The Science of Machine Learning vs. the Push for AI Deployment

The original premise was that artificial intelligence would revolutionize the world, transforming business processes, industrial mechanics, and creating entirely new industries. Elon Musk even suggested that money and work might become unnecessary, leaving only abundance. The promise held up in the era of Generative AI, with millions of near-perfect photos and video creation, ChatGPT answering almost all queries, and Claude handling complex workflows that previously took teams days.

Developers favored tools like Cursor, which simplified coding tasks, leading to rapid commercialization of frontier models and soaring valuations. However, the conversation began to shift towards reliability and determinism, as open-weight and open-source models emerged, challenging the notion that hundreds of billions of dollars were needed to perfect LLMs.

A new race began, focusing on the deployment of AI rather than its creation, led by Silicon Valley through influential gatekeepers like Y Combinator. Two kinds of companies emerged globally: those deploying AI across various industries and those taking a more scientific approach to make AI more deterministic. The challenge is that the technology was presented faster than it could be adopted, creating a mismatch between the technology's capabilities and businesses' ability to trust them.

The author, a founder, argues that while AI can change the world, it is not as simple as the hype suggests. AI relies on probabilities rather than fixed rules, making it difficult to replace determinism with probabilities. The author believes that AI is as significant as the Internet, empowering humans, fostering creativity, and helping create more powerful processes.

However, just as the Internet did not remove humans from the loop, AI will not either. The author sees a recent shift in messaging from Silicon Valley, with Dario Amodei calling for a slowdown in frontier AI progress, expressing concern about the technology's impact on the human race. Despite this, the author believes AI is still a significant opportunity and supports efforts to deploy AI in vertical industries, but warns that the deterministic expectations of these vertical AI companies run counter to the science of machine learning.