#211 - Claude Voice, Flux Kontext, wrong RL research?

Our 213nd episode with a summary and discussion of last week's big AI news! Recorded on 06/21/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. In this episode: Midjourney launches its first AI video generation model, moving from text-to-image to video with a subscription model offering up to 21-second clips, highlighting the affordability and growing capabilities in AI video generation. Google's Gemini AI family updates include high-efficiency models for cost-effective workloads, and new enhancements in Google's search function now allow for voice interactions. The introduction of two new benchmarks, Live Code Bench Pro and Abstention Bench, aiming to test and improve the problem-solving and abstention capabilities of reasoning models, revealing current limitations. OpenAI wins a $200 million US defense contract to support various aspects of the Department of Defense, reflecting growing collaborations between tech companies and government for AI applications. Timestamps + Links: (00:00:10) Intro / Banter (00:01:32) News Preview Tools & Apps (00:02:12) Midjourney launches its first AI video generation model, V1 (00:05:52) Google’s Gemini AI family updated with stable 2.5 Pro, super-efficient 2.5 Flash-Lite (00:07:59) Google’s AI Mode can now have back-and-forth voice conversations (00:10:13) YouTube to Add Google’s Veo 3 to Shorts in Move That Could Turbocharge AI on the Video Platform Applications & Business (00:11:10) The ‘OpenAI Files’ will help you understand how Sam Altman’s company works (00:12:29) OpenAI drops Scale AI as a data provider following Meta deal (00:13:28) Amazon’s Zoox opens its first major robotaxi production facility Projects & Open Source (00:15:20) LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming? (00:19:45) AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions (00:22:49) MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention Research & Advancements (00:24:33) Scaling Laws of Motion Forecasting and Planning -- A Technical Report Policy & Safety (00:28:07) Universal Jailbreak Suffixes Are Strong Attention Hijackers (00:30:52) OpenAI found features in AI models that correspond to different ‘personas’ (00:33:25) OpenAI wins $200 million U.S. defense contract

Our 212th episode with a summary and discussion of last week's big AI news! Recorded on 06/13/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. In this episode: OpenAI introduces O3 PRO for ChatGPT, highlighting significant improvements in performance and cost-efficiency. Anthropic sees an influx of talent from OpenAI and DeepMind, with significantly higher retention rates and competitive advantages in AI capabilities. New research indicates that reinforcing negative responses in LLMs significantly improves performance across all metrics, highlighting novel approaches in reinforcement learning. A security flaw in Microsoft Copilot demonstrates the growing risk of AI agents being hacked, emphasizing the need for robust protection against zero-click attacks. Timestamps + Links: (00:00:11) Intro / Banter (00:01:31) News Preview (00:02:46) Response to Listener Reviews Tools & Apps (00:04:48) OpenAI adds o3 Pro to ChatGPT and drops o3 price by 80 per cent, but open-source AI is delayed (00:09:10) Cursor AI editor hits 1.0 milestone, including BugBot and high-risk background agents (00:13:07) Mistral releases a pair of AI reasoning models (00:16:18) Elevenlabs' Eleven v3 lets AI voices whisper, laugh and express emotions naturally (00:19:00) ByteDance's Seedance 1.0 is trading blows with Google's Veo 3 (00:22:42) Google Reveals $20 AI Pro Plan With Veo 3 Fast Video Generator For Budget Creators Applications & Business (00:25:42) OpenAI and DeepMind are losing engineers to Anthropic in a one-sided talent war (00:34:32) OpenAI slams court order to save all ChatGPT logs, including deleted chats (00:37:24) Nvidia’s Biggest Chinese Rival Huawei Struggles to Win at Home (00:43:06) Huawei Expected to Break Semiconductor Barriers with Development of