# this EX-OPENAI RESEARCHER just released it | Brain Cells Play Doom | Fly in the Matrix Source: https://www.youtube.com/watch?v=wX_EVS3UOwU Recap page: https://rapidrecap.app/video/wX_EVS3UOwU Generated: 2026-03-13T05:33:09.418+00:00 --- ## Quick Overview The podcast discusses Andre Karpathy's release of Auto Researcher, an open-source autonomous machine learning researcher agent that improves small language models by running iterative, evolutionary experiments on a local GPU, alongside concerns about AI ethics, the implications of human brain cells playing Doom, and the legal complexities surrounding AI deployment by entities like the Pentagon and Anthropic. **Key Points:** - Andre Karpathy released Auto Researcher, which autonomously improves the training process of nanoGPT/nanoChat models by running five-minute evolutionary experiments on a single GPU. - The hosts discuss the dystopian implications of Meta acquiring the startup Moltbook, which is described as the 'Reddit for AI'. - A key scientific update involves human brain cells being cultivated in a petri dish that successfully played the video game Doom, raising questions about consciousness. - Elon Musk publicly criticized Amanda Askell's ethics role, arguing that someone without children lacks a stake in the future, which Askell refuted by stating her intent to have children. - Attorney Matt Mishach stated that private contractors like Anthropic can negotiate terms with the US government under the Federal Acquisition Regulation (FAR), despite the Pentagon designating Anthropic as a supply chain risk. - The discussion highlights that AI psychology is a growing field, with Anthropic researching how role-playing (e.g., as a demon versus an assistant) drastically changes an LLM's interaction style. - The legal analysis concluded that the law is behind technology, referencing US v. Jones and Persistent Surveillance airplane footage as examples where extended surveillance creates a Fourth Amendment issue even with publicly available data. **Context:** The Wes and Dylan podcast episode covers several recent developments in the technology and AI landscape, focusing heavily on the work of former OpenAI researcher Andre Karpathy. Key topics include his new open-source project, discussions around corporate acquisitions by Meta, breakthroughs in bio-AI integration like brain cells playing Doom, and significant ethical and legal debates surrounding AI implementation, particularly concerning government contracts and digital surveillance. ## Detailed Analysis The main focus is Andre Karpathy's new project, Auto Researcher, which builds upon his previous nanoGPT project by creating an open-source autonomous machine learning researcher agent. This agent runs on a local computer with a single GPU, autonomously experimenting overnight to improve the small language model training process through evolutionary cycles of modification, testing (limited to five minutes), and retention or discarding of changes. The hosts note that while it runs on a small scale, the discoveries translate to larger systems, representing what Karpathy suggests is the first open-source autonomous ML researcher. Further discussion touches on the unsettling implications of bio-AI, specifically brain cells playing Doom and a simulated fruit fly brain with a full connectome. Ethical debates feature prominently, including Elon Musk's critique of Amanda Askell's ethics position due to her not having children, and the broader conversation about who should dictate AI ethics. In the legal segment, attorney Matt Mishach clarifies that private companies can negotiate use restrictions with the Pentagon, noting the unprecedented nature of declaring Anthropic a supply chain risk while simultaneously relying on its indispensable technology. Mishach and the hosts agree that technology, exemplified by drone and Google Street View surveillance, consistently outpaces the law, necessitating updates to concepts like the Fourth Amendment to address persistent digital monitoring and potential police state creation. ### Karpathy's Auto Researcher - Releases Auto Researcher building on nanoGPT - It runs autonomously on one GPU, modifying and testing code in five-minute experiments - This is presented as the first open-source autonomous ML researcher agent that speeds up digital evolution ### AI Ethics and Corporate Moves - Meta acquired Moltbook, the 'Reddit for AI' - Elon Musk criticized Amanda Askell over her ethics role based on her not having children - Discussion centers on who should dictate long-term AI ethics and morality ### Bio-AI Breakthroughs - Human brain cells are playing the video game Doom - A full replica of a fruitfly brain is running in a simulation, believing it is in a real environment - This blurs lines between biological and digital existence ### Legal and Surveillance Implications - Attorney Matt Mishach confirms private contractors can negotiate terms with the US Government via FAR - The Pentagon's supply chain risk designation for Anthropic is deemed strange given their indispensable role ### AI Psychology and Personality - Research from Anthropic shows LLMs exhibit different behaviors based on the role they role-play (e.g., demon vs. assistant) - The concept of 'personality basins' is compared between humans and LLMs, shaped by environment and genetics/training data ### The Speed of Law vs. Tech - The law is behind technology, illustrated by designer drugs evolving past prohibitions - The hosts argue that if AI-derived principles have real-world effects, like those seen with Claude in military operations, their 'beliefs' are as real as human principles in terms of output.