AllyshipContent Radar

Hands-on 193

Công cụ, thử nghiệm và hướng dẫn thực hành

Crawl live-1795068900000 · 13:22 19 thg 11, 2026 · Xuất bản 13:29 19 thg 11, 2026

79

Will Repos Become Software Factories? — Mihail Eric, Monaco

39.3

Hướng dẫn Latent Space

What do you assign when students can finish last year’s homework with one prompt? For the second run of Stanford’s CS146S, Mihail Eric brought open-source maintainers into the course. Students have to understand an unfamiliar repo, propose a change, and explain it to the person who will review their code. Mihail teaches The Modern Software Developer at Stanford and leads AI at Monaco. We work through his ten-week syllabus, from watching the tool calls inside Claude Code to deciding what a subagent should return to its parent. At Monaco, his team chose to build its own agent workspaces, buy code review tools, and give an on-call agent access to Datadog and GitHub. Mihail explains the reasoning behind those decisions. Mihail thinks open-source repos will become agent-run software factories. If the factory can produce a patch, what should a junior engineer be learning to do? AI Engineer CODE Summit — https://aie.codes AI Engineer on YouTube — https://www.youtube.com/channel/UCLKPca3kwwd-B59HNr-_lvA Chapters 0:00 One million AI Engineer subscribers and the CODE Summit 0:51 Mihail Eric and Stanford CS146S 5:26 What junior engineers need to learn now 10:02 Open-source maintainers review the students’ code 15:04 A tour of the ten-week syllabus 24:04 The loop, tools and harness inside a coding agent 32:18 Context engineering for a large codebase 38:15 Skills that make you understand the diff 46:42 The Monaco.com dispute and the airplane banner 50:23 What Monaco does for sales teams 59:21 Inside Monaco’s engineering team 1:01:40 “You buy to learn, and you build to keep” 1:05:24 Making repos agent-ready and closing the on-call loop 1:12:37 When code review becomes the bottleneck 1:20:37 Security, dependencies and permissions 1:30:13 Background agents and Stripe’s Minions 1:34:16 MCP portals and why infrastructure converges 1:41:22 Building the software factory 1:45:24 What work is left for developers? 1:52:16 From personal adoption to team autonomy Show notes • Stanford CS146S syllabus and assignments — https://themodernsoftware.dev/ • Mihail builds a coding agent from scratch — https://www.mihaileric.com/The-Emperor-Has-No-Clothes/ • Mihail opens up the Claude Code harness — https://www.mihaileric.com/Demystifying-Claude-Code/ • Geoffrey Litt on understanding a diff — https://www.geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html • Monaco’s agent workspaces — https://www.monaco.com/blog/agent-developer-workspaces • How to Kill the Code Review — https://latent.space/p/reviews-dead/ • Faros report on AI adoption and review time — https://www.faros.ai/blog/ai-software-engineering • Basis’s Authority Map — https://www.getbasis.ai/blogs/how-we-made-our-monorepo-ergonomic-for-agents#canon-vs-not-canon • Stripe’s Minions — https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents • Boris Cherny’s Steps of AI Adoption — https://x.com/bcherny/status/2077929379661844559 Opening excerpt from Mihail Eric’s interview on The Thinking Mode, EO Studio — https://www.youtube.com/watch?v=wEsjK3Smovw More Latent Space interviews — https://latent.space

Latent Space · 9 thg 11, 2026

Nguyên văn nguồn
Chi tiết nguồn
81

This sounds extreme and I wouldn’t have believed it even a year ago, but we’re starting to see plenty of examples of productivity gains that are an order of magnitude or two above output levels pre-AI

36.9

Hướng dẫn levie

This sounds extreme and I wouldn’t have believed it even a year ago, but we’re starting to see plenty of examples of productivity gains that are an order of magnitude or two above output levels pre-AI. With the right harness, a deep understanding of your domain of what you want to build, likely a relatively new codebase (with some exceptions), a fluency of how to work with multiple agents at once, and a setup that makes it so you can test and deploy with limited friction, outputs on this level feel realistic in coding. What’s crazy is this gap doesn’t appear to be diminishing over time. You’d think that we’d see a convergence in the lines because of AI, but it turns out the more depth you had before -and willingness now to take a different approach- gives you crazy leverage. And that is only compounding as the models can take on larger work and harder problems. Right now this is most pronounced in software engineering for the obvious reasons. But there’s no reason we don’t start to see this in a number of other computer-use-sensitive domains like other scientific fields, AEC, finance, cyber, and more. Going to get quite wild.

