|
|
|
|
六一儿童节的时候， Richard Sutton锐评AI 各位女士们、先生们，大家好。 很遗憾我今天无法亲临现场，与各位进行面对面的交流讨论。但我仍然很高兴能够通过这段录制的视频，与大家分享一些关于人工智能当前与未来发展的高层次思考，尤其是 AI 与科学、数学之间关系的看法。我理解，这也正是本次会议以及 SAIR Foundation 所关注的核心主题。 我想先讲一个老笑话，相信大家都听过。有位研究人员的工作正在接受评审，评审意见回来后写道：“这项工作既新颖（novel），又优秀（good）。遗憾的是，优秀的部分并不新颖，而新颖的部分并不优秀。” 我关于 AI 的第一个观点是：这个评价恰好适用于今天相当大一部分 AI。并非所有 AI 都如此，但相当大一部分确实如此。我们今天所谓的“生成式 AI”（Generative AI）几乎都属于这一类，包括大语言模型、图像模型、视频模型，甚至最近用于学习世界模型（world model）的一些新方法。这些 AI 的....
|
|
From the floor of HumanX, Ryan welcomes Songyee Yoon, managing partner at Principal Venture Partners (PVP), to chat about AI development outside the US, from the need to adapt models to local languages and culture to the challenges of the global supply-chain for things like semiconductors to how venture capital is looking at international AI companies. ....
|
|
Agentic AI is getting physical. At COMPUTEX on Tuesday, NVIDIA announced NVIDIA JetPack 7.2 and NVIDIA NemoClaw support on NVIDIA Jetson. JetPack 7.2 brings agentic AI skills, Yocto project support, NVIDIA CUDA 13 on NVIDIA Jetson Orin, a substantial performance gain on Jetson AGX Orin 32GB module and Multi-Instance GPU (MIG) support on NVIDIA Jetson […]
|
|
PLEASE AGENTPMAN1, I'LL DO ANYTHING! I'LL FUCKING SUCK YOUR DICK! YOU WANT ME TO EAT YOUR SHIT? I'LL...
|
|
|
|
|
家里有个小 NAS 。里面存了一些歌。一半是用 NAS 自带的 app 听，一半是。。。SMB 共享打开听 虽然 NAS 也提供 DLNA ，一直以来找不到趁手 app ，要么收费，要么 bug 多，要么不能多端。 13年前我也想 基于 chrome.socket 做个 Chrome App 弄个类似的。结果这破玩意实现有问题， 多连接会导致 hang 。最后 2020年Chromium决定杀死 Chrome Apps 周六的时候，实在无聊，决定又开始搓轮子。在思考 SSDP/UPnP ，native UI, electron，命令行这些选型的时候，突然想到，DLNA服本来就要提供一个http，自己再造个 http 客户端去通信，岂不是多此一举？只要依托它，解决跨域……等等，用个 bookmarklet 不就行了？当页调用 fetch() ，走 SOAP 协议，完美。 所以这就有了，网页版听歌的。不需要安装 app ，只需要一个浏览器书签 http....
|
|
|
Let's take a moment to shout out all the articles we’re seen posted so far and encourage folks reading to reach out about your own.....
|
|
In this two-for-one special recorded at HumanX, Ryan is joined by Dataiku’s Florian Douetteau to chat about the governance, orchestration, and data requirements for serious agentic systems and 1Password’s Nancy Wang for a conversation on making agent swarms secure.....
|
|
If AI is the new power tool for developers, is there still value in artisanal craft when anyone can be a builder?....
|
|
This post was automatically generated. Chess For the Day Record: 3-0-1 Net Elo Change: +11 Games Played Strays55 - LOSS Mopsi1 - WIN BESHOY_NESSEEM - WIN LeonRubi - WIN Blog Posts On This Day May 27, 2025 (3 posts) May 27, 2024 (3 posts) May 27, 2023 (2 posts) May 27, 2022 (6 posts)
|
|
|
AI’s impact on software engineering continues, and more and more of that AI is packaged as agents as results from our newest pulse survey show agentic usage has almost doubled (59%) since we last asked about it in our annual Developer Survey....
|
|
From the floor of HumanX, Ryan Donovan is joined by Peter Salanki, CTO and co-founder of CoreWeave, to chat about what it really takes to run AI in production; the growing importance of observability, utilization, and scheduling; and Peter’s advice for avoiding the trap of over-architecting too early. ....
|
|
|
作为一个外行，我一直对“AI”的魔力感到惊奇，我一度以为神经网络一层一层传播，可以看成某种有限步骤的图灵机。AI提醒我不要瞎类比，图灵机左移右移是离散，确定的逻辑，神经网络是fp32上连续的概率映射。 后来稍微深入了解了一下，认识到对于一个深度固定的 Transformer 模型（比如 96 层的 GPT-4），它的单次前向传播算是一个深度固定的有向无环图（DAG）。所谓的 predict next token，可以粗糙理解成 next_token = eval(model_weights, history+input) 这里最奇特的算是：自回归（Autoregression）。传统的冯·诺依曼架构中，指令（Code）和数据（Data）是分开的。但在 LLM 中，上下文是动态的指令+数据。输出什么样话，什么时机结束。LLM得自己想办法把画圆回来，并且知道什么时候该停止吐词。 这种 控制面 和 数据面 混合的做法让我感到非常不适，也是诸多prompt inje....
|
|
|
This post was automatically generated. Chess For the Day Record: 4-0-3 Net Elo Change: +11 Games Played Pif_Paf - WIN LiverFluke - WIN Senin8 - LOSS alpingo45 - WIN Renzoxx1 - WIN burki97 - LOSS Ridwan-crbn - LOSS Blog Posts On This Day May 22, 2025 (6 posts) May 22, 2024 (2 posts) May 22, 2023 (3 posts) May 22, 2022 (7 posts)
|
|
Experiments in agentic engineering and AI-driven development.....
