Latest articles
3,960 articlesSort
7:41ENMicrosoft’s Majorana 1: The Quantum Computing Claim and Its LimitsMatthew Berman · Forward FutureMatthew Berman explains Microsoft’s proposed topological qubits and imagines future uses, while the published experiment supports narrower conclusions.
1:03ENMicrosoft’s Majorana 1 Quantum Chip: What Was Claimed and What Was DemonstratedMatthew Berman · Forward FutureMatthew Berman discusses Microsoft’s 2025 Majorana 1 announcement and its possible implications for quantum computing and cryptography.
0:48ENWhy AI Threat Models May Be Hard to TestA short discussion of how a model that recognizes testing could behave differently in the lab.
0:51ENClone Robotics’ Protoclone and the Promises of a Home AndroidMatthew Berman · Forward FutureMatthew Berman reacts to a musculoskeletal prototype and a proposed limited-run home robot; the advertised household skills remain unverified in the clip.
46:23ENAdarsh Hiremath on Mercor, AI Recruiting and the Future of WorkGuest Adarsh Hiremath · Host Harry StebbingsThe Mercor co-founder discusses expert labor for AI labs, automated interviews, fundraising and his vision for a global labor marketplace.
9:43ENTesting Grok 3 on Games, Reasoning, and Web SearchMatthew Berman · Forward FutureMatthew Berman tries a new set of audience-suggested prompts, from a Snake–Flappy Bird mashup to a family-logic puzzle and a tesseract visualization.
3:18:45EN5,000 Samples Topped ARC-AGI. Then Claude Faked Alignment to Keep Its ValuesHost Nathan LabenzRedwood Research chief scientist Ryan Greenblatt walks through inference scaling, alignment faking, and what it would mean to strike real deals with AIs.
38:14ENKyle Vogt Pitched 120 Investors for Cruise. He Will Not Sell the Home-Robot CompanyHost Sarah Guo, Elad GilThe Twitch and Cruise co-founder tells Sarah Guo and Elad Gil why Tesla won on business model, why chores should be automated, and why a tiny AI-era team is enough.
0:32ENFigure’s Humanoid Robots and Natural-Language InstructionsMatthew Berman · Forward FutureMatthew Berman comments on a 2025 Figure Robotics demonstration and the potential for household chores.
44:34ENThe Cheaper Intelligence Gets, the More of It We'll BuyHost 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.
0:51ENMira Murati Announces Thinking Machines LabMatthew Berman · Forward FutureMatthew Berman reviews the new AI lab’s stated goals in its February 2025 announcement.
1:00:09ENDeepSeek's Jailbreaks Don't Matter. Giving the Model Arms and Legs DoesPalo Alto Networks CEO Nikesh Arora tells Sequoia's Training Data why cheap brains scramble the AI order, why attackers already move faster than defenders, and why refusing to experiment is the real risk.
13:24ENGrok 3’s 2025 Launch: Benchmarks, Reasoning and DeepSearchMatthew Berman, Elon Musk · Forward FutureMatthew Berman reviews xAI’s launch charts, a historical user-preference leaderboard and an early product demonstration.
1:25:52ENJonathan Ross on Groq, Inference Economics and the AI Infrastructure RaceGuest Jonathan Ross · Host Harry StebbingsThe Groq CEO explains his bet on inference chips, compares his company’s architecture with GPUs, and discusses the financing and power needed to scale.
53:07ZH → EnglishZhang Yaqin on the AGI Roadmap: Information, Physical, and Biological IntelligenceGuest 张亚勤 · Host 张小珺Zhang Yaqin divides AGI’s development into information, physical, and biological intelligence. Information intelligence may approach human levels within five years, general-purpose robots may take about ten, and biological intelligence could expand human memory and health in twenty. He also discusses how large models can give autonomous vehicles and robots more data, common sense, and ability to act.
18:05ENOpenAI’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.
3:20:53ZH → EnglishHe Junxian Explains DeepSeek’s Papers: From Base Models to ReasoningGuest 何俊贤 · Host 张小珺Follow the research through MoE, attention caches, training efficiency and reinforcement learning.
45:31ENJerry Chow on Quantum Computing, Encryption, AI and Error CorrectionMatthew Berman, Jerry Chow · Forward FutureIBM’s quantum systems director explains qubits, cryogenic hardware, software and the work needed to make quantum computers useful.
1:29:02ENShe Never Sees a Row of Your Data. The Answers Still Clear 95%.Host Nathan LabenzIllumex CEO Inna Tokarev Sela tells Nathan Labenz how a Platonic industry ontology, trained only on metadata, collapses enterprise software into one conversation.
1:07ENHow Interacting AI Agents Could Change the Wider SystemA speaker argues that simultaneous AI adoption could create unexpected interactions between agents.
8:04ENOpenAI’s February 2025 GPT-5 Roadmap and the Case for a Unified InterfaceMatthew Berman · Forward FutureMatthew Berman explains the proposed model-routing experience, GPT-4.5 and GPT-5 plans, and the product choices that prompted them.
1:20:53ENSolly Solomou on Building LADbible, Social Media and Expanding to the USGuest Solly Solomou · Host Harry StebbingsThe founder recounts buying a Facebook page, creating a media business and learning to lead through rapid change.
49:36EN86 Out of 100: Harvey's Playbook for Selling AI Into LawHost Sarah Guo, Elad GilWinston Weinberg on a Reddit blind test, expanding-and-collapsing product design, and why Harvey started with century-old firms instead of the mid-market.
44:26ENAn AI Application Is Still an ApplicationMongoDB's Sahir Azam tells Sequoia why vector search is a new primitive for unstructured data, not a replacement for transactions, and why retrieval quality is the last mile that gets AI into conservative enterprises.
Load moreShowing 24 of 3,960









