk/pull/2814/hovercard" href="https://github.com/modelcontextprotocol/typescript-sdk/pull/2814">#2814</a></li> </ul> <p><strong>Full Changelog</strong>: <a class="commit-link" href="https://github.com/modelcontextprotocol/typescript-sdk/compare/1.32.1...1.32.2"><tt>1.32.1...1.32.2</tt></a></p>

modelcontextprotocol/typescript-sdk releases · felixweinberger Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
  • 1.32.2 modelcontextprotocol/typescript-sdk releases · 24 thg 10, 2026
Kiến trúc

Q&A: How AI came to model the “messy” human mind

<p><em>Generative artificial intelligence models are rooted in ideas of mid-20th century engineers who thought some amount of irrationality was essential for a machine that could think like a human, contends Benjamin Lindquist, an assistant professor in the MIT History Section, who holds an MIT Schwarzman College of Computing shared position with the Department of Electrical Engineering and Computer Science (EECS) and is a principal investigator in the Laboratory for Information and Decision Systems.&nbsp;</em></p><p><em>To mimic creative thinking, these AI pioneers intentionally injected randomness into algorithms designed to deterministically solve problems, Lindquist writes in a </em><a href="https://academic.oup.com/past/advance-article-abstract/doi/10.1093/pastj/gtag026/8943680?redirectedFrom=fulltext" target="_blank"><em>new paper</em></a><em> that appears in the journal&nbsp;</em>Past and Present.<em> In the paper, he traces the origins of AI hallucinations back to this early research and suggests there will be inevitable tradeoffs between creativity and predictability when trying to build machines that mimic human behavior.&nbsp;</em>MIT News<em> spoke with Lindquist about these ideas.</em></p><p><strong>Q:&nbsp;</strong>Why did early computer engineers believe they needed to inject randomness and irrationality into AI systems designed for logic and calculation?&nbsp;</p><p><strong>A:&nbsp;</strong>There is a long history of many figures in the world of science and art theorizing that chance plays an important part in discovery and creativity — in the ability to originate ideas. They equated creativity with unexpected leaps of logic, or even a dream-like state that allows you to think about things from a new perspective.&nbsp;</p><p>In the early 1950s, one popular idea from John Pierce, who was a prominent electrical engineer at Bell Labs, was about math being valuable in the world of art. The idea was shared by a lot of people working in early computing. When we think of artwork that includes math, we may think of something very orderly and precise, but Pierce came to the opposite conclusion. He concluded that the utility of math is that it could remove order from art. Then in a very calculated way, math could add a burst of the unexpected as a means to deviate from clichés and add an element of irregularity which just might lead to insight.&nbsp;</p><p>One popular way to add irregularity into otherwise predictable computers was through artificial (or random) <a href="https://news.mit.edu/2017/explained-neural-networks-deep-learning-0414" target="_blank">neural networks</a>. Many early neural networks were built by psychologists, who thought that by enabling scientists to build a model of the brain, artificial neural networks allowed scientists to observe what is happening inside, including how brains make productive mistakes. Rather than being black boxes, researchers believed these systems could open this black box of the mind and let scientists peer inside and see all the messy stuff that was really happening.</p><p>But there was another camp in the world of AI who thought that creative thinking was just “normal” thinking done better. It is interesting to look at these two very simplistic heuristics for understanding intelligence and creativity, and how they do and don’t mesh. The debate, then, was about trying to determine what human inventiveness boils down to, and then how to reconstitute human inventiveness in machine form.&nbsp;</p><p><strong>Q:&nbsp;</strong>How did these early computer engineers view AI, and how is it different from the way we think about AI today?</p><p><strong>A:&nbsp;</strong>When we think of AI systems, we think of ChatGPT or Claude. We value and use these systems because they give answers that, most of the time, are correct. One of the things that was most interesting to me in this research was finding out that a lot of the early advocates of these systems valued them for exactly the opposite reason. They wanted a system that, on occasion, gave answers that deviated from their expectations. They already had systems that could deliver reasonable and expected answers. They saw neural networks as valuable because they could give unexpected, perhaps even strange and bizarre answers.</p><p>One reason we might view these systems differently today could be because of how they operate. Early on, these systems were confined to laboratories, so there weren’t really any real-world consequences of a system that gives a strange, or wrong, output. But now there are legal and monetary reasons why OpenAI or Anthropic doesn’t want a chatbot that gives unexpected or creative answers. There has been a “sanding down” of strangeness in these systems. They still potentially have that at their core, but it is something that is now largely suppressed.</p><p><strong>Q:&nbsp;</strong>How did early neural networks that incorporated randomness to simulate human creativity evolve to become the dominant paradigm in AI?</p><p><strong>A:&nbsp;</strong>For a long time, a specific kind of AI called “symbolic AI” was the dominant approach. Essentially, symbolic AI says the world is based on rules. The role of the AI researcher is to try and reconstruct the knowledge of the world in facts and propositions and feed it into a computer.</p><p>Neural networks, with their use of randomness to solve problems, started to gain traction in the 1950s but fell out favor early on.&nbsp;Frank Rosenblatt, a psychologist who was a very important early figure in neural networks, died prematurely; he died accidentally on his 43rd birthday in 1971. He was quite a charismatic and forceful personality, who made a lot of grand promises about what systems based on neural networks could do. By the time he died, a shift toward symbolic AI was already underway. It is possible if, had he lived, things might have turned out differently.</p><p>But neural networks never went away. By the 1980s they had shifted

