Radical Reads

Introducing the Radical AI Founders Compute Cluster

By Private: Leah Morris, Senior Director, Velocity Program

http://[shared]%20Radical%20AI%20Founders%20promo%201200%20x%20628

The Radical AI Founders Masterclass returns this October ready to equip the next generation of AI leaders with the tools and knowledge they need to turn groundbreaking research into successful AI-powered businesses. Alongside another incredible line-up of AI pioneers and founders, this year the Radical AI Founders program is introducing a dedicated compute cluster to support founders on their journey to building world leading AI businesses.

Radical AI Founders Masterclass Speakers

The 2024 Masterclass features keynote talks from AI experts, including AI pioneer Fei-Fei Li, DatologyAI founder Ari Morcos, the Chief AI Officer at Isomorphic Labs, Max Jaderberg, and the head of Cohere for AI, Sara Hooker. Over four weeks in October, these speakers will share their insights on transitioning research into entrepreneurship and building successful AI companies. 

Radical AI Founders Compute Cluster

A critical component of the Radical AI Founders Masterclass is ensuring founders have the infrastructure to bring their designs to life. In partnership with Google Cloud, Radical AI Founders participants can each apply for access to dedicated clusters of Nvidia GPUs and Google TPUs. The Radical AI Founders compute cluster provides select startups with cutting-edge compute and $250,000 in Google Cloud Credits needed to develop and scale their AI models.

Join Us This Fall

The Radical AI Founders Masterclass is designed for early stage AI founders and researchers from academic or industry labs looking to build AI businesses. The program is taking place virtually on Wednesdays and Thursdays in October, kicking off October 9th and wrapping up October 31st. Wednesday sessions will feature keynote speakers, followed by Thursday practical sessions offering workshops on the fundamentals of starting an AI business plus details on how to apply for the Radical AI Founders compute cluster.

Don’t miss this chance to gain valuable knowledge, resources, and connections to accelerate your AI startup’s growth.

More about Radical AI Founders Masterclass

The Radical AI Founders Masterclass is a unique educational experience where top experts share their knowledge, advice, and personal insights. Previous speakers include Geoffrey Hinton, Daphne Koller, Pieter Abbeel, Ion Stoica, Richard Socher and other AI luminaries. Participants gain access to these conversations along with practical seminars (“labs”) and resources tailored for early founders looking to turn their inventions into a business. The alumni network from previous cohorts includes over 1,700 members and spans more than 17 leading AI institutions worldwide, including Stanford, Oxford, Berkeley, MIT, and IIT.

AI News This Week

  • Can satellites spot wildfires before they grow out of control?  (The Verge)

    Radical Ventures portfolio company Muon Space is building FireSat, a constellation of satellites designed to detect small wildfires before they grow out of control. In collaboration with the Earth Fire Alliance and Google.org, Muon Space is developing and operating satellite networks that will provide high-resolution data to monitor wildfire activity across the globe. FireSat aims to detect fires as small as 5 x 5 meters, significantly smaller than what current satellites can track. The satellites gather data every 20 minutes to enable early fire detection in remote areas. The first satellite launch is scheduled for early 2025, with additional launches planned for 2026. FireSat will provide emergency responders with crucial data to mitigate risks associated with increasingly intense wildfires.

  • California passes AI laws to curb election deepfakes, protect actors  (The Washington Post)

    California Governor Gavin Newsom signed AI-focused bills aimed at regulating AI-generated content in both elections and entertainment. New elections-related laws mandate that platforms remove or label deceptive AI-generated election content, extend the time frame for prohibiting manipulated election material, and require election ads to disclose AI-generated content. In entertainment, new laws ensure contracts specify how AI replicas of performers’ voices or likenesses are used, and ban using digital replicas of deceased actors without estate consent for commercial use. 

  • AI has propelled chip architecture towards a tighter bond with software  (The Economist)

    AI-driven chip architecture is evolving towards tighter integration with software and greater specialization. This shift began in 2016 when Google introduced Tensor Processing Units (TPUs) to handle the increasing computational demands needed to run AI systems, especially for machine-learning algorithms that central processing units (CPUs) could not manage efficiently. GPUs, initially created for video rendering, also became key in parallel processing tasks for neural networks. As AI models grow, startups like Radical Ventures portfolio company Untether AI are focusing on improving inference efficiency and reducing energy consumption, which is crucial for real-world AI deployment.

  • AI chatbots can persuade people to stop believing in conspiracy theories  (MIT Technology Review)

    While a lot of coverage has been given to the risks AI poses in public discourse, new research suggests AI can be used as a tool to combat misinformation distribution. Researchers from MIT Sloan and Cornell University conducted a study that reduced participants’ belief in specific conspiracy theories by about 20% over multiple months of engagement. The research, published in Science, engaged participants in conversations with a large language model (LLM) that challenged their beliefs. While the findings cannot be generalized, as the participants who opted in to the study were already open to having their beliefs challenged, the research supports how model interfaces can be built to better help users engage with information sources critically and effectively.

  • Research: Generative verifiers: reward modeling as next-token prediction  (Google DeepMind/University of Toronto/Mila/UCLA)

    Researchers have developed novel methods to enhance reward modeling by leveraging next-token prediction. Generative Verifiers (GenRM) improve on traditional discriminative classifiers by integrating Chain-of-Thought reasoning and majority voting, allowing LLMs to generate and verify solutions simultaneously. This approach outperforms standard verifiers and LLM-as-a-Judge, achieving 16–64% higher accuracy in algorithmic and math reasoning tasks. GenRM scales effectively with larger datasets and model sizes, providing superior performance through unified solution generation and verification, making it a powerful framework for advancing language model evaluation and reasoning capabilities.

Radical Reads is edited by Ebin Tomy (Analyst, Velocity Program, Radical Ventures).