The video discusses the anticipated return of Fable 5 with tighter enterprise controls, highlights the innovative Japanese Sakana AI’s Fugu Ultra model outperforming competitors in complex AI tasks, and announces OpenAI’s new Jalapeno AI chip designed for efficient large language model inference. It also critiques European AI regulations for hindering innovation, reflecting on the challenges and potential shifts in the AI landscape amid geopolitical and technological developments.
The video begins with exciting AI news, highlighting the potential return of Fable 5, a popular AI model, though no official confirmation has been made. Multiple signs suggest its comeback, such as updates in Cloud Code referencing Fable and its reappearance in Amazon Bedrock. The video discusses past negotiation issues involving Anthropic’s CEO Dario Amade, who has been replaced by co-founder Tom Brown in talks with the Trump administration, facilitating smoother discussions about re-releasing Claude Fable 5. The re-release is expected to have tighter access controls, focusing on enterprise use with limited accounts and capped usage.
Next, the video delves into Sakana AI, a Japanese AI lab known for its innovative research in recursive self-improving models. Sakana AI recently introduced the Fugu Ultra model, which reportedly matches the performance of Fable and Mythos without export control risks. Unlike traditional models, Fugu Ultra acts as an orchestrator, coordinating multiple models to handle different tasks, a concept that could represent the future of AI development. Benchmark tests show Fugu Ultra outperforming several leading models, including Opus 4.8 and GPT 5.5, especially in complex tasks like AI research automation using Andrej Karpathy’s open-source Auto Research project.
The video further explains the significance of the Auto Research benchmark, which tests a model’s ability to conduct iterative AI research by improving training recipes autonomously. Fugu Ultra excels in this benchmark, demonstrating a learning curve that surpasses other models after several iterations. Despite some initial underwhelming impressions from independent testers, the benchmark results suggest that Fugu Ultra’s orchestrating approach offers unique advantages. Additional experiments show Fugu Ultra performing well in tasks like classical Japanese kana reading, Rubik’s cube solving, CAD design, and chess, often outperforming other models in accuracy and strategy.
In other news, OpenAI has announced its first AI chip, named Jalapeno, designed specifically for large language model inference rather than training. Remarkably, the chip went from design to production in just nine months, partly accelerated by OpenAI’s own AI models. This custom silicon aims to address inference bottlenecks and enable large-scale deployment of AI services by 2026, partnering with major data center providers like Broadcom, Celestica, and Microsoft. The video also shares a personal anecdote about planting a jalapeno bush, humorously tying it to the chip’s name.
Finally, the video touches on European AI regulation and innovation challenges. It highlights a troll account parodying European regulatory enthusiasm and discusses criticisms from EU leaders about the abrupt cutoff of Fable 5/MOS 5 access, which they see as a catalyst for Europe to develop its own AI ecosystem. The video critiques EU regulations for stifling innovation and causing productivity losses, such as the massive time spent on cookie consent banners. It questions whether the EU’s recent realization of AI’s importance will lead to meaningful change or if it’s too late, inviting viewers to share their thoughts on these developments.