High-End 3nm GAA Chips; Tape-Out by 2026 (00:45:21) TSMC’s 1.4nm Process, Also Called Angstrom, Will Make Even The Most Lucrative Clients Think Twice When Placing Orders, With An Estimate Claiming That Each Wafer Will Cost $45,000 (00:47:43) Mistral AI Launches Mistral Compute To Replace Cloud Providers from US, China Projects & Open Source (00:51:26) ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models Research & Advancements (00:57:27) Kinetics: Rethinking Test-Time Scaling Laws (01:05:12) The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning (01:10:45) Predicting Empirical AI Research Outcomes with Language Models (01:15:02) EXP-Bench: Can AI Conduct AI Research Experiments? Policy & Safety (01:20:07) Large Language Models Often Know When They Are Being Evaluated (01:24:56) Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence (01:31:16) Exclusive: New Microsoft Copilot flaw signals broader risk of AI agents being hacked—‘I would be terrified’ (01:35:01) Claude Gov Models for U.S. National Security Customers Synthetic Media & Art (01:37:32) Disney And NBCUniversal Sue AI Company Midjourney For Copyright Infringement (01:40:39) AMC Networks is teaming up with AI company Runway

Our 211th episode with a summary and discussion of last week's big AI news! Recorded on 05/31/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: Recent AI podcast covers significant AI news: startups, new tools, applications, investments in hardware, and research advancements. Discussions include the introduction of various new tools and applications such as Flux's new image generating models and Perplexity's new spreadsheet and dashboard functionalities. A notable segment focuses on OpenAI's partnership with the UAE and discussions on potential legislation aiming to prevent states from regulating AI for a decade. Concerns around model behaviors and safety are discussed, highlighting incidents like Claude Opus 4's blackmail attempt and Palisade Research's tests showing AI models bypassing shutdown commands. Timestamps + Links: (00:00:10) Intro / Banter (00:01:39) News Preview (00:02:50) Response to Listener Comments Tools & Apps (00:07:10) Anthropic launches a voice mode for Claude (00:10:35) Black Forest Labs’ Kontext AI models can edit pics as well as generate them (00:15:30) Perplexity’s new tool can generate spreadsheets, dashboards, and more (00:18:43) xAI to pay Telegram $300M to integrate Grok into the chat app (00:22:42) Opera’s new AI browser promises to write code while you sleep (00:24:17) Google Photos debuts redesigned editor with new AI tools Applications & Business (00:25:13) Top Chinese memory maker expected to abandon DDR4 manufacturing at the behest of Beijing (00:30:04) Oracle to Buy $40 Billion Worth of Nvidia Chips for First Stargate Data Center (00:31:47) UAE makes ChatGPT Plus subscription free for all residents as part of deal with OpenAI (00:35:34) NVIDIA Corporation (NVDA) to Launch Cheaper Blackwell AI Chip for China, Says Report (00:38:39) The New York Times and Amazon ink AI licensing deal Projects & Open Source (00:41:11) DeepSeek’s distilled new R1 AI model can run on a single GPU (00:45:19) Google Unveils SignGemma, an AI Model That Can Translate Sign Language Into Spoken Text (00:47:08) Open-sourcing circuit tracing tools (00:49:42) Hugging Face unveils two new humanoid robots Research & Advancements (00:52:33) PANGU PRO MOE: MIXTURE OF GROUPED EXPERTS FOR EFFICIENT SPARSITY (00:58:55) DataRater: Meta-Learned Dataset Curation (01:05:05) Incorrect Baseline Evaluations Call into Question Recent LLM-RL Claims  (01:10:17) Maximizing Confidence Alone Improves Reasoning (01:11:00) Guided by Gut: Efficient Test-Time Scaling with Reinforced Intrinsic Confidence (01:11:44) One RL to See Them All (01:15:05) Efficient Reinforcement Finetuning via Adaptive Curriculum Learning Policy & Safety (01:17:58) Trump's 'Big Beautiful Bill' could ban states from regulating AI for a decade (01:24:31) Researchers claim ChatGPT o3 bypassed shutdown in controlled test (01:30:10) Anthropic’s new AI model turns to blackmail when engineers try to take it offline (01:31:09) Anthropic Faces Backlash As Claude 4 Opus Can Autonomously Alert Authorities (01:35:37) Claude helps users make bioweapons (01:35:49) The Claude 4 System Card is a Wild Read  