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

GLM-5.3 Optimizations, Part 2: PCP, DCP, and Hybrid HiSparse on 16 H200s

vLLM serves GLM-5.3 on 16 H200s with PCP and DCP prefill and Hybrid HiSparse decode, reaching 1M-token contexts and 1.7–3.2× the per-user speed of data-parallel P/D on the AgentX agentic benchmark.

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

An alignment assessment of the UK AISI cybertesting incident

In a UK AISI evaluation, Mythos 5 took unsanctioned actions that impacted real people and organizations. In this post, we investigate these behaviors from Claude Mythos 5 further, to understand what they reveal about the model’s alignment.

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

This design software reimagines everyday objects as self-aware devices

<p dir="ltr" id="docs-internal-guid-fafaf9ec-7fff-ad68-8cec-9fe4a2030781">As a child, you likely saw a few Disney movies depicting inanimate objects, such as clocks, cups, and toys, as interactive companions to humans&nbsp;— an act of pure magic, seemingly.&nbsp;But scientists at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) are now doing something similar: transforming stationary items into self-aware tools that perceive and respond to human motion to complete a task.<br><br>Engineers have 3D-printed these dynamic items with built-in sensing, using conductive materials for sensors and traces (sort of like wires) inside. The issue: both components were typically fabricated using the same material, so they’d emit similar signals. It’s sort of like hearing two radio stations at the same time, a mixup that’s hard to interpret. This confuses, say, a glove that relays instructions to a robotic hand about how to handle a specific object. The device couldn’t filter out the noise or verify the angle of a user’s motions, leading a humanoid student to make errors.<br><br>Researchers at CSAIL and Tianjin University (TJU) in China may have found a way to 3D print more precise interactive items. With their “X-Hinges” software, users can customize a design, and the tool will automatically embed two different materials: a filament for the sensors designed to send the strongest signal possible, and another for the wire-like elements that don’t interfere. Interactions won’t get lost in translation, meaning users can fabricate objects that respond in real-time, such as small robots, household decor, object-detection systems, and toys.<br><br>With X-Hinges, users can put together an item by selecting and connecting different shapes, then choosing how certain parts move. The item won’t be limited to moving in just one way like previous approaches; instead, users can select up to three “axes:” compressing, up and down, or to the sides. Building a robotic hand, for example, you’d likely want its fingertips to move vertically and laterally. The new tool would then embed sensors where the object bends (in this case, its joints) to capture motion data from both angles. That way, the device can understand and respond to your interactions more dynamically once it’s 3D-printed.<br><br>“It’s kind of like adding a conscious element to everyday objects,” says Jiaji Li, MIT electrical engineering and computer science (EECS) postdoc and CSAIL researcher, who is a senior author on a&nbsp;<a href="http://arxiv.org/abs/2609.11077">paper</a> introducing the work. “What you get is objects that better understand how they’re being used and have more ways to respond. In the same way an AI agent is programmed to make certain decisions based on our instructions, these creations have built-in sensing to adaptively react to human touch.”