AllyshipContent Radar

News 2839

Chủ đề cộng đồng đang thảo luận

Crawl live-1795104900000 · 23:15 19 thg 11, 2026 · Xuất bản 23:31 19 thg 11, 2026

109 / 2839 chủ đề đạt ngưỡng 55 trong khoảng này.

1

The Pi-adapted Qwen 3.8 27B actually works. Here’s the proof.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

aartiles24 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

The Pi-adapted Qwen 3.8 27B actually works. Here’s the proof. Lots of people have been saying the harness-hub Pi adapter for Qwen 3.8 27B is great. Most of what I found was benchmark scores and tok/s, with little about the actual tasks behind them. So I gave it a real task I had already run with the base 27B on my MacBook (M5 Pro, 64 GB): a full raycaster FPS from one brief (prompt by @ddalcu, shared by @Beamsters1), built by Tenaz + Pi with no human in the loop. The base 27B, 4 attempts: ❌ 2 blocked before writing a single line of code ❌ 2 stopped after ~4 h with 2 tasks done and no playable game It kept burning its whole output budget on reasoning. The pi-adapted 27B (OptiQ 4-bit on Splash, thinking off): ✅ COMPLETE in 2 h 3 min: 34 sessions, 0 failures, 0 human input 🎮 All 8 must-have features work when you play it ⚡ ~52 tok/s, 0.38 s median TTFT, 96% prompt cache 🧠 Memory pressure normal the whole time, with Chrome and Claude open Same brief, same laptop (different engine and harness too, details in the link): from no playable game to a finished one. The adapter does what it claims: short turns, no thinking, real tool use. Qwen 3.8 Flash Next is still the smarter model, but it needed ~7 h for this game. The adapted 27B got there in about 2, and left the Mac usable. That makes it the local model I’d hand to a frontier model as its worker. Every attempt, builds and numbers side by side: https://aartiles.github.io/local-model-benchmarks-site/challenges/raycaster-shooter/ Media

2

The Max Agency podcast with @hwchase17 is back for a second season!

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

langchain 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

The Max Agency podcast with @hwchase17 is back for a second season! Watch Episode 1 with @clay’s Head of AI @jeffbarg on your favorite platform: ⏯️ YouTube: https://www.youtube.com/watch?v=efBlbQahwng 🎧 Spotify: https://open.spotify.com/episode/2qVxd4mCdmMrbFUwkxvOHC 🎧 Apple Podcasts: https://podcasts.apple.com/us/podcast/inside-clays-gtm-agent-rebuild/id1891551672?i=1000802949274 Media

4

Media

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Anthropic 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Media

  • Media X · 19 thg 11, 2026
5

fun fact:

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

hwchase17 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

fun fact: as we were gearing up towards launch, we had our content agent take a stab at making a launch video we didn't end up using it - but it was really good and we definitely could have! attaching below this agent is Open Source as well! https://github.com/langchain-ai/design-agent Media

6

Ask HN: When was the last time you thought "that's bad API design"?

60

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · 4 tương tác đã đo · Nguồn crawl

Mitchem 1 bài viết · HN: LLM
Nguyên văn nguồn
Chi tiết nguồn

I used to be irrationally impressed when I saw good API design and docs. Now I can't remember the last time I sat down and spent meaningful time reading docs for a platform. I've just gotten in the habit of pointing Claude at them or using a service like Context7 for the past year or so. As a consumer, I've realized that I'm not even sure which services I use that have good or badly designed APIs. If you maintain an API, have you changed how you write docs now that the main reader is probably an LLM?

7

Show HN: Openspot – macOS launcher for disposable tools

61

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · 5 tương tác đã đo · Nguồn crawl

tniich 1 bài viết · HN: LLM
Nguyên văn nguồn
Chi tiết nguồn

Hi HN! We're Tony, Kolom and Anastasiia, a small team from Ukraine. Today we are open-sourcing Openspot, our research project on using generative UI in day-to-day work. The idea is simple: generate a tool (spot) for your current task using coding agent, and keep it if you like it or remove it after use if you don't. We are using Openspot daily and really enjoying the possibilities it unlocks. Some examples of spots: - caffeinate, which allows the Mac to work with the lid closed - local file and screenshot search in natural language using EmbeddingGemma 2 - draw on screen during calls or before screenshots - AI agents monitoring - AirDrop, but over the network (using iroh) We built it because of two things: 1. Over the past year, we (and almost every designer and engineer we know) have started making small tools using coding agents. But the DX for macOS applications is very frustrating even for coders, and you have to pay for a developer account to share them. 2. And we kind of miss the days of custom Winamp skins and Twitter clients. That's why every part of Openspot is customizable, so anyone can give it their own look and feel. This is a very early stage. The code is MIT-licensed, and the core will remain free forever. We're sticking to macOS 26+ for now (because of Liquid Glass). Ask us anything, or roast it. Any feedback is very appreciated.

