筛选结果
99 篇排序
1:13:33ENA Trillion Tokens Later, the Concept Finally Lands: Łukasz Kaiser on What Transformers Still Miss主持 Jacob EffronThe Attention Is All You Need co-author tells Unsupervised Learning that transformers will probably keep winning — and that they still do not learn the way people do.
2:02:17ENSAEs Find the Nouns. Parameter Decomposition Looks for the Verbs主持 Nathan LabenzGoodfire's Lee Sharkey argues that sparse autoencoders recover a network's variables, not the algorithms those variables run, and shows how stochastic parameter decomposition tries to recover the computations themselves.
1:37:05EN → 简体中文一位代数几何学家辞去终身教职,因为碗形的损失曲线图是假的主持 Nathan LabenzTimaeus 的 Jesse Hoogland 和 Daniel Murfet 认为,真正决定神经网络如何学习、如何泛化、又如何出错的,是损失景观中的奇点,而不是每份幻灯片里那些平滑的山谷。
2:12:00ENGradient 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.
2:33:39EN → 简体中文攻击者后出手:Nicholas Carlini 为什么总能攻破 AI 防御主持 Nathan Labenz谷歌 DeepMind 安全研究员谈简单的损失函数、70% 的稳健准确率,以及为什么修复必须落在系统层面,单靠模型本身不够。
1:40:22ENTwenty-atom crystals taught an AI the physics of a heartbeat主持 Nathan LabenzOrbital Materials' Jonathan Godwin and Tim Duignan tell Nathan Labenz how a neural network trained on tiny crystals simulated a potassium ion channel on a single 2017 GPU.
1:42:16EN267 Papers in Seven Months: Albert Gu Explains the State Space Model Revolution主持 Nathan LabenzThe Mamba co-author walks through five years of recurrent-model work, why selective compression beats a growing KV cache, and what Mamba 2 trades away to train faster.
1:54:01EN → 简体中文一个小网络长出了盐晶体,而训练数据里从来没有它主持 Nathan LabenzTim Duignan 用盐水训练了数万个参数。模型“发明”了一种晶体,还推出了一套连续介质溶剂理论(continuum solvent theory);而他读博期间花了整整一个博士阶段,也没能手工把这套理论写出来。
1:58:43EN → 简体中文一条只有 302 个神经元的线虫已经会权衡取舍,AI 仍停在同一座梯子上主持 Nathan LabenzMax Bennett 把 6 亿年的大脑演化对应到五次突破上,并追问今天的模型究竟爬上了哪几级台阶。
1:19:24ENA Toy Paper Did Not Prove LLMs Cannot Generalize. The Discourse Still Ran With It.主持 Nathan LabenzNathan Labenz and Erik Torenberg unpack a viral DeepMind result, a weekend reproduction, and a premature war over voluntary AI commitments.
43:45ENEight Years of No Growth, Then OpenAI and NVIDIA Became His Customers主持 Sarah Guo, Elad GilWeights & Biases CEO Lukas Biewald tells No Priors how a labeling company with no chasm to cross funded the ML developer tools that trained most of the large language models.
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.
1:32:22ENA Million Bytes and 256 Symbols: Meta's MEGABYTE Wants to Throw the Tokenizer Away主持 Nathan LabenzLili Yu of Meta AI explains a multiscale transformer that predicts million-byte sequences, skips tokenization, and aims to model any file as raw bytes.
1:52:46ENCharge Neural Nets by Connection Length, and the Black Box Splits Into Modules You Can See主持 Nathan LabenzMIT physicist Ziming Liu walks Nathan Labenz through BIMT, a brain-inspired training trick that sparsifies networks until you can read them by looking.
1:25:50ENPrompt Engineering Became Programming, and Riley Goodside Is Fine With That主持 Nathan LabenzScale AI’s first staff prompt engineer on tool use, fine-tuning without amnesia, generate-then-rank, and why the 2023 job-of-the-future story never arrived.
46:57EN95% of Drug Programs Fail. Daphne Koller Is Betting Machine Learning on the First Decision主持 Sarah Guo, Elad GilA new medicine still takes more than a decade and well over a billion dollars — and Insitro's founder argues the expensive failures start with the wrong target, indication, or patient.
2:31:57EN → 简体中文Jay McClelland 谈思维如何从神经网络中涌现嘉宾 Jay McClelland · 主持 Lex Fridman这位斯坦福大学的认知科学家,与 Lex Fridman 聊了联结主义(connectionism)、反向传播(backpropagation)、David Rumelhart,以及为什么直觉会先发现某些东西,之后逻辑才去证明它。
2:51:33EN → 简体中文自我是一段 GPT 提示词:Wojciech Zaremba 谈意识、Codex 与费米的沉默嘉宾 Wojciech Zaremba · 主持 Lex FridmanOpenAI 联合创始人 Wojciech Zaremba 对 Lex Fridman 说,他认为我们在宇宙中可能是孤独的;他还说,GPT-3、Codex 和一只单手机器人,已经暗示了智能是如何构建的。
2:18:29EN → 简体中文你的大脑皮层,是 15 万个模型在给一只咖啡杯投票嘉宾 Jeff Hawkins · 主持 Lex FridmanJeff Hawkins 对 Lex Fridman 说:智能是学习世界模型的能力,新皮层是成千上万张相互独立的地图在投票,而通往机器智能最快的路径,是对这层皮质做逆向工程。
2:30:30EN → 简体中文成功第一步的谬误嘉宾 Ishan Misra · 主持 Lex Fridman伯克利的 Jitendra Malik 解释了计算机视觉为何仍然看起来很简单、全自动驾驶为何还遥遥无期,以及孩子知道而今天的网络不知道的是什么。
1:56:17EN → 简体中文如果把地球重演一百万次,还会有像我们这样的生物出现吗?嘉宾 Risto Miikkulainen · 主持 Lex FridmanRisto Miikkulainen 谈神经进化、鬣狗围猎、从不睡觉的罗勒,以及为什么“变异加选择”仍能找到反向传播看不见的解。
1:49:56EN → 简体中文派对上那个 21 岁的年轻人,说对了 GameStop嘉宾 Richard Craib · 主持 Lex FridmanNumerai 创始人 Richard Craib 与 Lex Fridman 谈 WallStreetBets、众包机器学习,以及金融为何是一台把未来拉到当下的机器。
2:02:45EN → 简体中文偷代码:让算法跑起来本身一分不给主持 Lex FridmanCharles Isbell 和 Michael Littman 与 Lex Fridman 聊了为什么机器学习不只是统计,为什么挣扎必须保有希望,以及 6600 美元的在线硕士到底卖的是什么。
2:01:07EN → 简体中文Michael Littman 不等待失控的神,他在观察我们嘉宾 Michael Littman · 主持 Lex Fridman布朗大学的强化学习(reinforcement learning)研究者 Michael Littman 对 Lex Fridman 说,为什么意外出现的超级智能(superintelligence),在他看来就像一架 747 自行组装起来,而社交媒体、死记硬背式的语言和心智理论(theory of mind),更让他担心。
加载更多已显示 24 篇,共 99 篇