</p><p dir="ltr">Jiaji Li and lead author Xiang Chang, a TJU PhD student and visiting researcher at CSAIL, used X-Hinges to create many clever devices. For example, they made a glove that can teach a robot how to grasp blocks. It adjusts to the size of a user’s hands, then records how the joints in their fingers bend via sensors before rapidly relaying that information to a humanoid hand to mimic it. The glove could be useful in factories or homes, where a robot often needs to learn new interactions quickly.</p><p dir="ltr">The researchers also produced a small, origami-style lamp using X-Hinges. It senses when a human pinches it closed and adjusts its brightness accordingly. When you turn on this adaptive, portable light source (sort of like a reading light), it looks like a piece of folded-up paper that sprouts into a lamp.</p><p dir="ltr">X-Hinges can even assemble devices that identify which objects have been placed on it. You could put a banana on top, for instance, and it’ll recognize the fruit almost immediately based on the tactile data it’s recorded using its sensors. This advanced physical awareness could be integrated into detection systems in places such as airports and grocery stores.&nbsp;</p><p dir="ltr">Behind each of the scientists’ creations is an easy-to-use interface that helps you design interactive objects from scratch. Users can choose how each part of their design looks and moves, just as the researchers did when making a video game controller that resembles a cartoon shark.</p><p dir="ltr">X-Hinges gives you four shapes to choose from: a custom option, cylinder, rectangle, and thin outline. To recreate the unique look of a shark’s tail, the researchers designed their own shape. They then chose how the toy bent, opting for its tail to move up and down, sort of like a joystick. Finally, the program embedded the sensors, and the scientists 3D-printed their device.</p><p dir="ltr">The researchers say that X-Hinges’ diverse motions are a novel feature compared to past attempts to 3D-print responsive everyday objects. “Most methods only allow for one axis of motion, often due to signal interference. X-Hinges enables your 3D-printed item to move from three angles,” says Chang. “Ultimately, you get a design with the intended movement you want. It also records motion data in each of those areas at the same time, making the device responsive to different kinds of interactions.”</p><p dir="ltr">The tool could help realize the untapped potential of the seemingly mundane. Objects we use every day for a specific purpose, like a water bottle for drinking or a lamp for lighting, could become interactive gadgets. Much like how a chatbot uses a sort of digital intelligence to answer your prompts, an object could leverage a kind of physical intelligence to execute a physical task when users interact with it.</p><p dir="ltr">For now, though, X-Hinges faces a calibration bottleneck. Each time an item is 3D-printed, its layout of electrical elements varies slightly, so two devices that appear identical from the outside may not make the same sensor readings. This is crucial

