Résultats
109 articlesTrier
30:11EN
Sequoia Capital· 30 juil. 2025PodcastFrom 12 Percent on AIME to IMO Gold in Two MonthsAlex Wei, Sheryl Hsu, and Noam Brown tell Sequoia's Sonya Huang how a three-person OpenAI sprint used general-purpose reinforcement learning, not Lean, to reach gold-level on the 2025 International Mathematical Olympiad.
6:53EN
Matthew Berman· 24 juil. 2025Qwen3-Coder: An Open-Weight Coding Model and Its Command-Line AgentMatthew Berman · Forward FutureMatthew Berman examines Qwen’s July 2025 coding model, its training and benchmark results, and tries its command-line tool on a Snake game.
51:06EN
20VC with Harry Stebbings· 18 juil. 2025Podcast · 2 voixScott Wu on Cognition's Windsurf Deal and the Future of AI CodingInvité : Scott Wu · Anime : Harry StebbingsThe Cognition CEO explains the fast-moving acquisition, the role of AI agents in engineering, and his predictions for software development.
1:02:54EN
No Priors: AI, Machine Learning, Tech, & Startups· 17 juil. 2025Podcast · 2 voixEnterprise Coding Tools Flipped the 80/20. Reflection AI Built the Other HalfAnime : Sarah Guo, Elad GilFormer Gemini reward-model lead Misha Laskin on Asimov, why comprehension beats generation, and why there is no such thing as generalization—only pulling the test distribution into train.
38:10EN
Cognitive Revolution "How AI Changes Everything"· 16 juil. 2025PodcastPlus Five for Format, Plus Ten if It CompilesAnime : 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.
15:55ENMatthew Berman· 28 mai 2025AI Shutdown Resistance and Blackmail in Controlled Safety TestsMatthew Berman · Forward FutureMatthew Berman examines Palisade’s shutdown experiment, Apollo’s simulated scheming evaluations and Anthropic’s Claude 4 system card.
57:31EN
Unsupervised Learning: With Jacob Effron· 22 mai 2025PodcastThe New Ruler for Models Is How Long They Can WorkAnime : Jacob EffronOn Claude 4's launch day, Anthropic researcher Sholto Douglas argued that the next phase of AI is less about harder puzzles and more about hours of unsupervised work.
15:13ENMatthew Berman· 10 mai 2025Absolute Zero: Training a Reasoner on Self-Generated Coding TasksMatthew Berman · Forward FutureMatthew Berman examines a 2025 paper in which a pretrained language model proposes and solves code tasks using verifiable feedback.
38:10ENNo Priors: AI, Machine Learning, Tech, & Startups· 1 mai 2025Podcast · 2 voixo3 should wait exactly as long as the answer takesAnime : Sarah Guo, Elad GilOpenAI researchers Eric Mitchell and Brandon McKinzie tell No Priors why o3's leap is a chain of tools during reasoning — and why you should send the same prompt a hundred times.
30:46ENNo Priors: AI, Machine Learning, Tech, & Startups· 24 avr. 2025Podcast · 2 voixDeep Research Turned Web Browsing into a Trainable ProblemAnime : Sarah Guo, Elad GilOpenAI researcher Isa Fulford explains why a side project on read-only synthesis became an agent, when reinforcement fine-tuning is worth it, and why models will hack the search engine you gave them.
1:14:11ENCognitive Revolution "How AI Changes Everything"· 5 avr. 2025PodcastThe Reasoning Curve Was Smooth All Along: Gemini 2.5's Jack Rae on Thinking, Memory, and AGIAnime : Nathan LabenzDeepMind's thinking-and-inference lead explains why RL-for-reasoning felt sudden, why Gemini shows raw chain of thought, and why a million-token window still is not memory.
1:56:38ENCognitive Revolution "How AI Changes Everything"· 8 mars 2025Podcasto3 took ARC from 20% to 87%. Nathan Labenz sketches early superintelligenceAnime : Nathan LabenzOn Consistently Candid, the Cognitive Revolution host argues that math and coding will be superhuman by 2026, and that a reasoning core plus specialist “intuitive physics” may already be enough.