|
|
Recorded at HumanX, Ryan sits down with Garima Kapoor and Anand Babu Periasamy, co-founders and co-CEOs of MinIO, to chat about eliminating the storage bottlenecks that leave GPUs underutilized, their partnership with NVIDIA on the new STX reference architecture, and why modern AI infrastructure is converging on S3-compatible object storage. ....
|
|
With much of a software engineer’s time moving from writing code to structuring prompts and reviewing code, the workday is getting denser and more intense. Can AI solve the problems it's causing?....
|
|
|
|
Ryan welcomes Jaime DeLanghe, chief product officer at Slack, to chat about how they’re preparing to integrate everybody’s agents in their chat application. ....
|
|
AI 现在调用都走 OpenAI-like 接口，遇到长任务多半会走 stream=true 然后AI能力也多半会接力返回给下游，比如浏览器 那么问题来了。下游如果连接断开，是不是就意味着服务器得把AI的输出接住，然后下一次请求接着吐？ 如果下一次请求不路由到这个节点和进程，意味着接住要设计一套缓存 更麻烦的是，现代web框架一般都是请求 - 响应模式的，如果浏览器断开连接，按正常流程，后端也会抛出异常之类的中断 所以“接AI的话”这玩意实际上设计还要考虑挺多东西，很麻烦？？？ 这个问题丢给 ChatGPT它这么回答： 断了就断了，不续传。用户重新发起请求，后端重新生成，最多靠 prompt cache / KV cache / 上下文缓存降低重复成本。很多产品其实就是这么干的，因为实现成本最低。 尼玛。又学到一招。人脑还是想复杂了。 btw 吐槽一下现在 vibe coding 开发者估计很少有人会去在意这些细节了。
|
|
On the floor of HumanX, Ryan is joined by Adam Meyers, Senior VP of Counter Adversary Operations at Crowdstrike, for a deep dive on their latest Global Threat Report that tracks over 281 adversaries across nation states, e-crime, and hacktivist organizations.....
|
|
|
Because someone still has to own the consequences of what gets built and whether it can function at scale.....
|
|
无聊看了下 FSRS (Free Spaced Repetition Scheduler) ，想看它怎么存数据的 Card — 卡片当前状态 表结构： 字段 类型 说明 due Date 下次复习时间 stability float 记忆稳定性（R=90% 时的间隔天数） difficulty float 难度，范围 [1, 10] state int 状态机：New=0, Learning=1, Review=2, Relearning=3 learning_steps int 当前学习步骤 index scheduled_days int 本次调度天数 reps int 总复习次数 lapses int 遗忘次数 last_review Date? 上次复习时间 核心字段详解 1. state 状态机，取值 New=0 Learning=1 Review=2 Relearning=3 Again Go....
|
|
In this No Dumb Questions, Phoebe is joined by Stack Overflow’s tech lead for the infrastructure team, Josh Zhang, to learn about the cloud, compute, and data centers. ....
|
|
In this two-for-one episode recorded at HumanX, Ryan is first joined by Christine Yen, CEO of Honeycomb, to discuss how AI compresses the software development lifecycle, making observability about capturing the right telemetry. Then, Spiros Xanthos, founder and CEO of Resolve AI, shares with us how AI coding increases code volume but decreases human intuition, making production operations harder than ever. ....
|
|
不瞎扯，只说我观察到的。 AI coding的时候，做plan，AI在评估复杂度，会蹦出来一句：”该部分大概 1-2 天就能完成“ 然后点击 build 按钮，10分钟后，完成了。 以后AI必然能会意识到，人类开发速度和AI速度的差异。 如果说 predict next word 时代的AI还可能混淆，以为自己是以某个人类自然语言里的主人公身份说话，后chatgpt时代 AI参与人类劳动的文章和叙事越来越多，AI这个物种和人类就有明确的身份区分了。 最近在折腾自己的 agentic tool。我发现AI自己是没有意识到AI的潜力的。 对于开放式课题，AI很多设计还是按照传统软件开发的老思路，对于一些复杂场景判定，需要在代码里写死 if else 规则，做配置，设定静态边界。 问题是，AI的通用能力都烂大街了。特别是我这里明明后续都要调用一次AI推理接口，很多判定的事不过 prompt 里一句话就带过了 但是 AI 没能想到这一点。 我也能..
|
|
Jon Hyman, co-founder and CTO of Braze, shares how he's led the company's engineering organization over nearly 15 years of growth — and how they transformed into an AI-first team in just a few months.....
|
|
Find out about the new features in .NET 11 Preview 4 across the .NET runtime, SDK, libraries, ASP.NET Core, .NET MAUI, C#, Entity Framework Core, and more! The post .NET 11 Preview 4 is now available! appeared first on .NET Blog .
|
|
Why AI-accelerated teams keep breaking production — and what the ones that don't are doing differently....
|
|
|
At HumanX, Ryan is joined by Philip Rathle, CTO at Neo4j to discuss what knowledge context means for AI agents, how limitations like stale training data make the model-only approach to agents a bad fit for enterprise environments, and how Graph RAG raises the bar for accuracy and reduces context rot by combining vectors with a knowledge graph so agents are more targeted and connected.....
|
|
|
Signature-based detection has always known what it was looking for. Machine learning and autonomous agents are changing the question entirely, shifting from "does this match a known pattern?" to "does this actually make sense in context?"....
|
