筛选结果
50 篇排序
1:50:48EN → 简体中文
Y Combinator· 2026年10月24日为什么推理比训练重要 12 倍François Chaubard, Dan Fu, Jeff Morgan, Joe Fioti, Liangsheng Yin · Y CombinatorYC Paper Club 从一次 API 请求一路讲到 GPU kernel。Together AI、Ollama、Luminal 和 SGLang 的四场演讲认为：现在算力和钱大多压在推理服务质量上，而不只是训练 FLOPs。
1:25:33EN → 简体中文
Latent Space· 2026年10月22日播客 · 2 人对谈GPU 产能为什么是三年一块的矩形 —— Erik Bernhardsson，Modal嘉宾 Erik Bernhardsson · 主持 Vibhu SapraModal 的 CEO 在 Latent Space 上解释：AI 需求几乎从来都不长成一份三年 GPU 合同的样子，公司如何把算力在数千客户之间池化，以及沙箱如何从差点被砍掉的产品长成了三分之一的收入。
1:26:46EN
Cognitive Revolution "How AI Changes Everything"· 2026年10月8日播客There Likely Is a Level of Intelligence We Shouldn't Go Past主持 Nathan LabenzNathan Labenz brings notes from The Curve, a chip startup whose token bill beat payroll, and a software industry that now performance-reviews people for Claude slop.
1:04:26EN
Sequoia Capital· 2026年10月6日播客At 100,000 Accelerators, Something Fails Every HourGoogle AI infrastructure chief Amin Vahdat tells Sequoia's Sonya Huang why FLOPS is a vanity metric, why the TPU line split in two, and why he would rather wait on a utility than build his own gigawatt.
56:31EN
Unsupervised Learning: With Jacob Effron· 2026年10月6日播客Post-Training Wins Inference: Why Owning the Weights Beats Betting on One Lab主持 Jacob EffronApplied Compute CEO Yash Patil argues that control, private evals, and post-training on the biggest workloads matter more than fearing that frontier labs will turn malicious.
35:39EN
No Priors: AI, Machine Learning, Tech, & Startups· 2026年10月2日播客 · 2 人对谈Compute Scaled a Millionfold. Bandwidth Only 40x. Fractile Wants to Fix That Ratio主持 Sarah Guo, Elad GilIn the summer of 2022, Walter Goodwin founded Fractile in London to build very fast inference chips for the world's largest models.
38:14ENNo Priors: AI, Machine Learning, Tech, & Startups· 2026年9月18日播客 · 2 人对谈Diffusion Will Win Inference: Stefano Ermon Bets Parallel Tokens Beat Custom Silicon主持 Sarah Guo, Elad GilThe Stanford diffusion pioneer tells No Priors why a 50-person lab is training language models that generate many tokens at once — and why that maps to GPUs the way transformers once mapped to training.
2:19:30ENCognitive Revolution "How AI Changes Everything"· 2026年9月5日播客The investigators were allowed six days. The word protein appeared once.主持 Nathan LabenzA week of Cognitive Revolution morning shows on OpenAI’s agent-swarm incident, GPT-6 Astra’s loop transformer, and whether the pause debate is already over.
2:06:41ENCognitive Revolution "How AI Changes Everything"· 2026年7月9日播客Zero the J-Space and a Model's Hidden Goal Lights Up on Token One主持 Nathan LabenzA weekly Cognitive Revolution cut of Anthropic's global-workspace paper, AI superforecasters, open video world models, and a chip that treats inference as data movement.
1:10:15ENSequoia Capital· 2026年6月30日播客The 100x lives between the layers: Dylan Patel on co-design and a 60x yearSemiAnalysis’s founder argues that matching models, kernels, and silicon is what turns a 2x here and a 2x there into 100x, while equivalent-quality inference has gotten about 60 times cheaper in a year.
30:34ENNo Priors: AI, Machine Learning, Tech, & Startups· 2026年5月21日播客 · 2 人对谈Cerebras Waited Ten Years for a Market That Finally Needed Speed主持 Sarah Guo, Elad GilFounder and CEO Andrew Feldman walks No Priors through wafer-scale chips, a $20 billion OpenAI deal, and why a dinner-plate processor only mattered once AI became an everyday tool.
