The video discusses the recent developments in AI, highlighting the competition between academic institutions and tech giants like OpenAI, particularly with Stanford and the University of Washington creating advanced AI models at a fraction of the cost. Sam Altman describes OpenAI’s new GPT-4.5 as a “giant and expensive” model that does not significantly advance capabilities, raising concerns about its sustainability and the company’s future in a rapidly evolving landscape.
The video discusses the recent developments in the AI landscape, particularly focusing on the annual Institute for Economic Policy Research Summit at Stanford University. High-profile figures from banking and tech, including JP Morgan’s Jamie Dimon and Alphabet’s Ruth Porat, gathered to discuss various topics, including competition policy and the role of AI in national security. The timing of the summit is significant as it coincides with a new phase in the AI race, where universities like Stanford are finding ways to compete with major tech companies despite facing federal funding cuts.
Stanford and the University of Washington recently developed an AI reasoning model called S-1, which was created for less than $50 in compute credits. This model is reported to be on par with advanced models from OpenAI and DeepMind, showcasing how academic institutions can innovate at a fraction of the cost of their corporate counterparts. Similarly, researchers at Berkeley achieved comparable results for around $450, highlighting a trend where the costs of developing cutting-edge AI models are decreasing.
The video highlights a technique called distillation, which allows for the compression of large AI models into smaller, more efficient versions without sacrificing performance. This advancement is changing the narrative in the AI field, as it enables universities to produce competitive models while tech giants like OpenAI are still focused on developing larger and more expensive models. This shift raises questions about the sustainability of OpenAI’s approach, especially with the introduction of their latest model, GPT-4.5.
Sam Altman, CEO of OpenAI, described GPT-4.5 as a “giant and expensive” model that does not significantly advance reasoning capabilities and is not expected to outperform benchmarks. The costs associated with GPT-4.5 are notably high, being 15 times more expensive per million output tokens than GPT-4 and significantly more than other advanced models like Anthropic’s Claude and Google’s Gemini. This raises concerns about who will be willing to pay for such an expensive model that does not push the boundaries of AI development.
Despite the high costs and questions surrounding the viability of OpenAI’s strategy, the company continues to push forward with its projects, including a joint venture with SoftBank and Oracle aimed at investing heavily in GPUs and data centers. However, the market’s reaction has been cautious, as evidenced by Nvidia’s stock decline following the announcement of GPT-4.5. The video concludes by emphasizing the growing competition from academic institutions and the potential challenges that OpenAI may face in maintaining its leadership in the AI space amidst rising costs and innovative alternatives.