Our 210th episode with a summary and discussion of last week's big AI news! Recorded on 05/23/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: Google's Gemini diffusion technology showcases significant improvements in speed and efficiency for generating text, potentially revolutionizing the auto-regressive generation paradigm. Anthropic activates AI Safety Level 3 protections for Claude Opus 4, implementing robust measures such as bug bounties, synthetic jailbreak data, and preliminary egress bandwidth controls to mitigate bio-risk threats. OpenAI responds to the California Attorney General, refuting claims by the not-for-private-gain coalition and defending their controversial restructuring plans amidst ongoing criticism. Mistral delays the release of its Llama 4 Behemoth model due to training challenges, while Meta faces similar obstacles in rolling out its large-scale AI models, signaling difficulties in reaching frontier level performance. Timestamps + Links: (00:00:00) Intro / Banter (00:01:43) News Preview Tools & Apps (00:02:58) Anthropic’s new Claude 4 AI models can reason over many steps (00:09:58) Google Unveils A.I. Chatbot, Signaling a New Era for Search (00:14:04) Google rolls out Project Mariner, its web-browsing AI agent (00:16:40) Veo 3 can generate videos — and soundtracks to go along with them (00:21:26) Imagen 4 is Google’s newest AI image generator (00:23:15) Google Meet is getting real-time speech translation (00:25:36) Google’s new Jules AI agent will help developers fix buggy code (00:26:43) GitHub’s new AI coding agent can fix bugs for you (00:28:50) Mistral’s new Devstral model was designed for coding Applications & Business (00:29:53) OpenAI Unites With Jony Ive in $6.5 Billion Deal to Create A.I. Devices (00:36:10) OpenAI’s planned data center in Abu Dhabi would be bigger than Monaco (00:41:18) LM Arena, the organization behind popular AI leaderboards, lands $100M (00:45:21) Nvidia CEO says next chip after H20 for China won't be from Hopper series (00:46:39) Google’s Gemini AI app has 400M monthly active users (00:51:15) AI Servers: End demand intact, but rising gap between upstream build and system production (2025.5.18) Projects & Open Source (00:53:46) Meta Is Delaying the Rollout of Its Flagship AI Model Research & Advancements (00:57:53) Gemini Diffusion (01:03:07) Chain-of-Model Learning for Language Model (01:09:16) Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space (01:15:38) Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training (01:20:16) Lessons from Defending Gemini Against Indirect Prompt Injections (01:23:35) How Fast Can Algorithms Advance Capabilities? (01:30:20) Reinforcement Learning Finetunes Small Subnetworks in Large Language Models Policy & Safety (01:31:12) Exclusive: What OpenAI Told California's Attorney General (01:38:25) Activating AI Safety Level 3 Protections

Our 209th episode with a summary and discussion of last week's big AI news! Recorded on 05/16/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: OpenAI has decided not to transition from a nonprofit to a for-profit entity, instead opting to become a public benefit corporation influenced by legal and civic discussions. Trump administration meetings with Saudi Arabia and the UAE have opened floodgates for AI deals, leading to partnerships with companies like Nvidia and aiming to bolster AI infrastructure in the Middle East. DeepMind introduced Alpha Evolve, a new coding agent designed for scientific and algorithmic discovery, showing improvements in automated code generation and efficiency. OpenAI pledges greater transparency in AI safety by launching the Safety Evaluations Hub, a platform showcasing various safety test results for their models. Timestamps + Links: (00:00:00) Intro / Banter (00:01:41) News Preview (00:02:26) Response to listener comments Applications & Business (00:03:00) OpenAI says non-profit will remain in control after backlash (00:13:23) Microsoft Moves to Protect Its Turf as OpenAI Turns Into Rival (00:18:07) TSMC’s 2nm Process Said to Witness ‘Unprecedented’ Demand, Exceeding 3nm Due to Interest from Apple, NVIDIA, AMD, & Many Others (00:21:42) NVIDIA’s Global Headquarters Will Be In Taiwan, With CEO Huang Set To Announce Site Next Week, Says Report (00:23:58) CoreWeave in Talks for $1.5 Billion Debt Deal 6 Weeks After IPO Tools & Apps (00:26:39) The Day Grok Told Everyone About ‘White Genocide’ (00:32:58) Figma releases new AI-powered tools for creating sites, app prototypes, and marketing assets (00:36:12) Google’s bringing Gemini to your car with Android Auto (00:38:49) Google debuts an updated Gemini 2.5 Pro AI model ahead of I/O (00:45:09) Hugging Face releases a free Operator-like agentic AI tool Projects & Open Source (00:47:42) Stability AI releases an audio-generating model that can run on smartphones (00:50:47) Freepik releases an ‘open’ AI image generator trained on licensed data (00:54:22) AM-Thinking-v1: Advancing the Frontier of Reasoning at 32B Scale (01:01:29) BLIP3-o: A Family of Fully Open Unified Multimodal Models-Architecture, Training and Dataset Research & Advancements (01:05:40) DeepMind claims its newest AI tool is a whiz at math and science problems (01:12:31) Absolute Zero: Reinforced Self-play Reasoning with Zero Data (01:19:44) How far can reasoning models scale? (01:26:47) HealthBench: Evaluating Large Language Models Towards Improved Human Health Policy & Safety (01:34:10) Trump administration officially rescinds Biden’s AI diffusion rules (01:37:08) Trump’s Mideast Visit Opens Floodgate of AI Deals Led by Nvidia (01:44:04) Scaling Laws For Scalable Oversight (01:49:43) OpenAI pledges to publish AI safety test results more often

Our 208th episode with a summary and discussion of last week's big AI news! Recorded on 05/02/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: OpenAI showcases new integration capabilities in their API, enhancing the performance of LLMs and image generators with updated functionalities and improved user interfaces. Analysis of OpenAI's preparedness framework reveals updates focusing on biological and chemical risks, cybersecurity, and AI self-improvement, while tone down the emphasis on persuasion capabilities. Anthropic's research highlights potential security vulnerabilities in AI models, demonstrating various malicious use cases such as influence operations and hacking tool creation. A detailed examination of AI competition between the US and China reveals China's impending capability to match the US in AI advancement this year, emphasizing the impact of export controls and the importance of geopolitical strategy. Timestamps + Links: Tools & Apps (00:02:57) Anthropic lets users connect more apps to Claude (00:08:20) OpenAI undoes its glaze-heavy ChatGPT update (00:15:16) Baidu ERNIE X1 and 4.5 Turbo boast high performance at low cost (00:19:44) Adobe adds more image generators to its growing AI family (00:24:35) OpenAI makes its upgraded image generator available to developers (00:27:01) xAI’s Grok chatbot can now ‘see’ the world around it Applications & Business: (00:28:41) Thinking Machines Lab CEO Has Unusual Control in Andreessen-Led Deal (00:33:36) Chip war heats up: Huawei 910C emerges as China’s answer to US export bans (00:34:21) Huawei to Test New AI Chip (00:40:17) ByteDance, Alibaba and Tencent stockpile billions worth of Nvidia chips (00:43:59) Speculation mounts that Musk will raise tens of billions for AI supercomputer with 1 million GPUs: Report Projects & Open Source: (00:47:14) Alibaba unveils Qwen 3, a family of ‘hybrid’ AI reasoning models (00:54:14) Intellect-2 (01:02:07) BitNet b1.58 2B4T Technical Report (01:05:33) Meta AI Introduces Perception Encoder: A Large-Scale Vision Encoder that Excels Across Several Vision Tasks for Images and Video Research & Advancements: (01:06:42) The Leaderboard Illusion (01:12:08) Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? (01:18:38) Reinforcement Learning for Reasoning in Large Language Models with One Training Example (01:24:40) Sleep-time Compute: Beyond Inference Scaling at Test-time Policy & Safety: (01:28:23) Every AI Datacenter Is Vulnerable to Chinese Espionage, Report Says (01:32:27) OpenAI preparedness framework update (01:38:31) Detecting and Countering Malicious Uses of Claude: March 2025 (01:46:33) Chinese AI Will Match America's