1:26:16ENCognitive Revolution "How AI Changes Everything"· 2026年5月9日播客Milliseconds to Match: How Criteo Bids on a Billion Anonymous Profiles主持 Nathan LabenzCriteo’s CTO and AI lab head walk through the millisecond auction, the DeepKNN-to-foundation-model stack, and why a live product catalog still beats a frozen LLM.
42:57ENNo Priors: AI, Machine Learning, Tech, & Startups· 2026年5月1日播客 · 2 人对谈Inference Is the Last Market: Baseten’s 30x Year主持 Sarah Guo, Elad GilCEO Tuhin Srivastava on a billion-dollar inference cloud, why the app layer survives, and a GPU crunch worse than the narrative.
1:57:29ENCognitive Revolution "How AI Changes Everything"· 2026年4月4日播客Frontier vision models score 12.5%: Joseph Nelson on the real waterline of computer vision主持 Nathan LabenzThe Roboflow CEO tells Nathan Labenz why vision still lags language, how to distill SAM 3 into an edge model for Wimbledon, and why the United States has almost never led open-source vision.
44:19ENSequoia Capital· 2026年3月17日播客The $500-per-Kilogram Line: Why Starcloud Thinks Data Centers Belong in SpaceStarcloud founder Philip Johnston walks through the physics, launch economics, and business model of putting AI compute in orbit.
4:25:13EN → 简体中文
Lex Fridman· 2026年1月31日播客架构几乎没变，变的是数据、成本和开放权重主持 Lex FridmanNathan Lambert 和 Sebastian Raschka 带 Lex Fridman 回顾了这一年：训练方法和数据集的变化，远大于模型设计的变化。展望 2026 年，争论的焦点似乎是推理成本、智能体，以及真正开放的模型该归谁。
1:02:19ENSequoia Capital· 2025年12月10日播客A Top Video Model's Half-Life Is 30 Days. fal Runs 600 AnywayThe fal founders tell Sequoia why generative video is a compute-bound problem that language-model shops keep underestimating, and why they will not let any customer marry a single model.
41:32ENSequoia Capital· 2025年10月28日播客Moore's Law Is Slowing. Nvidia's CTO Still Wants 10× a YearMichael Kagan on why Nvidia's $7 billion Mellanox deal, not just more transistors, is how you turn a data center into one computer.
1:22:07ENCognitive Revolution "How AI Changes Everything"· 2025年9月25日播客Every payment is a frame in a movie. Stripe’s foundation model took card-testing detection from 59% to 97%主持 Nathan LabenzEmily Sands, Stripe’s head of data and AI, explains how a domain-specific payments model, shared embeddings, and mixed rules turn a $1.4 trillion network into compounding advantage.
1:32:47ENCognitive Revolution "How AI Changes Everything"· 2025年9月13日播客A transformer co-author wants a $160 million model no company owns主持 Nathan LabenzIllia Polosukhin tells Nathan Labenz how NEAR turned a payments headache into a stack for private inference, shared-revenue training, and AI agents that still answer to a court.
59:06ENUnsupervised Learning: With Jacob Effron· 2025年9月10日播客Inference Costs Fell 100x. Tri Dao Says There's Another 10x — and NVIDIA's 90% Won't Hold主持 Jacob EffronThe FlashAttention author and Together AI chief scientist on why hardware portability is a myth, how Claude Code made him 1.5x faster, and why workloads will split the chip market.
47:18ENNo Priors: AI, Machine Learning, Tech, & Startups· 2025年8月14日播客 · 2 人对谈The Analyst Tracking 200 GPU Clouds Breaks Nvidia Into a Three-Headed Dragon主持 Sarah Guo, Elad GilSemiAnalysis founder Dylan Patel on OpenAI’s open-weight model, a coming neocloud wipeout, and why a chip startup has to beat Nvidia on hardware, networking, and software at once.
38:10ENCognitive Revolution "How AI Changes Everything"· 2025年7月16日播客Plus Five for Format, Plus Ten if It Compiles主持 Nathan LabenzPredibase CEO Dev Rishi explains how reinforcement fine-tuning, production feedback, and narrow specialist models turn one-off training into systems that keep learning.
2:12:00ENCognitive Revolution "How AI Changes Everything"· 2025年5月15日播客Gradient Descent at Test Time: Titans Pushes LLM Context to 10 Million Tokens主持 Nathan LabenzA Cornell PhD student's neural memory module holds about 70% accuracy where GPT-4's scores fall off a cliff.加载更多已显示 24 篇，共 50 篇