783: Generative AI for Solar Power Installation — with Navdeep Martin

Classic vs. GenAI

Jon Krohn and Navdeep Martin explore the distinctions between classic AI and generative AI. Classic AI involves supervised and unsupervised learning approaches, such as tagging articles or clustering similar texts without labels. Generative AI, however, adapts to changing data and requires less data for training, enabling more dynamic applications like understanding community sentiment and evolving permitting codes 1 2.

Generative AI now allows you to deal with changing data, and you don't need a bunch of data to train a new machine learning model.

--- Navdeep Martin

This shift has expanded the use cases for AI, particularly in fields like climate change mitigation 2.

   

Recommendation Systems

The development of recommendation systems has evolved significantly over the years. Navdeep Martin shares his experience at the Washington Post, where they initially used topic models to track user interests, despite the challenges of keeping models up-to-date. This approach was later applied at Comcast for TV and movie recommendations 3.

Recommendation engines gained widespread popularity when Amazon came out with their recommendation engine in the late nineties.

--- Navdeep Martin

Generative AI has further revolutionized this field, enabling media organizations to write entire articles and adapt to changing data more efficiently 4.

   

Avoiding Hallucinations

Avoiding hallucinations in AI models is crucial for maintaining accuracy. Jon Krohn and Navdeep Martin discuss techniques like instructing models to leave answers blank if unsure and emphasizing expertise in prompts to reduce hallucinations 5 6.

The RLHF, at least in the last year or two that we see in the leading systems, there's some kind of priming there.

--- Jon Krohn

These strategies help ensure that AI systems provide reliable and accurate information, which is essential for their effective application in various fields 6.