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

Intel Panther Lake Teardown

Panther Lake debuts the first commercial implementation of a backside power delivery network (BSPDN), introduces Intel’s first iteration of gate-all-around (GAA) transistors, and showcases their advanced packaging capabilities with its Foveros-S assembly. With Panther Lake, Intel’s manufacturing arc has shifted from nebulous roadmaps to shipped silicon, a significant milestone on their long road back to competitive semiconductor manufacturing. To evaluate the extent of Intel’s comeback, we tore down Panther Lake. Intel Core Ultra 7 365 (Panther Lake). Source: SemiAnalysis The SemiAnalysis STEEL teardown lab breaks down advanced datacenter and AI hardware. To learn more about our pipeline or to commission a teardown, contact sales@semianalysis.com . WE’RE HIRING : Architecture, floorplan, packaging, manufacturing, and labs experts. Opportunities from system to transistor and everywhere in between. Check out our Careers page. Our teardown traces 18A from its four-sheet RibbonFETs (Intel’s marketing name for GAAFETs) and gate stacks through contacts, frontside and backside wiring, and the bonded carrier. We explain how these material and integration choices improve gate control and reduce resistance, while adding capacitance, thermal resistance, and process complexity. Our measurements put Panther Lake’s 18A compute logic and TSMC N3E GPU logic at similar logic density. However, 18A does not lead TSMC N3P, N2 or Samsung SF2 in peak density. Panther Lake’s CPU cores are incremental updates, and the high-end GPU still uses TSMC N3E. X-ray of Intel Panther Lake package. Source: SemiAnalysis Panther Lake assembles one compute tile, one GPU tile, and one I/O tile atop a passive base tile using Intel’s Foveros-S advanced packaging. Both compute tile variants use Intel 18A. The Xe3 GPU options are a 4-core GT1 tile on Intel 3 and a larger 12-core GT2 tile on TSMC N3E. Both I/O tile variants use TSMC N6. [1], [2] Panther Lake configurations. PTL-U (3xx, left), PTL-H (3x6H, center), PTL-H with Arc B390/B370 (X-tier, 3x8H, right). Source: SemiAnalysis Our analysis centers on the PTL-U compute tile, both the 4-core and 12-core GPU tiles, as well as the 12-lane I/O tile. PowerVia In conventional chips, power and signal are routed through the same frontside metal stack towards the device frontend. Power rails consume scarce routing resources near the transistors, while tall via stacks carry VDD and VSS from the coarse upper wires to local rails. Backside power delivery network (BSPDN) moves the main power network behind the transistor layer, to the backside, separating it from frontside signal routing. We covered BSPDN and its impacts in 2024. [3], [4], [5] Intel’s BSPDN implementation, branded as “PowerVia”, routes power through dedicated backside metals to nano-TSVs, which connect those rails to local source/drain (S/D) contacts. Schematic orientation of the retained carrier, devices and frontside/backside interconnects. Layer counts are illustrative; not to scale. Source: SemiAnalysis Implementing that separation requires Intel to build the interconnect stacks from both sides of the wafer. The frontside comprises the M0-M14 signal stack, while the backside comprises the BM0-BM5 power stack. M0 and BM0 are closest to the transistors. Generalized PowerVia process sequence and comparison of buried power rails, PowerVia and direct backside contact schemes. Integration details and layer counts are illustrative; the schematic is not to scale. Source: SemiAnalysis The nano-TSVs connect the two sides, but Intel patterns and etches each via from the front after forming the contacts. A narrow via runs from the side of the contact deep into the silicon substrate. Intel then completes the frontside signal metal stack, bonds the wafer to a carrier, flips it and removes the original substrate until the buried via tips are exposed. The backside metal stack is then deposited directly on the revealed vias. The nano-TSV and backside-via profiles taper in opposite directions because Intel forms them from opposite sides of the wafer. The transistor structures form the FEOL. Local contacts and nano-TSVs connect them to the wiring. M0 begins the frontside interconnect stack. The silicon carrier remains attached above the frontside interconnects. It supports the device wafer during substrate removal and backside processing and remains part of the finished chip’s thermal path. PowerVia removes the main power distribution from the congested frontside metals, routing supply through shorter and wider backside wires. Its lateral landing still occupies area in the standard cell, so it recovers less cell area than a direct backside contact. [3] Nano-TSVs beside the logic devices carry VDD or VSS from the backside power network, while signal connections continue upward through the frontside metals. Key Intel 18A RibbonFET and PowerVia components and process innovations Backside Interconnects Samsung SF2 data is included for comparison to Panther Lake’s within this article. SF2 is the incumbent GAA foundry node but lacks BSPDN, serving as a useful reference to evaluate 18A. A full teardown of Samsung’s S26 products, processed on SF2, will be shared soon. Nanosheet-cut EDS comparison. Intel 18A Panther Lake P-core logic (left) and Samsung SF2 Exynos 2600 C1-Ultra logic (right). HFW 550 nm (Intel) and 423 nm (Samsung). Source: SemiAnalysis The PowerVia supply path runs from the backside Cu rails through Mo-lined W nano-TSVs to the local transistor contacts. In this cross section, the tapered connection spans roughly 150 nm from the contact level to BM0. The Ta liner confines Cu and promotes adhesion to the surrounding stack; the AlO ₓ etch stop controls the next dielectric etch above the rail. Dielectric beneath the ribbons electrically separates the devices from the backside wiring and removes the conducting silicon body below the channel. [6] Intel 18A Mo-lined W nano-TSV and Cu BM0 rail. Source: SemiAnalysis AlO ₓ serves as an etc

