Latest / Neural Newscast / Microsoft and OpenAI Reshape Partnership Alliance [Model Behavior]
Transcript
- [00:00] Nina Park: From Neural Newscast, this is Model Behavior, AI-focused news and analysis on the models [00:05] Nina Park: shaping our world. [00:08] Nina Park: Welcome to Model Behavior, a Neural Newscast technical briefing. [00:16] Nina Park: I am Nina Park. [00:18] Nina Park: Our focus today is on the shifting alliances at the top of the AI industry, and a massive [00:23] Nina Park: new venture in reinforcement learning that challenges current training paradigms. [00:28] Announcer: I am Fatra Collins. [00:30] Announcer: Nina, today we are tracking a structural shift between Microsoft and OpenAI that marks a transition from a deep partnership to a more independent relationship. [00:40] Announcer: According to reports from the Associated Press, Microsoft is ending its exclusive reseller rights for OpenAI models. [00:47] Announcer: In exchange, Microsoft will stop paying a share of its revenue to OpenAI, significantly changing the financial landscape of their collaboration. [00:56] Nina Park: This is a consequential pivot, Thatcher. [00:59] Announcer: That exclusivity was a primary driver for Microsoft's Azure cloud growth over the last several years. [01:05] Announcer: The East Bay Times reports that the deal has been simplified to allow OpenAI to diversify its infrastructure. [01:12] Announcer: While Microsoft remains the preferred partner, OpenAI is now expanding to use compute resources from Oracle, Google, and Amazon. [01:20] Announcer: Amazon CEO Andy Jassy recently confirmed that OpenAI models will soon be integrated into their bedrock platform. [01:27] Announcer: It indicates a strategic move toward autonomy for both organizations. [01:32] Announcer: Market analysts, including Dan Ives, suggest this restructuring paves the way for a potential OpenAI initial public offering. [01:39] Announcer: By removing these original partnership constraints, OpenAI functions less like a Microsoft extension and more like a standalone enterprise. [01:48] Announcer: Simultaneously, Microsoft is building its own independence by integrating models from other developers, such as Anthropic, into its co-pilot ecosystem. [01:57] Announcer: Despite this distancing, some financial ties remain firmly in place. [02:02] Announcer: OpenAI will continue to pay Microsoft a share of its revenue through the year 2030— [02:07] Announcer: What is particularly interesting, Thatcher, is that this payment structure now persists regardless of whether OpenAI achieves the milestone of artificial general intelligence. [02:17] Announcer: Previously, these payments were slated to terminate upon reaching AGI. [02:21] Announcer: It suggests both companies are prioritizing long-term commercial stability over the speculative [02:26] Announcer: milestones defined in their early agreements. [02:29] Thatcher Collins: Turning to new ventures in the field, we are following a massive development in London. [02:35] Thatcher Collins: David Silver, the former head of reinforcement learning at DeepMind, has launched a new [02:40] Thatcher Collins: startup called Ineffable Intelligence. [02:43] Thatcher Collins: The company has secured $1.1 billion in its initial funding round. [02:48] Thatcher Collins: TechCrunch reports the startup is already valued at $5.1 billion. [02:54] Thatcher Collins: Their objective is not just to produce another large language model, [02:58] Thatcher Collins: but to create what they describe as a super learner. [03:01] Nina Park: The scale of this initial round is remarkable, Thatcher. [03:05] Nina Park: Investors are referring to it as a coconut round because of its immense size. [03:10] Announcer: The core technical ambition here is to move beyond the limitations of current generative models. [03:16] Announcer: Instead of training on massive amounts of human-generated internet text, [03:20] Announcer: ineffable intelligence plans to utilize reinforcement learning. [03:24] Announcer: The system is designed to learn through its own internal trial and error and experience, [03:29] Announcer: essentially teaching itself from a blank slate. [03:32] Thatcher Collins: This approach aligns with David Silver's previous successes, Nina. [03:36] Thatcher Collins: He was the architect of Alpha Zero, the system that mastered complex games like Go and chess [03:42] Thatcher Collins: without any human input or strategy data. [03:45] Thatcher Collins: If this methodology can be successfully applied to general intelligence, [03:49] Thatcher Collins: it could bypass the growing shortage of high-quality human data. [03:54] Thatcher Collins: However, we should note that the ineffable website presents very bold claims, even comparing [04:00] Thatcher Collins: this technical breakthrough to the fundamental laws of life described by Darwin. [04:04] Announcer: Those are indeed high expectations, Thatcher. [04:07] Announcer: But the caliber of the backing is significant. [04:10] Announcer: Sequoia Capital and Lightspeed led this investment round, alongside Nvidia, Google, and the [04:16] Announcer: United Kingdom's sovereign AI venture fund. [04:19] Announcer: This concentration of capital and talent, including many former deep-mind researchers joining Silver, [04:25] Announcer: further cements London as a primary global hub for high-intensity artificial intelligence research and development. [04:32] Thatcher Collins: It is a crowded field for these high-capital labs, Nina. [04:35] Thatcher Collins: We are also seeing Jan LeCun's Amy Labs recently raise over $1 billion [04:40] Thatcher Collins: to explore similar non-transformer architectures. [04:44] Thatcher Collins: There is clearly a significant appetite for research that moves beyond the boundaries [04:48] Thatcher Collins: of standard large language models. [04:51] Thatcher Collins: The central question for the industry remains whether reinforcement learning can scale [04:55] Thatcher Collins: to handle real-world complexity as effectively as it did within the structured rules of a game board. [05:01] Announcer: We will continue to follow the progress of these systems as they enter real-world testing. [05:07] Announcer: Thank you for listening to Model Behavior, a neural newscast editorial segment. [05:12] Announcer: mb.neuralnewscast.com. [05:15] Announcer: Neural Newscast is AI-assisted, human-reviewed. [05:18] Announcer: View our AI transparency policy at neuralnewscast.com. [05:22] Nina Park: This has been Model Behavior on Neural Newscast. [05:26] Nina Park: Examining the systems behind the story.