AI - Perspectives from a software engineer
Today I am sharing a Q and A with Benjamin Karst, founder and chief engineer at Essence AI. I asked Benjamin some of his thoughts about the development and use of AI by individuals and companies.
How is AI Evolving?
Benjamin Karst: Last year large language models (LLMs) were able to generate text in-differentiable from human writing. This year, leading AI companies are focused on expanding human-level performance into other modalities like generative speech and image generation.
More than mastering these modalities, companies are training AI to master specific professional skills. For instance, there are specialized large language models specifically trained for their coding ability. There are also a great number of startups focused on assisting lawyers and doctors. These applications are pragmatic because we know that intelligence in these areas is already of value.
On one end of the spectrum is pragmatic AI, something like an LLM that can help you code, and on the other end you have novelty AI, voice transfer singing AI, AI that is fun but doesn't have proven value. Many applications seem to be somewhere in between, like text-to-video generation. I would say that might be on the novelty end of the spectrum.
You can see AI scaling vertically, with increased abilities, and horizontally, as its usage becomes more widespread.
How are companies using AI?
Benjamin Karst: Put simply, companies are using AI to replace labor. You could also make the argument that it's making people more efficient but if a programmer is 10x more efficient and the demand for software stays constant, you need 90% fewer programmers.
How are consumers using AI?
Benjamin Karst: According to Sam Altman, ChatGPT is used by over 100 million people per week. Since ChatGPT’s, a number of analogous products have been released including Google Gemini, Bing Copilot, Anthropic’s Claude, and most recently Meta AI. Clearly, the “killer app” is the LLM chatbot.
Text-to-image generation is also popular. That is when you describe what image you want to see and AI generates it. Leading in this category is Midjourney. They claim to have over 40 million monthly visitors. Since Mid Journey’s success, Adobe has integrated a number of generative AI features into its products.
An up-and-coming (application) is text to music. It’s the same kind of idea where you describe what you want to hear and AI outputs what you just described. Udio and Sono are getting a lot of hype in the area of Gen AI but it remains to be seen whether or not demand will be sustainable.
What will the next few years bring in the AI field?
Benjamin Karst: I don’t know what’s next but all signs point to the American middle-class labor market being disrupted. European law has imposed strict regulations on what they deem “high-risk” use cases (education, employment, healthcare). I view these regulations as protecting jobs just as much as they are safety measures.
The dot-com era was an infrastructural-level change in commerce. It became over-invested and overhyped. The same is happening with AI. 98% of AI startups won’t be around in a couple of years but investors will 1000x or more from the few that survive so it’s still worth the investment.
As you might have guessed from his surname, Benjamin is my son.
If you have specific questions for Benjamin, find him on his LinkedIn profile.