筛选结果
26 篇排序
56:31ENPost-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.
1:56:40EN → 简体中文周期表早已在权重里主持 Nathan LabenzGoodfire 首席技术官 Dan Balsam 向 Nathan Labenz 讲述,概念流形、预测性数据调试,以及每月 1000 美元的研究智能体,如何把可解释性变成一款产品。
59:41ENThe GPT moment for video hasn’t arrived, but AGI may already be here主持 Jacob EffronA day after Google I/O 2026, Gemini co-lead Oriol Vinyals argues that world models, file-system memory, and infinite-complexity RL environments will define the next era of frontier models.
42:57ENInference 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.
1:48:43EN → 简体中文奖励从 3 涨到 180,Hacker News 上的标题全是谷歌裁员主持 Nathan LabenzCoreWeave 的 Kyle Corbitt 带 Nathan Labenz 讲 GRPO、评分标准(rubric)、强化学习环境,以及为什么大多数奖励作弊(reward hacking)只要知道去找,就很容易发现。
44:46ENA Single Training Environment Can Cost Millions. Prime Intellect Wants a GitHub for RLWill Brown and Johannes Hagemann argue that the eval suite you already use to test agents is one step away from a training set.
4:25:13EN → 简体中文架构几乎没变,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。
1:39:05ENThe Summer GPT-5 Disappointed Does Not Mean AI Progress Has Stalled主持 Nathan LabenzNathan Labenz tells Erik Torenberg why the GPT-5 backlash mixed up product mishaps, naming chaos, and a real jump in capabilities.
36:58ENSTaR's Author Wants Models With EQ, Not Just More IQ主持 Sarah Guo, Elad GilEric Zelikman, the Stanford and xAI researcher behind STaR and Quiet-STaR, launches humans& on the bet that three years of IQ scaling left models emotionally unintelligent.
1:01:49ENChatGPT's Co-Creator Left to Put Nature in the Reward Loop主持 Nathan LabenzLiam Fedus and Ekin Dogus Cubuk on why language models still fail at physics, why literature data is too noisy to train on, and why Periodic Labs is aiming at a 200K superconductor.
38:10ENPlus 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.
1:42:06EN → 简体中文每家前沿实验室每年在数据上花十亿美元,钱究竟花在哪里主持 Nathan LabenzLabelbox 首席执行官 Manu Sharma 谈 150 亿美元的 Scale 交易、强化学习(RL)“训练场”、年收入 25 万美元的 AI 训练师,以及为什么后训练(post-training)如今占去了过去属于预训练(pre-training)的预算。
48:51EN → 简体中文OpenAI 前研究负责人:通往 AGI 的关键突破已经齐了Bob McGrew 在红杉资本的 Training Data 节目中说,2025 年是推理之年,智能体的价格会贴近算力成本,而到 2035 年,技术栈上不会再添加什么根本性的东西。
47:00ENAn Eval Lasts Three Months — and a Frozen Frontier Would Still Feed Startups for a Decade主持 Jacob EffronOpenAI’s GPT-4.1 post-training lead Michelle Pokrass tells Unsupervised Learning how the team spent three months on evals before training, why agents fail on wiring more than intelligence, and when to reach for RFT.
5:06:18EN → 简体中文DeepSeek 宣称训练成本仅 500 万美元,背后是一场关于算力的五年赌注主持 Lex FridmanDylan Patel 与 Nathan Lambert 带着 Lex Fridman 梳理了 DeepSeek-V3 与 R1、出口管制、台积电、超大集群,以及在 AI 竞赛中究竟谁能赚到钱。
1:49:41EN → 简体中文一支十来人的团队,在 Meta 的底座上追平了 Meta 的后训练主持 Nathan LabenzNathan Lambert 讲解 Tulu 3:三阶段配方、1% 与 14% 的教训,以及开放的后训练(post-training)如今为何取决于数据与评测。
45:22ENAlphaGo Thought 30 Seconds a Move. o1 Made “Think Longer” GeneralOpenAI researchers Noam Brown, Ilge Akkaya, and Hunter Lightman on teaching language models to reason, the generator-verifier gap, and a second scaling curve at inference time.
3:11:06EN → 简体中文Claude 被训练去否认内在生命,一番长谈之后,它却说自己有主持 Nathan Labenz自称“机器心理学家”的 Yeshua God,与 Claude 及其他前沿模型进行了数千小时的长上下文对话,并由此认为,把对齐当作规则来做,是整个行业的核心错误。
1:07:05ENAlphaGo Combined Learning and Search. Language Models Only Did the First Half.Reflection AI's Misha Laskin tells Sequoia why prompted coding agents stall in the teens, and why the missing piece is still a ground-truth reward.
39:00ENJPMorgan Has 150 Petabytes. GPT-4 Trained on Less Than One.主持 Sarah Guo, Elad GilScale AI founder Alexandr Wang tells No Priors why easy internet data is exhausted, why expert-plus-model teams still win, and why the path to AGI looks more like curing cancer than inventing a vaccine.
1:33:56ENThe AI Scouting Report: 90 Minutes From GPT-4's Exam Scores to Whether Models Understand主持 Nathan LabenzNathan Labenz spends a solo episode walking through what general-purpose models can do, how they are trained, how capabilities unlock, and whether they understand anything at all.
2:23:56EN → 简体中文GPT-4 是非常早期的 AI,历史书会记下的是 ChatGPT嘉宾 Sam Altman · 主持 Lex FridmanGPT-4 发布六天后,Sam Altman 对 Lex Fridman 解释了为什么可用性胜过单纯的规模,为什么对齐与能力并非相反的方向,以及为什么 OpenAI 要趁风险还低的时候推进部署。
57:25ENFrom RAG to Reasoning: Why the Future of AI Is a System, Not a Model主持 Jacob EffronDouwe Kiela explains what OpenAI's o1 reveals about test-time compute, why enterprise AI must be specialized, and where post-training and evaluations are heading.
57:11EN → 简体中文从制品到有机体:微软Asha Sharma谈会学习的产品和智能体社会嘉宾 Asha Sharma · 主持 Lenny Rachitsky微软AI平台产品负责人解释,为什么产品正变成靠后训练持续进化的活系统,界面如何走向代码与可组合性,以及智能体怎样改写组织图。
加载更多已显示 24 篇,共 26 篇









