AllyshipContent Radar

Hands-on 278

Công cụ, thử nghiệm và hướng dẫn thực hành

Crawl live-1795090500000 · 19:15 19 thg 11, 2026 · Xuất bản 19:35 19 thg 11, 2026

163

Beyond matching DeepInfer's performance, we also put vLLM to the test on @SemiAnalysis_ AgentX with real-world agentic workloads.

8.9

Hướng dẫn vllm_project

Beyond matching DeepInfer's performance, we also put vLLM to the test on @SemiAnalysis_ AgentX with real-world agentic workloads. At DeepSeek's off-peak pricing, vLLM delivers $38.96/GPU-hour revenue, with 954K tokens/s/GPU total throughput (cached+input+output) at 196 tokens/s P90 per user. This is the best DeepSeek-V4.1-Flash result on AgentX ever, powered by recent advancement in vLLM! Check out our blog for the full optimization story and how to reproduce the results 👇 Enjoy the read.

X · 25 thg 10, 2026

Nguyên văn nguồn
Chi tiết nguồn
164

At MIT, Martha Minow speaks on sustaining democracy, both by using and overcoming AI

6.7

Hướng dẫn Peter Dizikes | MIT News

<p>Legal scholar Martha Minow delivered this fall’s Compton Lecture at MIT on Thursday, examining the civic foundations of democracy — and describing how artificial intelligence is undermining it, but could be used to help reinforce traditional U.S. governance.&nbsp;</p><p>“It’s not a good time for democracy or constitutionalism,” said Minow, who is the 300th Anniversary University Professor at Harvard University, and a prominent expert in constitutional law and human rights. She noted a litany of challenges to democracy in recent years, including violations of laws and norms, the stifling of free speech, a diminution of congressional authority, disregard of legal rulings, the common vilification of others in society, and more.&nbsp;</p><p>Recent Gallup polling, Minow observed, shows 77 percent of people saying that the founders of the U.S. would be disappointed in the contemporary condition of the country.&nbsp;</p><p>“Many people genuinely wonder whether this nation’s system of government and rule of law, ignited 250 years ago, will survive,” Minow said.&nbsp;&nbsp;</p><p>Her remarks assessed the state of the U.S. in terms of what she called four key “ingredients” of democracy: education, reliable and accessible information, security from violence, and fairness. Minow evaluated how AI has, among other things, influenced these areas of life — but also suggested ways people could use contemporary technology to help.&nbsp;</p><p>“It’s very familiar to hear blame attributed to digital technologies and experts of all stripes, but a turnaround, I believe, is genuinely possible if people at MIT and elsewhere pour time and talent towards real and much-needed solutions,” Minow said.&nbsp;</p><p>A substantial portion of her remarks focused on the pitfalls of AI in education. That issue compounds some other troubling trends: The U.S. already has large local and regional disparities in educational funding, Minow noted, while schools have actually become significantly more segregated in economic and ethnic terms over the last three decades.&nbsp;</p><p>In this context, AI threatens to further weaken independent-thinking skills, Minow suggested, just when greater understanding of civic issues is absolutely vital.&nbsp;</p><p>“As AI surges, I think we have to learn to be smarter about how to use it, how to regulate it, and how to actually educate people about the costs of mindless using,” Minow said. She added: “Dominant AI platforms supercharge the strategies that are deployed by social media companies, inducing passivity in people … Leading AI models and interfaces and sycophantic elements endanger the development particularly of young people’s capabilities to think, to remember, and to engage in critical thinking.”&nbsp;</p><p>That trend away from cognitive effort, she suggested, can be changed — perhaps through AI tools that act more like tutors and are designed to enhance people’s deep thinking.</p><p>“Existing LLMs, model training, and interface design can offer nudges and questions and hints rather than replacing student work, but there’s so much more that could be developed, and more effectively, to be trustworthy, and to deepen human activities and curiosity, rather than replacing people,” Minow said.</p><p>To the audience, she added, “AI tools could do much more of that … And I believe that you, the people here, could develop better tools than the ones that are out there. And make better assessments, more meaningful and cheating-resistant.”&nbsp;</p><p>Minow’s talk, “Constituting Democracy in the Era of AI,” was delivered to a large audience in MIT’s Huntington Hall, Room 10-250 on campus.</p><p>MIT President Sally A. Kornbluth delivered introductory remarks to the audience.&nbsp;The&nbsp;Karl Taylor Compton Lecture Series, introduced in 1957, is the highest-profile event of its kind at MIT. Karl Taylor Compton was MIT’s ninth president, serving from 1930 to 1948, and then becoming chair of the MIT Corporation from 1948 to 1954.</p><p>“He guided MIT through the Great Depression and World War II,” Kornbluth said, “transforming the Institute from an outstanding school for training job-ready engineers to a great global research university.”&nbsp;</p><p>She added: “He was also a deeply committed patriot who helped harness the intellectual power of America’s research universities to serve the nation during World War II.”&nbsp;</p><p>Kornbluth called Minow “a legal scholar of extraordinary intellectual scope,” and a “dedicated educator whose work has shaped how we think about the law and our own humanity, as individuals and as a society.”&nbsp;</p><p>Minow has been a faculty member at Harvard Law School since 1981, and served as dean of the school from 2009 through 2017. Minow earned her BA in history from the University of Michigan, an MA in education from Harvard, and a JD from Yale University.</p><p>Minow’s books include “Making All the Difference” (1990), “Partners Not Rivals” (2002), and “When Should Law Forgive?” (2020).&nbsp;</p><p>In her talk, Minow also discussed the civic effects of misinformation, threats of violence, and multiple dimension of societal fairness. Modern media technology has led to a 75 percent reduction in journalists in recent decades, Minow observed, with an attendant rise in corruption in places without local media. Civic-minded technologists, she proposed, should work towards establishing a widely-used “information commons,” or many forms of them, to help people access more reliable information.&nbsp;</p><p>Meanwhile, when it comes to basic security in society, Minow warned, “We’re on the brink right now. I know a lot of judges, and every one of them has a credible threat of violence against them. In America. This is not acceptable.”&nbsp;</p><p>For all the challenges to democracy and traditional civic norms, however, Minow sounded positive notes throughout her remarks, pointing out that there are many opportunities to create new forms of reliable information, educa