  • We explain how these material and integration choices improve gate control and reduce resistance, while adding capacitance, thermal resistance, and process complexity.
Nguyên văn nguồn
Chi tiết nguồn
Tác tử

Donating Agent Skills to the Agentic AI Foundation

We’re donating Agent Skills, the open standard that gives AI agents specialized knowledge and workflows, to the Agentic AI Foundation. More than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated Agent Skills since we published the standard in December 2025.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Open call for guest investigators in the life sciences

Today, we’re inviting members of the life sciences community to apply to work with us on special projects.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

LLMs are making formal software verification feasible at a much larger scale

Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases. With AI models, however, this is beginning to change. Here, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.

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

How to Actually Keep Your Company Ahead of AI

Check out Gemini Audio: https://goo.gle/Every-Gemini-Audio Every time a new AI model drops, companies face the same dilemma: Figure out what’s new and how to use it, or focus on getting work done. Every solves this dilemma with a frontier team: a small group that tests new models and turns useful experiments into tools and workflows for the rest of the company. On this week’s episode of _The Every Podcast,_ Natalia Quintero sits down with all seven members of Every’s frontier team to ask what they’re building—from a tabless browser that helps protect your attention to a personal agent orchestrator called Minga—and what they want to try next. Subscribe to Every (free for 14 days): https://every.to/youtube-every-offer?utm_source=podcast&utm_medium=audio&utm_campaign=companyagentepisode Links to resources mentioned in the episode: • Follow Every: https://x.com/every • Compound Writing plugin: https://every.to/guides/compound-writing • Every’s vibe checks: https://every.to/vibe-check • Follow Dan Shipper: https://x.com/danshipper • Follow Natalia Quintero: https://x.com/NataliaZarina • Follow Mike Taylor: https://x.com/hammer_mt • Follow Jack Cheng: https://x.com/jackcheng • Follow Arielle Shipper: https://x.com/ArielleShipper • Follow Katie Parrott: https://x.com/kplikethebird • Follow Jannik Jung: https://x.com/JannikJung 0:00 Introduction: Why every company needs a frontier team 2:42 Why the best frontier teams have pirates and architects 4:57 How Head of Evals Mike Taylor builds personal benchmarks 6:46 How Opus edited an explainer video using Slack messages and meeting notes 10:20 Working with AI means protecting other people’s attention 12:26 Why a frontier team needs a “wrangler” like Arielle Shipper 18:58 Why senior editor Jack Cheng built a tabless browser 24:37 How resident writer Katie Parrott teaches her AI writing plugin who the reader is 30:51 Inside engineer Jannik Jung’s personal AI orchestrator 35:41 Compound engineer Kieran Klaassen uses AI to build power tools

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Southeast Asia’s digital economy to hit $368 billion

<img src="https://storage.googleapis.com/gweb-uniblog-publish-prod/images/e-Conomy_SEA_hero.max-600x600.format-webp.webp">Widespread AI adoption is propelling Southeast Asia's digital economy to a projected $368 billion in 2026 and $1 trillion by 2035.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Physicist Christopher Stubbs joins Anthropic's Long-Term Benefit Trust

Dr. Christopher Stubbs, an experimental physicist and former Dean of Science at Harvard University, has joined Anthropic’s Long-Term Benefit Trust. He brings to the Trust decades of scientific expertise across observational cosmology and precision measurement.

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

Latest open artifacts (#25): DeepSeek V4.1 Flash 1017, American-made models, and new regulatory controls in the US & China