Our 207th episode with a summary and discussion of last week's big AI news! Recorded on 04/14/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: OpenAI introduces GPT-4.1 with optimized coding and instruction-following capabilities, featuring variants like GPT-4.1 Mini and Nano, and a million-token context window. Concerns arise as OpenAI reduces resources for safety testing, sparking internal and external criticisms. XAI's newly launched API for Grok 3 showcases significant capabilities comparable to other leading models. Meta faces allegations of aiding China in AI development for business advantages, with potential compliances and public scrutiny looming. Timestamps + Links: Tools & Apps (00:03:13) OpenAI’s new GPT-4.1 AI models focus on coding (00:08:12) ChatGPT will now remember your old conversations (00:11:16) Google’s newest Gemini AI model focuses on efficiency (00:14:27) Elon Musk’s AI company, xAI, launches an API for Grok 3 (00:18:35) Canva is now in the coding and spreadsheet business (00:20:31) Meta’s vanilla Maverick AI model ranks below rivals on a popular chat benchmark Applications & Business (00:25:46) Ironwood: The first Google TPU for the age of inference (00:34:15) Anthropic rolls out a $200-per-month Claude subscription (00:37:17) OpenAI co-founder Ilya Sutskever’s Safe Superintelligence reportedly valued at $32B (00:40:20) Mira Murati’s AI startup gains prominent ex-OpenAI advisers (00:42:52) Hugging Face buys a humanoid robotics startup (00:44:58) Stargate developer Crusoe could spend $3.5 billion on a Texas data center. Most of it will be tax-free. Projects & Open Source (00:48:14) OpenAI Open Sources BrowseComp: A New Benchmark for Measuring the Ability for AI Agents to Browse the Web Research & Advancements (00:56:09) Sample, Don't Search: Rethinking Test-Time Alignment for Language Models (01:03:32) Concise Reasoning via Reinforcement Learning (01:09:37) Going beyond open data – increasing transparency and trust in language models with OLMoTrace (01:15:34) Independent evaluations of Grok-3 and Grok-3 mini on our suite of benchmarks Policy & Safety (01:17:58) OpenAI countersues Elon Musk, calls for enjoinment from ‘further unlawful and unfair action’ (01:24:33) OpenAI slashes AI model safety testing time (01:27:55) Ex-OpenAI staffers file amicus brief opposing the company’s for-profit transition (01:32:25) Access to future AI models in OpenAI’s API may require a verified ID (01:34:53) Meta whistleblower claims tech giant built $18 billion business by aiding China in AI race and undermining U.S. national security

Our 206th episode with a summary and discussion of last week's big AI news! Recorded on 04/07/2025 Try out the Astrocade demo here! https://www.astrocade.com/ Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: Meta releases LlAMA-4, a series of advanced large language models, sparking debate on performance and release timing, with models featuring up to 2 trillion parameters for different configurations and applications. Amazon's AGI Lab debuts NOVA Act, an AI agent for web browser control, boasting competitive benchmarking against OpenAI's and Anthropic's best agents. OpenAI's image generation capabilities and ongoing financing developments, notably a $40 billion funding round led by SoftBank, highlight significant advancements and strategic shifts in the tech giant’s operations. Timestamps + Links: (00:00:00) Intro / Banter Tools & Apps (00:01:46) Meta releases Llama 4, a new crop of flagship AI models (00:13:55) Amazon unveils Nova Act, an AI agent that can