Darren LI

10. A lot of founders are too addicted to operation numbers. The number of followers on Discord, the number of Twitter followers. These are the worst vanity tricks, and I ask founders to ignore these numbers because often quality is more important than quantity.

4. This may be the best time to launch a product. There is too much noise in a bull market. As these startups face reality, they begin to understand that without the traction of the product, the token price will eventually drop and be blamed by the holders.

目前做大模型+机器人的 最大难点,在于获取数据 。谷歌和微软都选择了从收集家用动作入手,未来其他场景,比如办公室、街道、商场、工厂,如果能训练出来一个个小模型,融入到底层大模型里面,将会是创业公司的重要壁垒。

自动驾驶汽车公司尝试过这种方法,但尚未完全成功。许多自动驾驶汽车公司,比如特斯拉,尝试通过在大量数据上训练这样一个端到端的大模型来实现5级自动驾驶,他们拥有训练模型所需的资源,资金和数据。特别是特斯拉在实际部署的10几万辆汽车中,不断收集并注释数据。尽管如此,特斯拉迄今尚未能生产出5级自动驾驶系统。这并不是说他们的方法根本不起作用,它能够处理大量驾驶情况,然而,它远未达到99.X%的可靠性。

4. Today, the stack for LLM APIs can feel separate from the custom model training stack, but these are blending together over time.

That’s a key mental shift. The cost of failure is as close to zero as it’s ever been, and it will continue to fall. That’s true for entrepreneurship, and it’s even more true for the Great Online Game. Because entrepreneurs are trying to build a business; when you start to play the Great Online Game, you’re just building optionality.