Show HN: Directionally bad – a newsletter about risks of AI centralization (deaination.substack.com)

In the latest issue of the newsletter "Directionally Bad," DeAI Nation focuses on the implications of Andrej Karpathy's recent move to Anthropic, a centralized AI lab known for its restrictive stance on open-source AI. Karpathy, a prominent figure in the AI community and co-founder of OpenAI, has previously championed decentralized AI initiatives. His decision to join Anthropic raises questions about the future trajectory of open-source contributions in an environment perceived as hostile. Some industry leaders express hope that Karpathy may advocate for more open practices at Anthropic, while others anticipate a continued trend toward centralization. Additionally, the newsletter highlights significant economic shifts in AI development, including a substantial price increase for renting Nvidia's H100 GPUs, which could complicate access to computational resources for developers. On a more optimistic note, Pluralis introduces Agora, a technological breakthrough that enables model pretraining using consumer-grade GPUs, potentially revolutionizing the cost dynamics of AI training. This could democratize access to advanced AI training capabilities significantly, as illustrated in a pilot project involving an 8 billion parameter model.