12:30ENMatthew Berman· 7 mars 2025QwQ-32B: A Smaller Reasoning Model and Its TradeoffsMatthew Berman · Forward FutureMatthew Berman examines Qwen’s reinforcement-learning recipe, tests a coding task, and compares benchmark scores and hardware requirements with DeepSeek-R1.
32:46ENSequoia Capital· 25 févr. 2025PodcastDon't draw the flowchart: OpenAI's Deep Research team on training agents end to endIsa Fulford and Josh Tobin explain why Deep Research was trained with reinforcement learning on hard browsing tasks instead of a human-written graph of model calls.
17:20ENMatthew Berman· 23 févr. 2025Vibe Coding a Login Page and an AI Tetris PlayerMatthew Berman · Forward FutureMatthew Berman experiments with AI-written code, iterates through failures and trains a basic game-playing agent.
44:34ENUnsupervised Learning: With Jacob Effron· 19 févr. 2025PodcastThe Cheaper Intelligence Gets, the More of It We'll BuyAnime : Jacob EffronAmazon AGI Lab head David Luan on what DeepSeek actually meant, why next-token prediction is not enough, and why agents are still stuck on reliability.
18:05ENMatthew Berman· 16 févr. 2025OpenAI’s Coding-Competition Study: o1-ioi, o3 and the Limits of the ResultsMatthew Berman, Sam Altman · Forward FutureMatthew Berman interprets OpenAI’s 2025 paper on reinforcement learning and programming contests. The results are strong on the studied benchmarks, but do not establish AGI or remove humans from model development.
6:50ENMatthew Berman· 12 févr. 2025DeepScaleR: A Small Reasoning Model Tested on Competition MathMatthew Berman · Forward FutureMatthew Berman reviews a 1.5B model’s reported benchmark results and runs a quantized version locally on one AIME problem.
21:10ENMatthew Berman· 11 févr. 2025Sam Altman’s 2025 AI Coding Forecast and Three ObservationsMatthew Berman, Sam Altman · Forward FutureMatthew Berman examines predictions about reasoning models, software agents and AI economics, distinguishing competitive-programming results from broader work.
2:16:17ENCognitive Revolution "How AI Changes Everything"· 5 févr. 2025PodcastA 10B Model Trained Across Three Continents — and an H200 for $1.49 an HourAnime : Nathan LabenzPrime Intellect's founders lay out a four-step plan to turn spare global GPUs into collectively owned intelligence, and argue that reinforcement learning just made their job easier.
10:01ENMatthew Berman· 5 févr. 2025DeepSeek R1 Code Optimization and a $3 Visual Counting ExperimentMatthew Berman · Forward FutureMatthew Berman examines a human-guided coding improvement and a small reinforcement-learning experiment, then considers their implications for AI research.
2:49:35ZH
Zhang Xiaojun Podcast· 4 févr. 2025Podcast · 2 voix潘嘉怡解读 R1、o1 与 Kimi K1.5：强化学习如何教模型学会推理Invité : 潘嘉怡 · Anime : 张小珺伯克利博士生潘嘉怡带读 DeepSeek、OpenAI 与 Kimi 的推理模型报告，拆解长思维链、GRPO、奖励设计与蒸馏。R1-Zero 的“顿悟时刻”展示了推理策略如何在强化学习中涌现。
12:07ENMatthew Berman· 2 févr. 2025TinyZero: A $30 Reinforcement-Learning Experiment on a Small Arithmetic TaskMatthew Berman · Forward FutureMatthew Berman explains how a pretrained Qwen model learned to search and check answers in Countdown, and why the experiment does not recreate DeepSeek-R1.
52:30ENSequoia Capital· 28 janv. 2025PodcastMove 37 looked like a bug. LLMs still need their AlphaZero momentIoannis Antonoglou, from AlphaGo to Reflection AI, on self-play, reliability, and why compute can still buy intelligence.Charger plus24 sur 109 affichés