8

At Kong we run half a quadrillion AI tokens a year, and I kept seeing agents stuck on clouds built for websites.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

subnetmarco 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

At Kong we run half a quadrillion AI tokens a year, and I kept seeing agents stuck on clouds built for websites. So we built a cloud for agents. Volcano is live today: the self-healing agent cloud. Sandboxes, durable workflows, Postgres, hosting. Media

9

Great paper from Meta on ML engineering agents, tested with a large-scale A/B test on production search rankers.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

dair_ai 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Great paper from Meta on ML engineering agents, tested with a large-scale A/B test on production search rankers. (bookmark it) They also show how important the harness design is. The agents propose a model change, implement it, train it, and evaluate it. Engineers keep control of the objective, the search direction, and the final review. In one campaign, the system completed 1,097 experiments and accepted 325 of them. That is a 29.6% acceptance rate, against 15.6% in an earlier campaign with a different objective. The best model raised search volume by 0.23% in the live test. They share that three problems made this hard. Each experiment takes hours, offline metrics often disagree with online impact, and gains flatten after the early wins. The team answered with parallel training runs, an objective calibrated to online impact, and careful context engineering. Several accepted changes were architectures the team had never tried. The authors also report that a stronger model was not enough on its own. The loop around it mattered just as much. If you plan to put agents on your training pipeline, calibrate the objective first. https://arxiv.org/abs/2611.23910 Chat with Paper: https://academy.dair.ai/papers/mle-agents-autonomous-ml-engineering-in-large-scale-recommendation-systems-2611.23910

10

Rant: Gemini 4 does not make any sense

62

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · 9 tương tác đã đo · Nguồn crawl

sambejk 1 bài viết · HN: GPT
Nguyên văn nguồn
Chi tiết nguồn

On November 18th, Google released another model, this one called Neon. Making it the fourth model in just 2 months. OpenAI makes sense: Luna-Terra-Sol-Astra. Moon is smaller than earth, earth is smaller than sun and I can forgive them for not knowing that sun is also a star. Anthropic is even better: Haiku-Sonnet-Opus-Mythos/Fable. Haiku is an extremely short poem, Sonnet is a bit longer, Opus is very long and Mythos is extremely long. Ok, Fable does not exactly really fit in there, but they came up with it after they only had the three models, so we had time to adjust. Also, Claude told me: Fable is basically Mythos with extra safety, and a fable is a story with a moral. Now, I want you to think back to those high school chemistry classes. Which is bigger neon or carbon or hydrogen or argon? Impossible right? The correct answer, hydrogen, carbon, neon, argon, doesn't even match with model capabilities. To make things worse, there is no consistency, they are combining noble gases with other non-metals. You can't even say they went for easy names, helium, is not in there as an example. They really should have gone with just noble gasses: helium, neon, argon and xenon (skipping krypton to not confuse with kryptonite) I'm glad that Google is sort of back in the race, but they should really invest in someone who is not AI-pilled to come up with these names. PS don't get me started on Meta, wtf is a glimmer?

12

Neon prompt:

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

fofrai 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Neon prompt: > Wow that monstera plant is cool, can we give it a face, turn it into a character, show it walking on its leaves, jumping, waving, etc? 😍 Media

13

GUARD: Stabilizing Recursive Self-Improvement with Regret Guidance

71

2 tiếng nói · chưa gắn nhãn trưởng lab · 2 nền tảng · chưa có số tương tác · Nguồn crawl

