Join 📚 Edwin's Highlights
A batch of the best highlights from what Edwin's read, .
One reason for the focus on understanding over prediction is the lack of available data. Until recently, sociological datasets were small. Machine learning works best when applied to large datasets, so sociologists have historically relied on simple statistical models like linear regression.
AI Snake Oil
Arvind Narayanan and Sayash Kapoor
The way to catch on to what your inner roommate is really like is to personify it externally. Make believe that your roommate, the psyche, has a body of its own. You do this by taking the entire personality that you hear talking to you inside and imagine it as a person talking to you on the outside. Just imagine that another person is now saying everything that your inner voice would say. Now spend a day with that person.
The Untethered Soul
Michael A. Singer
Perhaps the clearest takeaway for model providers, so far, is that commercialization is likely tied to hosting. Demand for proprietary APIs (e.g. from OpenAI) is growing rapidly. Hosting services for open-source models (e.g. Hugging Face and Replicate) are emerging as useful hubs to easily share and integrate models — and even have some indirect network effects between model producers and consumers. There’s also a strong hypothesis that it’s possible to monetize through fine-tuning and hosting agreements with enterprise customers.
Who Owns the Generative AI Platform? | Andreessen Horowitz
a16z.com
...catch up on these, and many more highlights
Readwise