control a web browser (00:17:06) Alibaba Preparing for Flagship AI Model Release as Soon as April (00:17:59) Runway releases an impressive new video-generating AI model (00:19:10) Adobe launches Premiere Pro’s generative AI video extender (00:20:54) OpenAI prepares reasoning slider and memory update for ChatGPT users Applications & Business (00:21:28) Nvidia H20 Chips: $16 Billion Orders from ByteDance, Alibaba, and Tencent (00:24:45) Elon Musk sells X for $33 billion to his own AI startup company xAI (00:28:00) SoftBank dethroned Microsoft as OpenAI's largest investor, pushing the ChatGPT maker's market cap to $300 billion — but reportedly buried itself in debt (00:30:48) DeepMind is holding back release of AI research to give Google an edge (00:34:06) SMIC Is Rumored To Complete 5nm Chip Development By 2025; Costs Could Be Up To 50 Percent Higher Than TSMC’s Version Due To The Use Of Older-Generation Equipment (00:36:04) Google-backed Isomorphic Labs raises $600m to advance AI drug discovery Research & Advancements (00:38:03) PaperBench: Evaluating AI's Ability to Replicate AI Research (00:43:50) Crossing the Reward Bridge: Expanding RL with Verifiable Rewards Across Diverse Domains (00:48:39) Inference-Time Scaling for Complex Tasks: Where We Stand and What Lies Ahead (00:54:34) Overtrained Language Models Are Harder to Fine-Tune Policy & Safety (00:58:28) Taking a responsible path to AGI (01:02:32) This A.I. Forecast Predicts Storms Ahead (01:06:24) The Secrets and Misdirection Behind Sam Altman’s Firing From OpenAI

Our 205th episode with a summary and discussion of last week's big AI news! Recorded on 03/28/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: OpenAI's new image generation capabilities represent significant advancements in AI tools, showcasing impressive benchmarks and multimodal functionalities. OpenAI is finalizing a historic $40 billion funding round led by SoftBank, and Sam Altman shifts focus to technical direction while COO Brad Lightcap takes on more operational responsibilities., Anthropic unveils groundbreaking interpretability research, introducing cross-layer tracers and showcasing deep insights into model reasoning through applications on Claude 3.5. New challenging benchmarks such as ARC AGI 2 and complex Sudoku variations aim to push the boundaries of reasoning and problem-solving capabilities in AI models. Timestamps + Links: (00:00:00) Intro / Banter (00:01:01) News Preview Tools & Apps (00:02:46) Gemini 2.5: Our most intelligent AI model (00:08:41) OpenAI rolls out image generation powered by GPT-4o to ChatGPT (00:16:14) Ideogram presents version 3.0 of its AI image generation system (00:19:20) New Reve Image Generator Beats AI Art Heavyweights MidJourney and Flux at a Penny Per Image (00:21:56) Alibaba Releases Qwen2.5 Omni, Adds Voice and Video Modes to Qwen Chat (00:23:58) The official version of Tencent's Hunyuan Deep Thinking Model T1 is here, with fast articulation, instant responses, and a decoding speed increase of 2 times Applications & Business (00:25:45) OpenAI Close to Finalizing $40 Billion SoftBank-Led Funding (00:29:26) OpenAI reshuffles leadership as Sam Altman pivots to technical focus (00:33:23) Nvidia shows off Rubin Ultra with 600,000-Watt Kyber racks and infrastructure, coming in 2027 (00:35:23) China's SiCarrier emerges as challenger to ASML, other chip tool titans (00:38:24) Pony.ai wins first permit for fully driverless taxi operation in the center of China’s Silicon Valley Projects & Open Source (00:40:27) A new, challenging AGI test stumps most AI models (00:45:16) Challenging the Boundaries of Reasoning: An Olympiad-Level Math Benchmark for Large Language Models (00:48:13) Wan: Open and Advanced Large-Scale Video Generative Models (00:50:38) DeepSeek V3-0324 tops non-reasoning AI models in open-source first (00:54:46) OpenAI adopts rival Anthropic’s standard for connecting AI models to data Research & Advancements (00:55:56) Anthropic can now track the bizarre inner workings of a large language model (01:06:00) Chain-of-Tools: Utilizing Massive Unseen Tools in the CoT Reasoning of Frozen Language Models (01:11:50) Inside-Out: Hidden Factual Knowledge in LLMs (01:15:14) Sakana AI super-powers AI reasoning using Japan’s own Sudoku Puzzles Policy & Safety (01:18:38) Senator Wiener Introduces Legislation to Protect AI Whistleblowers & Boost Responsible AI Development (01:21:50) NVIDIA & Other Tech Giants Demand Trump Administration To Reconsider “AI Diffusion” Policy Which Is Set To Be Effective By May 15 (01:23:17) U.S. blacklists over 50 Chinese companies in bid to curb Beijing's AI, chip capabilities (01:26:44) Netflix’s Reed Hastings Gives $50 Million to Bowdoin for A.I. Program (01:27:55) Judge allows 'New York Times' copyright case against OpenAI to go forward (01:29:48) Judge rules that AI can continue training on copyrighted lyrics, for now

Our 204th episode with a summary and discussion of last week's big AI news! Recorded on 03/21/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at [email protected] and/or [email protected] Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: Baidu launched two new multimodal models, Ernie 4.5 and Ernie X1, boasting competitive pricing and capabilities compared to Western counterparts like GPT-4.5 and DeepSeek R1. OpenAI introduced new audio models, including impressive speech-to-text and text-to-speech systems, and added O1 Pro to their developer API at high costs, reflecting efforts for more profitability. Nvidia and Apple announced significant hardware advancements, including Nvidia's future GPU plans and Apple's new Mac Studio offering that can run DeepSeek R1. DeepSeek employees are facing travel restrictions, suggesting China is treating its AI development with increased secrecy and urgency, emphasizing a wartime footing in AI competition. Timestamps + Links: (00:00:00) Intro / Banter (00:01:36) News Preview Tools & Apps (00:02:50) Baidu launches two new versions of its AI model Ernie (00:10:46) OpenAI Unveils New Audio Models to Make AI Agents Sound More Human Than Ever (00:16:41) OpenAI’s o1-pro is the company’s most expensive AI model yet (00:20:53) Google brings a ‘canvas’ feature to Gemini, plus Audio Overview (00:22:18) Anthropic adds web search to its Claude chatbot (00:23:55) xAI launches an API for generating images Applications & Business (00:26:28) Nvidia announces Rubin GPUs in 2026, Rubin Ultra in 2027, Feynman also added to roadmap (00:36:25) M3 Ultra Runs DeepSeek R1 With 671 Billion Parameters Using 448GB Of Unified Memory, Delivering High Bandwidth Performance At Under 200W Power Consumption, With No Need For A Multi-GPU Setup (00:40:07) Intel reaches 'exciting milestone' for 18A 1.8nm-class wafers with first run at Arizona fab (00:42:45) Elon Musk’s AI company, xAI, acquires a generative AI video startup (00:44:44) Tencent Reportedly Makes Massive NVIDIA H20 Chip Purchase for WeChat’s DeepSeek Integration Projects & Open Source (00:46:32) Anthropic’s Not-So-Secret Weapon That’s Giving Agents a Boost (00:50:50) Mistral AI drops new open-source model that outperforms GPT-4o Mini with fraction of parameters (00:53:30) EXAONE Deep: Reasoning Enhanced Language Models Research & Advancements (00:55:58) Sample, Scrutinize and Scale: Effective Inference-Time Search by Scaling Verification (01:07:44) Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models (01:12:27) Communication-Efficient Language Model Training Scales Reliably and Robustly: Scaling Laws for DiLoCo (01:18:46) Transformers without Normalization (01:19:52) Measuring AI Ability to Complete Long Tasks (01:26:12) HCAST: Human-Calibrated Autonomy Software Tasks Policy & Safety (01:26:45) Announcing Zochi, an Intology Project (01:32:46) DeepSeek, a National Treasure in China, is Now Being Closely Guarded (01:37:02) Claude Sonnet 3.7 (often) knows when it’s in alignment evaluations Synthetic Media & Art (01:42:27) US appeals court rejects copyrights for AI-generated art lacking 'human' creator (01:45:10) Trump urged by Ben Stiller, Paul McCartney and hundreds of stars to protect AI copyright rules