Minghe Zhang, Xuetao Kang, Yifei Ming, Yanfei Chen, Han Yu, Yufan Zhuang, Chen-Yu Lee, omarsar0 1 bài viết · arXiv cs.AI, X
Nguyên văn nguồn
Chi tiết nguồn

arXiv:2611.23476v1 Announce Type: new Abstract: Autonomous AI agents pair a frozen base model with an editable harness of prompts, skills, tool rules, and memory. Recursive self-improvement (RSI) adapts this harness from experience without weight updates, yet self-editing without external graders often backfires: reflection and scalar scores conflate what an edit newly broke with what remains unsolved, causing local regressions, skill forgetting, and memory bloat. We introduce GUARD (Grader-free Unsupervised Environment Design [UED] with Adaptive Rule Distillation), a closed-loop framework for reference-free continual harness evolution. Our key insight is to decompose each task into pre-execution self-checks and guide harness updates via directional regret. Each round grades trajectories against these checks, pinning mastered rules while prioritizing newly broken behaviors, and selects among parallel candidate edits by critic-free group-relative fitness. A Soft-HURT gate then promotes an edit only when gains outweigh regressions, scoping specialized fixes to prevent over-triggering and guaranteeing non-regression on its gated buffer. At test time, one frozen harness runs in a single pass. Across benchmarks on memory, skill evolution, continual task streams, and domain reasoning, GUARD consistently improves performance without catastrophic forgetting or memory pollution while substantially cutting deployment token usage.

16

6 分钟短片:针头里的三千个星系

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

xiaohu 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

6 分钟短片:针头里的三千个星系 用 Claude 和即梦的 CLI,制作了一部 6 分 46 秒的纪录短片,讲 1995 年哈勃望远镜盯着一块空天看了十天,拍出三千个星系的事 为什么选这个故事,因为这正是人类为了看清宇宙一点一点努力的过程见证 为了能观测幽黑的宇宙,人类一直不懈努力这也是 AI 视频现在努力的方向,还原真实的世界 Media

17

实测即梦 Seedance 2.5 4K

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

xiaohu 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

实测即梦 Seedance 2.5 4K 确实很强.. 确实非常的高清... AI 生成视频直接进入了超清领域 我这两天用它做了 6 部片子: 一部 6 分 46 秒的哈勃纪录片加 5 条 30 秒短片有品牌广告、情绪表演、MV,还有两条让“真人”小互来演... 用下来最大的感受是质感:金属有层层的反光,布能看到织纹,皮肤有毛孔,水珠是一颗一颗圆的 画面不再是 AI 那种干净但发虚的"塑料感",更像真拿摄影机拍出来的 这个广告TVC 片非常的高清,以后广告完全是可以 AI 生成了 Media

18

Le Chonk in London! 🇬🇧

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Mistral AI 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Le Chonk in London! 🇬🇧 For the first time since our worldwide hackathon, join us for a special Le Chonk hackathon! 📅 Saturday, Nov 21 📍 London📷 https://luma.com/le-chonk-in-london

