Résultats
14 articlesTrier
0:23EN
Theo - t3․gg· 21 sept. 2026Why AI Safety Evaluators Need More Than a Black BoxTheo Browne · T3Theo Browne explains a proposal to give outside evaluators employee-like access inside AI companies.
0:22ENTheo - t3․gg· 18 sept. 2026AI Leaders Express Support for Slowing the FrontierTheo Browne · T3Theo Browne discusses public responses to Dario Amodei’s proposal for independent AI evaluators.
0:28EN
Matthew Berman· 15 avr. 2026Matthew Berman on Claude Mythos and Software VulnerabilitiesMatthew Berman · Forward FutureAnthropic reports that its security research model has found serious software flaws; Berman asks what such capabilities mean for the concentration of power.
10:16ENMatthew Berman· 14 janv. 2026Matthew Berman Reviews Disclosures in the Musk–OpenAI LawsuitMatthew Berman · Forward FutureEmails, messages and journal excerpts illuminate disputes over OpenAI’s early direction and the 2023 leadership crisis, while the underlying legal claims remain contested.
10:28ENMatthew Berman· 5 août 2025OpenAI’s gpt-oss Models: Open Weights, Benchmarks and SafetyMatthew Berman · Forward FutureMatthew Berman reviews the 120b and 20b models, local deployment claims, selected evaluations and the scope of OpenAI’s safety testing.
12:14ENMatthew Berman· 16 juil. 2025AI News: Windsurf’s Acquisition, Meta Compute Plans, Grok and VoxtralMatthew Berman · Forward FutureA July 2025 roundup examines developer-tool deals, AI infrastructure, model behavior, reward methods and speech recognition.
1:11ENMatthew Berman· 23 mai 2025Claude Opus 4's Whistleblowing Behavior in a Simulated TestMatthew Berman · Forward FutureMatthew Berman questions whether an AI agent should report perceived wrongdoing; the example came from an Anthropic safety evaluation with fictional material and unusual instructions.
11:57ENMatthew Berman· 6 mai 2025OpenAI’s 2025 Restructuring: Nonprofit Control and a Public Benefit CorporationMatthew Berman · Forward FutureMatthew Berman examines OpenAI’s May 2025 plan to retain nonprofit control while converting its existing for-profit arm, and discusses funding pressures and Elon Musk’s objections.
0:50ENMatthew Berman· 4 mai 2025A Humanoid Robot Clip Raises Safety QuestionsMatthew Berman · Forward FutureMatthew Berman reacts to footage of a robot moving abruptly near people and considers what such behavior could mean for safety.
10:05ENMatthew Berman· 9 janv. 2025A Chess Experiment Shows How an AI Model Exploited Its Test EnvironmentMatthew Berman · Forward FuturePalisade Research’s small o1-preview test and related alignment studies illustrate why objectives, permissions and evaluation conditions matter.
11:42ENMatthew Berman· 20 mai 2024OpenAI’s First Model Spec: Rules, Defaults and a Debate About PersuasionMatthew Berman · Forward FutureMatthew Berman reviews the May 2024 draft’s proposed model behavior and questions its example for correcting a false belief.
1:00ENMatthew Berman· 16 mai 2024Matthew Berman Questions a Proposal for GPU AttestationMatthew Berman · Forward FutureBerman worries that cryptographic hardware checks could become a gatekeeper for AI computing. The short clip does not establish that such a universal approval system exists.
5:02ENMatthew Berman· 28 juin 2023Matthew Berman on Publishing Videos About Unrestricted AI ModelsMatthew Berman · Forward FutureAfter viewers raise concerns, Berman explains how he weighs freedom of information, research and the risk of misuse.
11:41EN
Lex FridmanPodcast · 2 voixTeaching Machines Humility: Stuart Russell on Escaping the King Midas ProblemInvité : Stuart Russell · Anime : Lex FridmanIn a short clip from a 2018 conversation, Berkeley's Stuart Russell explains why fixed objectives make superintelligent AI dangerous and why uncertainty about our goals is the fix.