筛选结果
27 篇排序
2:00:51ENResample Every Token and You Can Watch the Model Decide主持 Nathan LabenzGoodfire's Eric Bigelow on forking paths, in-context learning, and why chain-of-thought monitoring is getting harder to trust.
2:06:11ENLindy Runs on DeepSeek. Its CEO Still Wants Chinese Models Banned主持 Nathan LabenzFlo Crivello tells Nathan Labenz how Lindy Teammate lives in Slack, reaches two billion tokens of memory in two model calls, and why he would still outlaw the cheap Chinese models that currently power the product.
2:16:29EN → 简体中文从每月两篇到每月六十篇:两周中国行看到的 AI 安全主持 Nathan LabenzNathan Labenz 认为,那句耳熟能详的“中国绝不会慢下来”,一接触到中国真实的 AI 安全生态就站不住了。
1:47:38EN19 Neurons Can Drive a Car. Liquid AI Wants Foundation Models on Every Chip主持 Nathan LabenzRamin Hasani traces liquid neural networks from a worm's 300-cell brain to a 600-megabyte model that will talk from inside a Mercedes, and explains why architecture search now beats architecture avengers.
54:26ENTen Billion Active Parameters, and the Unsexy Work of Making Them Agent-Ready主持 Nathan LabenzMiniMax researcher Olive Song explains how M2 was trained with expert developers as reward models, interleaved thinking, and a perturbation pipeline that treats every scaffold change as the real generalization test.
1:54:30ENFrom One in Ten Cancer Cells to One in a Million主持 Nathan LabenzNathan Labenz's solo AMA updates his son's lymphoma treatment, calls Claude Opus 4.5 software AGI, and ranks the live players of early 2026.
1:18:40ENWhen AI Becomes the Oil Well: Luke Drago on the Intelligence Curse主持 Nathan LabenzThe Workshop Labs co-founder tells the FLI Podcast why a world of non-human production could recreate the resource curse — and how private data, open weights, and moonshot careers might still break it.
1:32:38ENPolish Is 1% of Frontier Training Data, So Poland Built Its Own Model主持 Nathan LabenzMarek Kozlowski of Poland's National Information Processing Institute explains how PLLuM uses language adaptation, organic instructions, and on-prem small models to keep a mid-size country in control of its AI.
1:25:54ENHours after training ends, Z.ai puts the weights on the internet主持 Nathan LabenzZixuan Li of Z.ai tells ChinaTalk how a Chinese lab ships GLM, treats open weights as a ticket into Western ecosystems, and still sees itself as an upstart.
2:14:55ENFrom Character AI to a 1% Warning Shot: Gabriel Weil's Case for Liability as AI Governance主持 Nathan LabenzA Touro law professor tells Nathan Labenz how negligence, strict liability, and punitive damages could make frontier labs treat public risk as a hit to their own balance sheet.
1:30:57EN → 简体中文蚂蚁集团 3200 亿美元 IPO 在最后一刻被叫停:Angela Zhang 谈中国科技的走钢丝主持 Nathan Labenz一位南加州大学法学教授梳理了支配中国科技的那套动态金字塔,从滴滴上市讲到 DeepSeek 的价格战,并解释了她为什么认为美国正开始越来越像中国。
1:08:51ENChip Controls Are Meant to Buy Ten Years, Not Win the AI Race主持 Nathan LabenzEconomist Noah Smith tells Nathan Labenz that U.S. export controls are a piece of conventional deterrence against China — not a plan to kneecap DeepSeek.
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。
1:48:20ENFrom 15% to 70% on Right-or-Wrong Scores: DeepSeek-R1 Laid Reinforcement Learning's Cards on the Table主持 Nathan LabenzOn Inauguration Day 2025, DeepSeek and Moonshot dropped reasoning-model results that Nathan Labenz reads as proof the reinforcement-learning recipe is simple, public, and still climbing.
3:53:49EN → 简体中文数据只有十二分之一,一个开源视觉模型击败了 GPT-4o主持 Nathan LabenzVeratai 的 Will Hardman 带 Nathan Labenz 梳理了两年来的视觉语言模型(VLM),从 ViT 和 CLIP 讲到 InternVL、BLINK 和 Transfusion。
1:57:00ENLess Than 1%: Why OpenAI's roon Thinks the Good Guys Get AGI First主持 Nathan LabenzDoom Debates host Liron Shapira presses OpenAI technical staffer roon on extinction odds, alignment-by-default, and a plan that even he admits is balancing on razor wire.
1:49:41ENA Team of Fifteen Matched Meta’s Post-Training on Meta’s Own Base主持 Nathan LabenzNathan Lambert walks through Tulu 3: the three-stage recipe, the 1 percent versus 14 percent lesson, and why open post-training now lives or dies on data and evals.
2:24:18ENMMLU Stays at Chance Until Late Training. Loss Is Only an Alarm.主持 Nathan LabenzZyphra’s training lead walks through Zamba 2’s hybrid architecture, the evals that actually matter, and the trade secrets small labs still have to rediscover.
1:54:01ENA Tiny Network Grew a Salt Crystal the Training Data Never Showed主持 Nathan LabenzTim Duignan trained tens of thousands of parameters on salt water. The model invented a crystal — and a continuum solvent theory he had spent a PhD failing to write by hand.
1:42:17ENThe Cheapest Inference Provider Is the Most Dangerous主持 Nathan LabenzFireworks AI cofounders Lin Qiao and Dmytro Ivchenko explain why subsidized GPU bills vanish, how LoRA serving can cut cost a hundredfold, and why OpenAI is the rival that actually matters.
2:16:27ENSuperhuman Go Has a Blind Spot. Balaji Bets on a War of the Gods主持 Nathan LabenzBalaji Srinivasan tells Nathan Labenz why adversarial inputs, missing actuators, and cryptography make a single runaway AI god less likely than a contest among many that still need priests.
1:14:36ENZulu Bypasses GPT-4, and a Hundred Examples Undo the Guardrails主持 Nathan LabenzNathan Labenz reads seven papers — four hopeful, three not — on whether anyone can actually keep a language model under control.
52:48ENOpenAI Cut Fine-Tuning 90 Percent, and Founders Crossed Llama 2 Off the List主持 Nathan LabenzNathan Labenz and Erik Torenberg walk through GPT-3.5 Turbo fine-tuning, why self-hosting still loses on total cost, and what is actually working in code, school, and medicine.
59:28ENFrom $500,000 to $50,000: MosaicML Made Training Your Own Model a Priced Product主持 Nathan LabenzJonathan Frankle and Abhi Venigalla explain who is training custom LLMs, why extraction beats chat, and how ALiBi stretches a 65k context window.
加载更多已显示 24 篇,共 27 篇