19

Season’s Streamings and Savings: Black Friday Deals Come to GeForce NOW

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

GeForce NOW Community 1 bài viết · NVIDIA Blog
Nguyên văn nguồn
Chi tiết nguồn

GeForce NOW is kicking off Black Friday with more reasons to make time for gaming. Six-month memberships are back for a limited time. Save 30% on the first six-month term of a Performance or Ultimate membership . Big gaming plans, smaller price tag. Keep the holiday savings going: Check out more NVIDIA holiday deals for gamers, creators and AI enthusiasts. The living room gets a little more playtime, too. GeForce NOW arrives on Amazon Fire TV Stick HD, Fire TV Stick 4K Select and Fire TV Stick 4K (2026 model) running Vega OS, alongside an Amazon offer launching later today for 90% off the first month of an eligible Performance or Ultimate membership in participating countries. It brings console-style gaming to the TV, straight from the Amazon cart. Bethesda fans have more in store, with five favorites adding GOG support as part of nine games joining GeForce NOW this week. Big RTX Power, Black Friday Price Thousands of games are calling from the cloud. Six-month Performance and Ultimate memberships are back for Black Friday, with 30% off the first six months for new memberships and eligible upgrades in select regions. The sale runs through Monday, Nov. 30, while supplies last. Go Ultimate for GeForce RTX 5080-class performance across PCs, Macs, Steam Deck and other supported handhelds, browsers, mobile devices, compatible TVs and more . Sessions lasting up to eight hours and first-priority server access leave more room for marathon play. Ultimate brings the latest NVIDIA innovations to the cloud, including DLSS 5 technology with the 3D-Guided Neural Rendering feature for lifelike lighting and materials in NBA 2K27 and Hogwarts Legacy , with support coming to War Thunder and Where Winds Meet . The experience goes big with streaming at up to 5K resolution or up to 360 frames per second, plus high dynamic range, ultrawide support and saved in-game settings. DLSS Super Resolution, Multi Frame Generation and Ray Reconstruction enhance supported games, including the path-traced missions of 007 First Light and the reality-bending action of CONTROL Resonant , while NVIDIA Reflex and Cloud G-SYNC technologies help keep gameplay responsive and smooth. GeForce NOW handles downloads, patches and updates for Ready-to-Play games, freeing local storage and leaving more time for gaming. Choose a membership and start the season streaming. More Screens, Major Savings Press play on more possibilities. GeForce NOW officially expands to all three of Amazon’s Vega OS streaming media players including Fire TV Stick HD, Fire TV Stick 4K Select and Fire TV Stick 4K (2026 model) bringing thousands of supported games to more another screens in the home. Vega OS keeps Fire TV fast and responsive, so launching GeForce NOW and jumping into games feels quick and snappy, with no console required. Getting started is simple: download the GeForce NOW app on a supported Fire TV device , sign in and connect a compatible controller. From cozy stories to epic adventures, turn-based tactics to co-op challenges, there’s plenty to build, explore and share with friends — all from the comfort of the couch. There’s a fire deal to get the games going, too. Launching later today through Thursday, Dec. 31, get 90% off the first month of a Performance or Ultimate membership purchased directly through Amazon on supported Fire TV devices in participating regions. The cloud is ready. The next quest is calling. GOG Big or Go Home Five favorite worlds. Extra ways to play. Five Bethesda titles expand their GeForce NOW support this week with their GOG versions: Indiana Jones and the Great Circle , Fallout 4 , The Elder Scrolls V: Skyrim , DOOM and DOOM Eternal . Uncover an ancient mystery in Indiana Jones and the Great Circle , a first-person adventure set between Raiders of the Lost Ark and The Last Crusade . Explore the wasteland in Fallout 4 or trade the Pip-Boy for dragons and open-world fantasy in The Elder Scrolls V: Skyrim . For a faster change of pace, DOOM brings relentless enemies, destructive weapons and fast-moving first-person combat, while its sequel, DOOM Eternal , turns up the intensity as the Slayer takes on demonic forces. Members who own the newly supported GOG versions can stream those titles through GeForce NOW without purchasing them again. From treasure hunts to dragon hunts, the collection can follow across nearly any supported device. Familiar worlds, with a little more room to roam. Members can look for the following games joining the GeForce NOW library this week: Volvy’s Adventure: Reslimed (New release on Steam , Nov. 19) Baldur’s Gate 3 ( Xbox Play Anywhere) DOOM (2016) ( GOG ) DOOM Eternal ( GOG ) Fallout 4 ( GOG ) Graveyard Keeper 2 ( Epic Games Store and GOG ) Indiana Jones and the Great Circle ( GOG ) Moon Sweepers ( Steam ) The Elder Scrolls V: Skyrim Anniversary Edition ( GOG ) Dates listed above reflect when games are released on their respective stores. GeForce NOW availability may vary, as games are onboarded after release and added throughout the week. Keep an eye on GeForce NOW channels and GFN Thursdays for availability updates on announced titles. Make it a cloud gaming weekend. Grab a day pass to try Performance or Ultimate before committing to a monthly membership, with the day pass cost applicable toward that first month. What’s on the big-screen gaming lineup this weekend? Let us know on X or in the comments below.

20

Study reveals data centers’ long-distance damage to air quality

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Adam Zewe | MIT News 1 bài viết · MIT News: AI
Nguyên văn nguồn
Chi tiết nguồn