MIT News: AI · 24 thg 10, 2026

Nguyên văn nguồn
Chi tiết nguồn
165

The MOST obvious (but REALLY good) STARTUP ideas right now:

6.6

Hướng dẫn gregisenberg

The MOST obvious (but REALLY good) STARTUP ideas right now: 1. Zestimate for company data (what your data is worth to an AI lab) 2. Google Analytics for agents (see which agents visit your site and what they want) 3. Y Combinator for 1 person companies (back solo founders running a team of agents) 4. Okta for agents (who each agent is and what it's allowed to touch) 5. Marie Kondo for your company's data (clean it up so agents can use it) 6. FICO for agents (a trust score before an agent can spend money) 7. Geico for agents (insurance for when an agent makes an expensive mistake) 8. Datadog for agents (see every step your agents took) 9. Twilio for agents (phone numbers and texting for agents) 10. Upwork for agents (hire an agent for a job instead of a freelancer) 11. ESPN for AI (every model launch covered like a big game). 12. Shutterstock for robot training videos (get paid to film yourself doing chores) 13. Credit Karma for what AI knows about you (see it, fix it, delete it) 14. Norton for AI scams (protect your parents from fake voices and emails) 15. PayPal Buyer Protection for agent purchases (fixes it when your agent buys the wrong thing) 16. Geek Squad for AI (set up personal agents for families and small businesses) 17. Dell for private AI (AI in a box for firms that won't use the cloud) 18. Carfax for used GPUs (a health report before you buy secondhand) 19. Airbnb for idle GPUs (rent out computing power you're not using) 20. Rotten Tomatoes for AI models (real reviews instead of benchmarks) 21. G2 for skills (find the best skills for your agent) 22. Dropbox for your AI memory (take what your agent knows about you to any agent) 23. Canva for physical products (design it with AI and get it made) 24. ADT for your personal agent (alerts when your agent does something it shouldn't) 25. Hertz for robots (rent one for a move, a deep clean or an event) 26. Harvard Business School for internet businesses (a real education in building online) 27. Workday for agents (onboard, manage and offboard your digital workers) 28. Getty Images for your face and voice (license your likeness to AI safely) 29. Meta Muse for older adults (and other high LTV niche) 30. Google Ads for agents (pay to be what the agent recommends) 31. Loom for agents (a walkthrough of everything your agent did) 32. Plaid for agents (approved access to banks, airlines and stores) 33. Jiffy Lube for vibe coded apps (quick fixes when your AI built app breaks) 34. Ancestry for datasets (where your training data came from) 35. The organic label for human-made content (proof a person made it) 36. Strava for builders (track what you ship every week) 37. Honey that haggles for you (your agent negotiates the price) 38. SEM Rush for personal agent platforms (like Muse etc) 39. Frequent flyer miles for agents (loyalty programs built for agents)

