# Jan 1, 2026: Nvidia Buys Groq ($20B), Meta Acquires Manus, Qwen-Image-2512 & Will Brown Interview Source: https://www.youtube.com/watch?v=Aa2eElhsXEw Recap page: https://rapidrecap.app/video/Aa2eElhsXEw Generated: 2026-01-01T21:33:37.299+00:00 --- ## Quick Overview Alex Volkov of Weights & Biases provides a quick recap for the first ThursdAI episode of the year (Episode 53), noting that despite being on vacation, he covered major AI news including Meta acquiring Manus AI and Nvidia allegedly buying Groq for $20B, and featured an interview with Will Brown from Prime Intellect discussing RL environments and the importance of verifiable signals in training models. **Key Points:** - Alex Volkov hosts the first ThursdAI episode of the year (Episode 53) while technically on vacation, prioritizing consistency. - Major AI news covered includes Meta acquiring Manus AI (an agentic runner company previously covered in March 2025) and Nvidia reportedly buying Groq (a chip processor for fast inference) for a rumored $20 billion. - The episode features an interview with Will Brown, a researcher at Prime Intellect and notable figure on X (formerly Twitter), known for his takes and memes. - Will Brown explains that RL environments, when applied to tasks like math problems, provide a verifiable signal (like a score of 1 for right, 0 for wrong) that guides the model's weights, contrasting this with the less structured nature of creative tasks. - Brown notes that the underlying principle of RL in LLMs is about exploring possible ways the model could approach a problem (like explicit chain-of-thought reasoning) to find the optimal path. - The discussion touches on the importance of having a structured, verifiable way to score model outputs, contrasting the explicit scoring in math with the fuzziness of creative writing evaluation. - The episode concludes with Alex Volkov expressing gratitude to his co-hosts (Nit, Yam Peleg, LDJ Wolfram, Raven Wolf) and supporters (Weights & Biases, Cor Reef) for continuing to deliver AI news. ![Screenshot at 00:05: Alex Volkov, host of ThursdAI, introduces the first episode of the year while speaking into a microphone, setting the stage for the news recap and interview segment.](https://ss.rapidrecap.app/screens/Aa2eElhsXEw/00-00-05.jpg) **Context:** This video is an episode recap of the 'ThursdAI' podcast by Alex Volkov from Weights & Biases, marking the first episode of the new year (Episode 53). Volkov opens the episode solo, explaining he is recording while on vacation but still wants to maintain consistency. The main segment transitions to an interview with Will Brown, a researcher at Prime Intellect known for his commentary on AI developments, particularly regarding Reinforcement Learning (RL) and model training methodologies. ## Detailed Analysis Alex Volkov opens the first ThursdAI episode of the year (Episode 53) from his room, despite being on vacation, to deliver the week's news and an interview. The major news item is the reported acquisition of Groq, the chip processor company focused on fast inference, by Nvidia for an alleged $20 billion, a deal supposedly led by Jensen Huang himself. Additionally, Meta reportedly acquired Manus AI, an agentic runner company previously featured on the show. Volkov then introduces the main interview with Will Brown of Prime Intellect, who is known for his insightful 'hot takes' and memes on X. Will Brown discusses Reinforcement Learning (RL) environments, explaining that they provide a structured signal (like a verifiable score of 1 for correct or 0 for incorrect math answers) to guide model training. He notes that this structured feedback, unlike vague creative writing evaluations, allows for clear optimization. Brown points out that the key benefit of RL in LLMs is exploring various reasoning paths (like chain-of-thought) to find the most efficient route to the correct answer, contrasting this with older methods that required excessive human labeling or time-consuming manual checking. He mentions that the adoption of these structured methods by major labs like Google has been rapid. The conversation concludes with Volkov thanking his co-hosts and supporters for enabling the continued production of the podcast. ### Intro & Year Recap - Welcome to the first ThursdAI of the year (Episode 53) - Alex Volkov is recording from his room while on vacation - He plans to cover news and an interview with Will Brown. ### AI Industry News - Nvidia reportedly bought Groq (chip processor for fast inference) for $20 billion, led by Jensen Huang - Meta acquired Manus AI (agentic runner company). ### Interview Setup & Guest Intro - Interview with Will Brown, researcher at Prime Intellect and known for X takes/memes - Acknowledgment of the Code Summit event where the interview took place. ### RL Environments Explained - RL environments are like a setup of math problems plus a grader (e.g., a verifier function) that outputs a score (1 or 0) based on correctness, similar to train/test sets. - The key idea of RL for LLM reasoning is using this verifiable signal to incentivize the model to think longer and explore better solutions (like chain-of-thought) rather than just producing the first answer. ### Hot Takes & Twitter Fame - Discussion on Will Brown's popular Twitter content, noting his style involves poking fun at things lightly. - Reference to the early days of LLMs like Sydney where models were overly verbose or struggled with simple tasks. ### Future of RL and Evals - The structured nature of RL environments (tasks, harness, grader) makes them more verifiable than general benchmarks or creative tasks. - The goal is to incentivize the model to find the sweet spot of reasoning length that yields correct answers efficiently. ### Closing Remarks - Alex Volkov thanks Will Brown, his co-hosts (Nit, Yam Peleg, LDJ Wolfram, Raven Wolf), and sponsors (Weights & Biases, Cor Reef) for their support. ![Screenshot at 00:05: Alex Volkov, host of ThursdAI, introduces the first episode of the year while speaking into a microphone, setting the stage for the news recap and interview segment.](https://ss.rapidrecap.app/screens/Aa2eElhsXEw/00-00-05.jpg) ![Screenshot at 00:48: Alex Volkov summarizing recent AI news, including acquisitions and model updates, as part of the episode's opening segment.](https://ss.rapidrecap.app/screens/Aa2eElhsXEw/00-00-48.jpg) ![Screenshot at 05:05: Transition screen featuring Alex Volkov and a title card introducing the interview with Will Brown of Prime Intellect.](https://ss.rapidrecap.app/screens/Aa2eElhsXEw/00-05-05.jpg) ![Screenshot at 07:14: Alex Volkov passionately discussing the recent developments and the need for structured feedback, gesturing emphatically.](https://ss.rapidrecap.app/screens/Aa2eElhsXEw/00-07-14.jpg) ![Screenshot at 11:21: A graphic slide appears labeling the next topic: "SCALING LAWS" during the discussion with Will Brown.](https://ss.rapidrecap.app/screens/Aa2eElhsXEw/00-11-21.jpg)