<p>The rapid expansion of data centers for artificial intelligence causes negative health and air quality impacts in regions far from where the brick-and-mortar facilities are located, according to a new study by MIT researchers.</p><p>The team estimates that additional air pollution from the electricity demands of data centers in 2023 resulted in about 650 premature deaths across the United States.&nbsp;</p><p>Some Midwestern states had among the highest estimated rates of premature deaths per capita attributable to data centers, even though much of the data center electricity demand was concentrated in Virginia and Texas.</p><p>The researchers built a modeling framework that captures how the electricity demanded by data centers propagates through the nation’s interconnected power grid. The model also considers how atmospheric conditions impact the formation of air pollution in different regions.</p><p>The framework&nbsp;identifies power plants across the country whose emissions increase in response to data center electricity demand. In parts of the Midwest, these emissions combine with atmospheric conditions that favor the formation of fine particulate matter, a pollutant that can cause heart attacks, stroke, and lung disease.</p><p>The researchers say these results underscore the importance of considering unintended regional health impacts of data centers, and how the location and operations of these facilities affect public health on a broader scale.</p><p>“When people think about the negative impacts of data centers, they usually focus on the neighbors. But our study indicates that the air quality impacts of data centers are very widespread across a large region. Even if you don’t think you are affected because there is not a data center near you, think again,” says Noelle Selin, a professor in the MIT Institute for Data, Systems, and Society (IDSS) and the Department of Earth, Atmospheric and Planetary Sciences (EAPS); director of the Center for Sustainability Science and Strategy; and senior author of the study.</p><p>Selin is joined on the paper by Yuang (Albert) Chen, an MIT graduate student; as well as others at MIT, the University of California at Irvine, and Yale University. The research <a href="https://iopscience.iop.org/article/10.1088/1748-9326/aebbd4" target="_blank">appears today in <em>Environmental Research Letters</em></a>.</p><p><strong>Capturing regional impacts</strong></p><p>The Lawrence Berkeley National Laboratory&nbsp;<a href="https://escholarship.org/uc/item/32d6m0d1" target="_blank">estimated</a> that electricity demands of data centers represented about 4 percent of total U.S. energy demand in 2023 and resulted in about 4.4 percent of the nation’s overall emissions.&nbsp;</p><p>That proportion has likely increased as new data centers continue to be built at a rapid pace.</p><p>Many prior studies&nbsp;estimate impacts using average electricity-grid emissions. Like those studies, the researchers assume data centers draw power to run computing tasks 24/7.&nbsp;Their approach also identifies which generators respond to that additional demand hour by hour, allowing them to trace where the resulting emissions and health impacts occur, Chen explains.</p><p>“Generators that respond to additional electricity demand are often natural gas plants, although the response varies by place and hour,” he says.</p><p>To drill down on regional health impacts, the MIT researchers developed a modeling approach that enables them to identify which generators are contributing to data center electricity demand within each modeled region.&nbsp;</p><p>This allows them to incorporate a spatial component into the analysis.</p><p>They built an electricity system model that simulates how much electricity each generator produces and pumps into the grid every hour. This model also considers how that electricity is transmitted between regions.&nbsp;</p><p>It gives the researchers data on emissions of nitrogen oxides (NOx) and sulfur dioxide (SO<sub>2</sub>) for each generator across the United States. NOx contributes to the formation of ground-level ozone, or smog, while both NOx and SO<sub>2</sub> contribute to the formation of fine particulate matter (PM<sub>2.5</sub>) in the atmosphere.</p><p>“It was a challenge to gather all the data,” Chen says. “There are more than 10,000 generators in the U.S. grid, and we needed to make sure the data provided for each generator is valid and validated across other data sources.”</p><p>The researchers incorporated these data into a chemical transport model. This model simulates how pollutants form and travel through the atmosphere across regions.&nbsp;</p><p>From there, they translated the additional air pollution from data center operations into premature mortality rates using evidence from epidemiological literature.</p><p>The model estimates that emissions from data centers caused about 650 premature deaths in 2023 (513 from PM<sub>2.5</sub> and 137 from smog), which represents about 5.3 percent of the nation’s total electricity-related mortality.&nbsp;</p><p>“While it is not surprising that any additional source of air pollution will result in mortalities, with data centers, when and where these deaths are occurring are different than you might expect,” Selin says.</p><p><strong>A pollution mismatch</strong></p><p>The premature mortality rates were highest in Iowa, Nevada, South Dakota, and Kansas, but the bulk of the electricity demand was concentrated in northern Virginia and in Texas. Premature mortality rates in Virginia and Texas are only about half as much as most Midwestern states.</p><p>One reason for this mismatch is the interconnected nature of the power grid, Chen says.&nbsp;</p><p>Generators that respond to the increased electricity demand in Virginia are likely located in the Ohio Valley, where there is a higher proportion of older power plants that generate more emissions from burning coal and natural gas.&nbsp;</p><p>“Different regio

1–20 / 109
Trang 1 / 6