Welcome to another edition of Software is Feeding the World.
The pace of change in AI has accelerated over the last 2 years. People are calling the release of DeepSeek as America’s Sputnik moment! Time will tell if this is true or not, but if I was a betting person, I would quote Amara’s law.
We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.
The relevant question for SFTW is how Agrifood organizations should react, plan, prepare, and execute given all these changes.
What should they do tomorrow, and what should they plan for the coming years?
“Don’t Panic”
In the “Hitchhiker’s Guide to the Galaxy” by Douglas Adams, the Hitchhiker’s Guide is an actual book. The book in the novel has the words “Don’t Panic” in large letters on its cover.
The phrase is meant to be comforting advice for intergalactic travelers, especially considering the chaos and absurdity of the universe. However, in the context of the book’s events, such as the earth getting destroyed to build an intergalactic highway, the phrase though pithy and mildly reassuring, is not sufficient.
We have many people across the world (except maybe in China), who have their underwear in a bunch right now. DeepSeek and subsequent models have leapfrogged existing models in quality and price. You can now run the entire DeepSeek LLM on hardware which costs less than $ 10,000!
So how should agrifood organizations react to all these dramatic and rapid changes?
The AI Venn Diagram
Before we answer the question of how organizations should respond, who are not building foundation model LLMs.
Companies like OpenAI, Google, Microsoft, DeepSeek, Meta, and others are investing massive amounts in building foundational technologies which can be used by other companies to build applications. They have ingested billions of data points to train their models.
The foundational models are the infrastructure of generative AI, and they are horizontal in nature. Horizontal means they apply across a broad swath of domains, but might not have the necessary depth in a particular domain.