X · 24 thg 10, 2026

Nguyên văn nguồn
Chi tiết nguồn
166

Amazon’s Trainium Takes a Strike at the Training King

6.6

Hướng dẫn Kimbo Chen

Trainium has already proven itself in the realm of inference. Anthropic is running production inference and RL rollouts on Trainium2 and Trainium3. Historically, Trainium has been great for inference but not the best fit for research training. In inference, the model architecture changes at a much slower pace compared to training, where researchers iterate on many different experiments and ablations of model architecture and need a stack that covers a broad API surface. In the past, that is where Trainium fell short. The AWS Neuron team is aggressively working to change this pitfall with the biggest shift being moving towards a first-class, native PyTorch training experience. Gone are the days of sprinkling xm.step() through your training loop and wrestling with a bespoke XLA-style execution model. Trainium now runs PyTorch the way PyTorch users expect, in eager mode, with torch.compile and the standard APIs. The SemiAnalysis engineering team has been trying out the Trainium3 racks for the past couple of weeks and has found the experience to be great – we love the overall direction of where it is heading. To be clear, it isn’t perfect yet, and we have personally run into a couple of bugs ourselves. It’s nowhere close to the maturity of the CUDA ecosystem, but the direction is right. Their team is making rapid progress with the help of agentic coding. Furthermore, unlike other accelerator companies, where we had to fight tooth and nail to get their engineers a reasonably stable CI fleet, Trainium uses up to 80,000 physical cores allocated towards automated quality control. In this article, we walk through our experience on the Trainium3 TorchTitan and TorchNeuron stack, along with our experience of Trainium’s Neuron Explorer profiler, which is way better than both Nvidia’s and AMD’s profilers. We created a bunch of demos to show how Trainium’s profiler is “GPU Impossible.” Visualizing Nafea Bshara, VP at AWS In Trainium Profiler Showcasing Trainium’s Finegrain Profiler. Source: SemiAnalysis In an upcoming article, we will go deep on writing speed-of-light NKI kernels and microbenchmarking on Trainium3. For our deep dive into the Trainium hardware, check out our previous article where we do a complete hardware breakdown. Source: SemiAnalysis We Are Hiring SemiAnalysis is hiring for various Member of Technical Staff roles. If you find this type of analysis interesting, apply below: Apply Here What we’re looking for: Highly motivated & skilled Member of Technical Staff to join our growing engineering team. Member of Technical Staff will play a crucial role in developing inference benchmarks & system modeling You will also play a role in writing and preparing our newsletter articles for publishing, with the opportunity for direct authorship recognition. We are hiring individual contributors at all experience levels. Compensation will be competitive As part of the interview process, you’ll complete a paid coding challenge designed to reflect typical daily tasks at SemiAnalysis Competitive Compensation depending on experience, skillset, location & business needs Qualifications Proactive, self-motivated, and capable of working independently in a global team. Demonstrated experience in ML frameworks such as PyTorch or JAX through professional experience, personal projects, or personal Substack blogs Solid understanding of at least 1 of the following: transformer architecture, LLM parallelism strategies, and/or CUDA parallel programming Strong research skills and the ability to synthesize information from various sources to draw insights. English language proficiency required, Chinese language proficiency preferred Moving Towards a First-Class PyTorch Experience Amazon has made a major course correction to its software strategy, and we believe it will meaningfully broaden Trainium3 adoption. TorchNeuron represents a native PyTorch experience instead of duct-taping Trainium onto PyTorch through the PyTorch/XLA project. The new backend is aimed at everyone who isn’t a frontier-lab kernel engineer. AWS has even hired the lead of Google’s native Torch on TPU effort to work on native Torch for Trainium. He was the TPU SW lead from the start of the program at Google, including XLA, xprof, Cloud TPU, etc. The old approach made Trainium feel foreign to anyone coming from CUDA. PyTorch/XLA relied on lazy tensor graph capture and had no first-class eager mode. It also didn’t support PyTorch’s native distributed APIs (torch.distributed.*) or its native parallelism APIs (DTensor, SimpleFSDP, DDP). Users had to rely on out-of-tree XLA SPMD APIs such as torch_xla.experimental.spmd_fsdp and torch_xla.distributed.spmd. For researchers who are used to native PyTorch on CUDA, switching to Trainium meant relearning how to write a training loop. Source: AWS The pain of converting GPU research codebases by adding XLA markers everywhere was a massive turnoff to anyone who used PyTorch/XLA before. Source: AWS At a high level, the only change needed is replacing .to(”cuda”) with .to(”neuron”), and in theory the vast majority of the PyTorch ATen operators will work decently well out of the box. Source: AWS In theory, with PyTorch’s torch.accelerator hardware-agnostic API, lines of code like bf16 autocast won’t even need to be changed, though some APIs in PyTorch and some research codebases still haven’t migrated to the torch.accelerator namespace and still directly call the torch.cuda.* namespace. As most academics and researchers across robotics, biotech, world models, and self-driving mainly use BF16, Trainium Torch will not support TorchAO MXFP8 on day 0 until they have made the BF16 training experience great. We don’t disagree with this sequence, as most people doing MXFP8 training are big Tier 1 labs that have enough kernel engineers to write custom kernels for the big training run. Trainium Torch Ecosystem PyTorch is popular not only because of core PyTorch but also because of the entire ecosystem of libraries, incl

SemiAnalysis · 24 thg 10, 2026

Nguyên văn nguồn
Chi tiết nguồn
167

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

5.3

Hướng dẫn Adam Zewe | MIT News

<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

MIT News: AI · 23 thg 10, 2026

Nguyên văn nguồn
Chi tiết nguồn
168

Live From SF! We’re teaching these folk how to train their own open weight agents. Sorry, but we won’t live stream. Full recording coming tomorrow!

5.1

Hướng dẫn sergiopaniego

Live From SF! We’re teaching these folk how to train their own open weight agents. Sorry, but we won’t live stream. Full recording coming tomorrow!

X · 23 thg 10, 2026

Nguyên văn nguồn
Chi tiết nguồn
163–168 / 179
Trang 28 / 30