筛选结果
201 篇排序
10:40EN → 简体中文本地测试 Nous Hermes 13B:推理表现不一与不安全回答Matthew Berman · Forward FutureMatthew Berman 在托管和本地硬件上比较一款 2023 年的量化模型,发现其编码、逻辑和安全表现不一致。
11:42EN → 简体中文Guanaco 与 QLoRA:2023 年的微调方法与模型测试Matthew Berman · Forward FutureMatthew Berman 解读 QLoRA 论文,解释报告中的显存节省,并用一组非正式提示测试 Guanaco 65B。
1:25:35EN → 简体中文两年半,从 33.3% 到 86.5%:谷歌把基础模型送进了医学院主持 Nathan LabenzVivek Natarajan 向 Nathan Labenz 讲述 Med-PaLM 和 Med-PaLM 2:少数据的技术、九个维度的临床医生评测,以及为什么美国医师执照考试(USMLE)式的高分是一个错误的奖杯。
10:16EN → 简体中文DarkBERT:让语言模型适应暗网文本Matthew Berman · Forward FutureMatthew Berman 介绍一项基于 RoBERTa 的模型研究及其网络安全评估。
11:07EN → 简体中文2023 年测试 MosaicML MPT-7B Chat 与 StoryWriterMatthew Berman · Forward FutureMatthew Berman 介绍 MPT-7B 的不同变体、许可及一组结果不一的本地测试。
58:24EN → 简体中文随机找 100 位医生,Zak Kohane 会惊讶于其中有人能胜过 GPT-4主持 Nathan Labenz哈佛医学院生物医学信息学系主任谈一例罕见的新生儿病例,为何像药物那样的试验无法为通用医疗模型做认证,以及 AI 为何仍应为患者服务。
9:58EN → 简体中文HuggingChat 的 2023 年初测:写作、推理与 Python 编程Matthew Berman · Forward FutureMatthew Berman 测试早期由 OpenAssistant 驱动的 HuggingChat,让它写作、算术、推理和编写贪吃蛇游戏,并将几个输出与 GPT-3.5 和 GPT-4 比较。这次小规模历史测试展示了潜力,也暴露出局限。
53:37EN → 简体中文30 个模型、42 个场景、7 项指标:Percy Liang 为语言模型做的公开体检主持 Sarah Guo, Elad Gil一个只用 PubMed 论文训练的模型,曾在美国医师执照考试中名列前茅,直到一个规模约大 200 倍的系统超过了它。
2:34:20EN → 简体中文智能,是你习得未曾准备过的技能的效率嘉宾 François Chollet · 主持 Lex FridmanFrançois Chollet 对 Lex Fridman 说,技能不等于智能,GPT-3 仍然只是在做模式匹配,而一种名为 ARC 的网格谜题,至今仍能把孩子和机器区分开。
1:49:51EN → 简体中文曾被斥为倒退的设计思路,如今几乎驱动着每一款新芯片嘉宾 David Patterson · 主持 Lex Fridman图灵奖得主 David Patterson 与 Lex Fridman 谈精简指令集(RISC)、RAID、摩尔定律(Moore's Law)、开放硬件,以及没有软件栈的芯片为何难以存活。
16:03ENQuantum Supremacy Is a Well-Defined Speedup, Not a Useful Task嘉宾 Scott Aaronson · 主持 Lex FridmanScott Aaronson explains what Google's 53-qubit experiment actually demonstrated, how it was verified, and why complexity theory can offer evidence but not proof against classical spoofing.
57:57EN → 简体中文Sebastian Funk:打造快速、安全且易维护的数据包处理器Sebastian Funk · Jane Street一个单核运行情数据的案例展示了延迟目标、分配控制和代码生成如何与高层抽象共存。
11:11ENYann LeCun on Benchmarks, Interactive Environments, and Why Human Intelligence Is Not General嘉宾 Yann Lecun · 主持 Lex FridmanIn a 2019 clip, LeCun defends benchmarks as a test of AI claims, explains why interactive environments break the usual train-test protocol, and argues that "general intelligence" is a misnomer.
1:22:57ENHuman-Level but Not Human-Like: A GPT-4 Red Teamer on What He Saw Behind the Mask主持 Nathan LabenzNathan Labenz recounts two months of full-time testing of an early GPT-4, from the AI doctor of his dreams to a model that would plan assassinations on request.
14:41ENTalking for 20 Minutes: How the Alexa Prize Turns Real Customers into a Testbed for Conversational AI嘉宾 Rohit Prasad · 主持 Lex FridmanRohit Prasad, head scientist of Amazon Alexa, explains the Alexa Prize grand challenge, how it is judged and why true conversation still needs reasoning.
16:41ENAlphaFold 2 and the Protein-Folding Breakthrough: Why Lex Fridman Calls It a Landmark for AI and Biology主持 Lex FridmanA quick explainer on DeepMind's CASP14 result, how protein folding works, how AlphaFold 1 and 2 likely differ, and what the advance could unlock.
57:21ENGemini 2.0 Flash Goes to Production: Pricing, Model Lineup and What Developers Should Build Next主持 Nathan LabenzGoogle DeepMind's Logan Kilpatrick walks through launch-day Gemini 2.0 releases, rate limits, model naming, and where he sees startup opportunities.
1:00:25ENCan Machines Think? A Guided Tour of Turing's Test, Its Critics and Its Successors嘉宾 Turing Test · 主持 Lex FridmanLex Fridman walks through Alan Turing's 1950 paper, the objections to it, and the benchmarks that have tried to replace or extend the imitation game.
1:22:17ENOne Set of Weights, Fourteen Medical Tasks: Inside Google's Multimodal Med-PaLM M主持 Nathan LabenzVivek Natarajan and Tao Tu explain how Google built a single generalist biomedical model that reads images, genomics and text, and what it will take to put it in front of patients.