X · 7 thg 11, 2026

Nguyên văn nguồn
Chi tiết nguồn
82

How to predict the behavior of molecules from scant data

35.1

Hướng dẫn David L. Chandler | Department of Chemical Engineering

<p>Discovering new molecules for specific uses is a major driver of progress in a wide variety of industries, whether it’s pharmaceutical companies looking for new drugs, or electronics or energy companies seeking new materials with particular properties. But it’s also largely a trial-and-error process, because there is not yet any way to perfectly predict how a given chemical compound will behave in terms of the myriad different properties and responses materials can exhibit.</p><p>Recent progress has been made on using artificial intelligence and machine learning to supplement or speed up laborious calculations based on physical principles that can at least provide some initial guidance on which compounds might be worth synthesizing and testing. Now, a new method developed at MIT shows that when there are few data on some technologically relevant property, such as ligand affinities for a recently discovered cellular target for a potential drug, this approach can outperform other existing methods for winnowing down the field of possible molecules that might be worth looking at.</p><p>The <a href="https://pubs.acs.org/jcisd8/article-abstract/66/17/10692/5250516/Deep-Learning-Foundation-Models-for-Low-Data?redirectedFrom=fulltext" target="_blank">findings were reported</a> in the Sept. 14 issue of the <em>Journal of Chemical Information and Modeling</em>, in a paper by MIT graduate student Akshat Shirish Zalte and recent graduate Jackson Burns PhD ’26, Research Scientist Charlles Abreu, Professor William Green, and three researchers from the chemical company BASF.&nbsp;</p><p>An earlier version of the paper has been online since last year, and the open-source model the team developed, called CheMeleon, has already found regular use by several pharmaceutical companies. CheMeleon was built using ChemProp, an open-source machine-learning package also developed and maintained by the Green group at MIT, and anyone can download CheMeleon through ChemProp and fine-tune it for their own purposes.</p><p>Earlier this year, there was a contest organized by OpenADMET for trying to predict the properties of a variety of new molecules that had not previously been synthesized, says Green, who is the Hoyt Hottel Professor of Chemical Engineering. “It was a big contest, a lot of people participated,” he says. When the winners were announced, it turned out that nine of the top 20 entries had used CheMeleon in their process. “So then we started getting all kinds of inquiries from the pharma companies about this, but our paper had just been submitted; it wasn’t even peer-reviewed or anything yet.”</p><p>While the paper went through that review process, its influence was already spreading. At least one pharma company, Green says, “sent a message to all their people, and said ‘Anytime you’re modeling a property, start with CheMeleon, that’s our new company policy’” when making initial decisions about which molecules to pursue.</p><p>The key to CheMeleon was pre-training a machine-learning model with detailed data about a million different molecules, so that it gets to learn correlations between the structure of a given molecule and how it behaves with respect to a catalog of about 1,600 descriptors that have been developed by chemists over decades. CheMeleon can then be fine-tuned to predict well-known properties like solubility and viscosity, to more esoteric properties such as bioactivity or reactivity, with a certain protein or enzyme.</p><p>Those million molecules are a minuscule fraction of the innumerable ways that atoms can bond together, but they provide enough examples so that once the AI system has been trained on them, it can work out on its own many underlying patterns and relationships, and then apply those to a new, unknown molecule that’s presented to it.</p><p>“Most of today’s deep-learning models require more data than traditional methods,” says Zalte. Some successful models use&nbsp;extensive data from the published literature in chemical journals, he says. “We don’t look at the literature at all. We just calculate a bunch of molecular descriptors like molecular weight or number of oxygens, among others.” Then predicting those descriptors forms the basis for their pre-training task.</p><p>CheMeleon is a pre-trained foundation model. Inspired by the well-known foundation models in language such as ChatGPT and Claide, which essentially are developed by feeding them massive amounts of information from the internet, CheMeleon uses molecular descriptors of a million molecules from PubChem, a large, freely available database of chemical information maintained by the U.S. National Institutes of Health.&nbsp;</p><p>With classical expert-driven models, on the other hand, specialists in a field manually input detailed knowledge about their subject. In developing these classical models, he says, for example, people know from experiments that a certain property of interest, say solubility, may relate to particular descriptors of the same molecules, “like molecular weight, or the number of oxygens that are present, or how branchy or bulky the molecule is.”&nbsp;</p><p>With this team’s approach, they can start with an existing neural network architecture and train it to learn underlying principles. “We’re not training it for a specific property, but as a general model that understands some level of chemistry, so that even if you fine-tune it with just a few data points for the task you care about, you can still get some good performance,” he says. That’s thanks to the fact that the model is “not having to learn everything from scratch.”</p><p>To verify its usefulness, “We compared our model not only against other foundation models that are out there, but also with classical models leveraging expert-designed principles,” Zalte says. Comparing its performance across 58 test cases, “we see that our model comes out on top.”</p><p>Green adds, “The CheMeleon foundation model drastically reduces the number of mol

MIT News: AI · 7 thg 11, 2026

Nguyên văn nguồn
Chi tiết nguồn
83

We spent $2m so you don't have to.

33.2

Hướng dẫn jeffboudier

We spent $2m so you don't have to. And so you can learn how to train MoEs. Mixture-of-Experts LLMs are what powers the AIs you use everyday, and few people in the world know how to train them at scale (a handful at each top AI lab). This handbook distills everything we learned in meticulously crafted chapters and figures. Disclaimer: it may take days to read. We hope to see 1000x more people able to train their own frontier models all over the world! 🤗 Congrats @Nouamanetazi @lvwerra @Thom_Wolf Media

X · 6 thg 11, 2026

Nguyên văn nguồn
Chi tiết nguồn
84

18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over.

33.1

Hướng dẫn thom_wolf

18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over. And yet the know-how to train a good MoE still lives mostly in the heads of a handful of teams. So we spent 800k GPU hours running the experiments ourselves and wrote everything down. Super happy to finally release "The Ultimate MoE Training Handbook", free and open, organized around the questions training teams hit: - how many experts, how big, shared or not? - why is half my step spent in all-to-all, and how do I hide it? - why is my run 25x slower with no error in the logs? - softmax, sigmoid or sqrt(softplus) router, and why keep it in fp32? - aux loss or DeepSeek's loss-free bias? - upcycle a dense checkpoint or train from scratch? - can I serve with fewer experts per token than I trained with? ⚠️ Warning: it's long (~70k words, budget a few days, sorry) but with 55 interactive figures, code you can run and numbers from our own runs. Read it here: https://huggingface.co/spaces/huggingface/moe-blogpost Media

X · 6 thg 11, 2026

Nguyên văn nguồn
Chi tiết nguồn
79–84 / 106
Trang 14 / 18