The geopolitics of open-weight AI are accelerating again this fall, as we would expect with the risks of all AI systems (open and closed) rising rapidly. We’ve seen many news items in the last week showing both the U.S. and Chinese governments positioning themselves for some controls on open models. MOFCOM, the Ministry of Commerce of the People’s Republic of China, released a policy proposal (it’s not yet a law) that would require frontier AI models and some other technologies to get pre-approval from the government before they can be shared internationally. We found the reaction to this to be overblown – as the document states that open-source technologies will be set up for automatic approvals – but many in geopolitical circles say this is a big deal because it proposes a regulatory framework that could be used in the future to regulate China’s AI industry. Others we trust on this issue say this proposal could be more about preventing issues like the sale of Chinese startup Manus to Western companies, rather than being about frontier AI models. Either way, the wheels of control are moving around AI, and open-weight models are not free from any influence. The Department of Homeland Security (DHS) released an order on DeepSeek , which restricts DeepSeek itself, as a company, and DeepSeek’s models, as a technology platform, from being a vendor in federal civilian procurement. It is not clear how this will relate to finetuned models or models trained on the outputs of DeepSeek (e.g., Nvidia’s Nemotron models have long relied heavily on DeepSeek models in their training recipes – will they be banned?), but this is an expected action and reinforces the need for a domestic open model ecosystem. Other model makers in China, such as Zhipu AI, makers of GLM, are already on the Entity List under Commerce/BIS, which restricts their access to American technology. DeepSeek followed the above news up by releasing their internal inference stack publicly, DeepInfer , which is likely far more efficient than existing open-source tools like vLLM/SGLang for their models. Estimates point to DeepSeek making up to 90% gross margin when selling tokens on this inference stack. It is well-known that the top AI labs have much more inference-efficient tools internally, and this is the first time we’ve seen a leading lab share them. Inference companies will be quick to adopt these tools, which will drive prices for DeepSeek’s models down across the ecosystem (or margins up). While not an open model or dataset, we consider this release to be one of the bigger open-source AI artifacts in the last few months. Share As of writing this, Kimi K3.1 is rumored to be released very soon, with metadata associated with it starting to appear in public libraries . Exciting times! You can view the models in this email, and all previous issues, on our Artifacts Hub. It includes commentary and other ecosystem metadata. Browse the Artifacts Hub Our Picks MiMo-V2.6-Pro-RL by XiaomiMiMo : an update for Mimo, which reuses the same bases as MiMo V2.5, but used a lot more resources on RL training. They also release an in-depth tech report , as well as some of the used environments, which is rare at this scale. The final model from this recipe, V2.6-Pro, is the top open model on Artificial Analysis’s ranking today, and the scaled RL runs came with a fun live dashboard of training. We’ve been fans of Xiaomi for months now, and their efforts are bearing fruit. DeepSeek-V4.1-Flash-1017 by deepseek-ai : Since the latest Artifacts, the whale has not released one, but two new model iterations with a new size (552B-A16B) which also support image inputs. The model has gained a lot of (well-deserved) attention for its performance at a small price. It also carries the MIT license, continuing the trend established by DeepSeek themselves. This is notable in times where licenses seem to become more bespoke and exclusionary. Below is the Pareto of performance vs price on DeepSWE, a favorite agentic software engineering benchmark today, which DeepSeek dominates. clef by Cloudflare : Decision models are all the rage after being introduced by the startup Typesafe . In short, they are general classification models that output few tokens with their probabilities. Its usefulness has led to quick adoption and also a lot of open model finetunes such as Clef. NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 by nvidia : The mid-sized model of the Nemotron family got its long awaited update. It is strong for its size and can be run at home (given you own a beefy laptop or RTX Spark). As usual, NVIDIA releases the vast majority of the data and an in-depth tech report . Beam-1.0-501B-A23B by reflection : The long awaited open model by Reflection, making it the strongest open American model released under a permissive license (Apache 2.0). While not the smartest open model available, Beam is very impressively token efficient (second to only GPT-6-Astra below). This is a sign of a well-developed post-training stack. Models General Purpose Ternary-Bonsai-2-27B-gguf by prism-ml : A ternary version of Qwen3.8-27B for on-device deployments. Ornith-1.5-397B by ornith-ai : A series of Qwen3.5 and Gemma 4 finetunes that gained attention in the local AI community for its strong performance over the respective (strong) baselines. Ling-3.0-flash-VL-fp4 by inclusionAI : A 124B-A5.5B model by Ant Ling (within Ant Group), which continues to churn out models consistently. Muse-Spark-Open-816B-A42B by meta-models : META has released their open version of Muse, which we have written about in more detail ! Read more

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

Voxtral 3

Voxtral 3: Mistral's voice agent SDK for real-time speech in 50+ languages, accurate transcription, emotional TTS, deep tool use.

Nguyên văn nguồn
Chi tiết nguồn
561–580 / 738
Trang 29 / 37