최신 글
3,570개정렬
1:00:48ENThe Phrase That Alarms a16z’s AI Policy Chief: “It Only Covers a Handful of Companies”진행자 Nathan LabenzMatt Perault argues that AI rules should target harmful use, not the science of building models, and that “modest” thresholds can lock in an oligopoly.
22:05ENApple’s “Illusion of Thinking”: Puzzle Limits and a Code-Assisted TestMatthew Berman, Ilya Sutskever · Forward FutureMatthew Berman reviews Apple’s reasoning-model experiments and tests a different way to solve Tower of Hanoi using generated code.
1:16:43ENMicha Kaufman on AI, Work and Fiverr’s Next Chapter게스트 Micha Kaufman · 진행자 Harry StebbingsThe Fiverr CEO discusses adapting to AI, startup competition, creative work and the uncertainty facing leaders.
2:02:27ENShe set out to build mirror life. Then she talked the field into stopping진행자 Nathan LabenzUniversity of Minnesota geneticist Kate Adamala tells Nathan Labenz why every known organism shares one chemical alphabet — and why flipping that alphabet would be a branch of synthetic biology better left unexplored.
0:54ENMatthew Berman Demonstrates ChatGPT Connectors and Deep ResearchMatthew Berman · Forward FutureIn a 2025 demonstration, connected services supply information for a natural-language question about recent HubSpot deals.
9:10ENReddit v. Anthropic: The Data-Scraping Allegations and What Happened NextMatthew Berman · Forward FutureMatthew Berman reads Reddit’s 2025 complaint alleging unauthorized access and use of its content. Anthropic disputes the claims, and subsequent procedural rulings have not decided liability.
1:27:13ENSam Lessin, Jason Lemkin and Rory O’Driscoll Debate DPI and the Future of Venture게스트 Sam Lessin, Rory O’Driscoll, Jason Lemkin · 진행자 Harry StebbingsThe investors examine cash distributions versus paper marks, fund size, AI spending and the changing economics of startup investing.
35:47ENFei-Fei Li’s new bet: AI is incomplete without spatial intelligence진행자 Sarah Guo, Elad GilThe ImageNet pioneer tells Sarah Guo and Elad Gil why World Labs is building 3D generative foundation models, and why language was only one piece of the puzzle.
2:12:04ENFrom a Five-Year Phone Queue to 480 Trillion Tokens a Month게스트 Sundar Pichai · 진행자 Lex FridmanSundar Pichai talks with Lex Fridman about Chennai, the AI package, Search, Chrome, Waymo, and why he stayed calm when the headlines said Google was finished.
1:12:29ENGoogle's AI Co-Scientist Solved a Phage Puzzle Humans Missed for Years진행자 Nathan LabenzImperial College scientists José Penadés and Tiago Costa told the Podovirus podcast how a still-unreleased Google system recovered their unpublished answer: tailless capsids that pick up stray phage tails.
44:21ENMercor's CEO: Knowledge Work's Next Job Is Writing Evals진행자 Jacob EffronBrendan Foody tells Redpoint how Mercor turned AI hiring and expert data into a $100 million round — and why he thinks a huge slice of white-collar work will become creating tests for models.
53:51ENVeo 3 Took First Place by Beating Veo 2. Google Labs Says Chapter Two Just StartedThree Google Labs leads walk through generative cameras, background computer-use agents, and notebooks that adapt knowledge for an audience of one.
18:13ENDarwin Gödel Machine: How a Coding Agent Tests Changes to Its Own SoftwareMatthew Berman · Forward FutureMatthew Berman explains Sakana AI’s evolutionary search over agent code, its benchmark results and the limits of claims about recursive self-improvement.
1:04:56ENVarun Mohan on Windsurf, Coding Agents and Startup Focus게스트 Varun Mohan · 진행자 Harry StebbingsThe Windsurf co-founder discusses pivots, rapid product experimentation, coding agents and competition among AI model providers.
47:18ENThomas Dohmke on GitHub Copilot, Coding Agents, and the Future of SoftwareMatthew Berman, Thomas Dohmke · Forward FutureGitHub’s then-CEO explains how Copilot grew from code completion into agent workflows, why programming education still matters, and where he expects human review to remain essential.
10:09ENDeepSeek R1 0528: Benchmark Gains and Two Coding TestsMatthew Berman · Forward FutureMatthew Berman examines the May 2025 DeepSeek reasoning-model update, comparing reported benchmark gains with his own Rubik’s Cube and Snake programming tests.
1:19:24ENOpenAI’s Jony Ive Deal, YC’s Scale and the Math Behind Venture Returns게스트 Rory O’Driscoll, Jason Lemkin · 진행자 Harry StebbingsHarry Stebbings, Rory O’Driscoll and Jason Lemkin debate hardware, IPO terms, dilution and the changing economics of venture investing.
25:13ENThe AI Table Is Being Cleared: Two or Three Coding Winners, and Forty Years Without a Biotech Giant진행자 Sarah Guo, Elad GilSarah Guo and Elad Gil walk through which AI markets have already consolidated, why rival startups still refuse to merge, and the biotech science sitting unused.
1:54:08ENWhy Anthropic's first check went to a company building microscopes for neural nets진행자 Nathan LabenzGoodfire's Dan Balsam and Tom McGrath argue that mechanistic interpretability now has a proto-paradigm, two stubborn gaps, and three products worth shipping today.
1:05ENSatya Nadella on AI, Productivity, and Cancer CareMatthew Berman, Satya Nadella · Forward FutureIn a short interview excerpt, the Microsoft CEO discusses the potential of cheaper AI to help coordinate healthcare work.
1:19:40ENIan Henry on Building Trader Tools, Functional UIs and Creative CodingYaron Minsky, Ian Henry · Jane StreetFrom high-density options interfaces to a 3D playground, Ian Henry and Ron Minsky explore the software choices that make complex tools usable.
15:55ENAI 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.
26:01ENTesting Veo 3: Convincing Video, Generated Audio and Physics FailuresMatthew Berman · Forward FutureMatthew Berman and a collaborator test Google’s 2025 video model with street interviews, cinematic scenes, visual puzzles and edited short films.
38:25ENA thousand airplanes will never train an AI engineerFormer Airbus CTO Paul Eremenko explains why physical-world engineering still has no coding-AI moment, and how P-1 AI's agent Archie is supposed to get there anyway.
더 보기3,570개 중 24개 표시









