Ask HN:我做出了一种新型模型(不是大语言模型),但不知道怎么卖出去
作者开发了一种非大语言模型的新型推理模型,每月运行成本不到10美元,但需要像宠物一样逐步训练和引导。最初曾将其作为游戏内机器人任务分配系统推出,但效果不佳。作者困惑于如何让用户愿意尝试一个初期尚不成熟的系统,并寻求推广与商业化的建议。
作者开发了一种非大语言模型的新型推理模型,每月运行成本不到10美元,但需要像宠物一样逐步训练和引导。最初曾将其作为游戏内机器人任务分配系统推出,但效果不佳。作者困惑于如何让用户愿意尝试一个初期尚不成熟的系统,并寻求推广与商业化的建议。
Leila Clark defends Jared Friedman's decision to give an AI agent full access to YC's production database, arguing that professional engineers with proper guardrails—backups, documented prohibitions, ORMs, and post-mortems—can safely unlock significantly greater productivity from AI agents, which are often smarter than average new developers.
In an interview, Yuval Noah Harari discusses how AI may serve as a force for truth, among other interesting topics.
The author argues that while no single math field has high ROI for the average programmer, each developer works in a domain where some branch of math could help. They suggest broad exposure to many fields (e.g., via textbooks) is more practical than deep study of one topic, since you need to know a field exists before deciding to learn it in depth.
Garry Tan says Jo uses GBrain under the hood, calling it one of the fastest ways to set up a personal or company AI. Kevin Li describes building an AI assistant that works proactively, completing tasks before the user wakes up, and shares a real-life example of his own family using it.