Ben’s Perspective Part 2: A Few Thoughts About Chatgpt With Ben Horowitz

Ben answers a few questions about a revolutionary new technology, and gives some perspective on crypto at the end!

Ben Horowitz is the cofounder and CEO of Andreessen Horowitz, one of Silicon Valley’s leading venture capital firms, and the author of two best selling books, The Hard Thing About Hard Things, and What You Do Is Who You Are. To read the first part of the Ben’s Perspective series, click the link here.

To find more of Ben, follow him on his Twitter.

My first question is about ChatGPT. What are some practical applications of the platform? What domains are most likely to be benefited by such an advanced large language model? 

Wow, there are so many practical applications that I feel like picking one or two won’t do it justice. One of the simplest and most obvious will be to handle routine customer support calls. I think that ChatGPT can probably already do this better than humans and by GPT 4, will definitely surpass humans. As performance improves and the cost goes down, one could easily imagine it replacing conventional search. Why ask for the website when you can ask for the answer directly? More importantly, there are incredible applications in the medical domain as AI helps us diagnose and treat many previously deceptive and deadly diseases. I believe there will be practical applications in almost every problem domain we face.

I had a question last interview about AI potentially ruining certain areas of academia by turning domain expertise into a function of algorithms instead of a product of schooling. What I didn’t predict is that AI would actually make academic expertise itself obsolete and this is something I don’t think enough people are thinking about. YouTube, Google and even Twitter are huge learning resources that make the whole notion of college seem questionable, and now we have an even better teacher on the horizon with ChatGPT. I can ask this machine anything and learn anything, there’s a level of dialectic here far beyond what was possible before online. Is academia in trouble? 

We’ve recently had this big debate over whether or not college loans should be forgiven. The conversation has all been around “is it fair to burden kids with so much debt or is it fair to effectively punish the people who repaid their loans?” I think both of those are besides the point. The real issue is that the value proposition of going to college is horribly broken. If it was worth the price, we wouldn’t have a student loan problem because students would easily be able to pay back their loans. So, to answer your question, Academia is already in big trouble as it is choking the financial life out of its customers. I think that ChatGPT for general education plus specialized education, mapped to real economic opportunity is a possible solution. In any case, a system where a 4-year college costs $300,000 and doesn’t improve your earning power needs to die. In the future, what we now call “Academia” should be targeted at true scholars only and the rest of us should have a better, more practical, cheaper option.

And do you still feel that knowledge-heavy labor markets are unlikely to be affected by these wild advancements? 

I don’t think I ever  said that any labor markets won’t be impacted. They definitely will and it looks like the knowledge-heavy ones will be impacted first. ChatGPT will replace so-called knowledge workers long before robotics replace people who clean, work on oil rigs, etc. 

Having said that, I would not underestimate the need for new things and new kinds of work. It’s easier to predict the kinds of work that will be eliminated, but much more difficult to identify the kinds of work that will emerge. Everyone easily predicted the elimination of typesetting jobs with the advent of the computer, but nobody predicted the emergence of millions of graphic design jobs. 

The other side of this is that students don’t have to learn anything since there’s no question a teacher can ask that sufficiently advanced language models can’t answer. Huge swaths of kids in high school and university can just switch their minds off and let a machine do their schoolwork for them, they don’t have to learn anything. How should educators respond? 

When I was a kid, cheap electronic calculators became available for the first time. There was a huge panic in academia that nobody would ever learn arithmetic again. I remember in junior high school, there was a play about all the calculators breaking and nobody being able to figure anything out. I think things will change and we will have to change the way we teach (which is long overdue by the way), but humans for the most part are curious and want to learn things. 

With human creativity being rapidly eclipsed by machines I sense a nascent yearning in the ether for new ways to express the human spirit. I feel like there will either be a Butlerian Jihad, reinvigorated interest flowing into sports as the last area where humans can excel, AI becoming supplemental in creating new forms of literature and digital art that we can’t yet imagine, or something else. Where and how will humans show the eminence of our species when AI becomes better than us at everything? I feel sort of desperate for some human-centric optimism here because the whole project of being human feels really weird as our specialness becomes automated. 

A couple of observations. First, I do think the last mile is the most difficult when it comes to AI creativity. The Substack team often talks about the need for “non-fungible writers” as opposed to the 95% of things that get written and can now literally be created by a machine. If what we’ve learned from self-driving cars is any indication, the last 1% is more difficult than the first 99%. Having said that, AI may still get there quickly :-).

Second is that we’ve seen all the historical forms of musical creativity get obsoleted by machines, which has mostly resulted in more human creativity. Every instrument can be played extremely well by a computer without decades of study. Somehow this hasn’t obviated the need for drummers or guitar players or violinists, but rather created a whole new class of creatives. When rappers started sampling existing music, many claimed that would be the end of musicians, but that definitely has not been the case. I don’t know what the post-post-modern creatives will do, but I am fairly confident that they will do something and people will love it :-).

One last question about the SBF situation and crypto in general. Crypto has seen more harmful scandals in its popular history than practical developments like we’ve seen with AI. Are we likely to ever see anything comparable to ChatGPTcome out of the crypto space? Does Venture Capital still believe in crypto despite the recent SBF fiasco and general crypto market collapse?

First off, I think that comparing Crypto to AI in this way is really weird. It would be like comparing the Internet to the invention of high-level programming languages. Crypto and AI are really different and will lead to different kinds of issues and breakthroughs. Crypto provides technology that enables a much better social and economic system through things like native Internet money, native Internet property rights, real privacy, strong authentication (as opposed to passwords and large corporations like Meta and Google claiming to keep your things private), ownership of your things, etc. In order to enable the power of these inventions, a great deal of infrastructure and consumer adoption must take place. At the same time, because money is involved, there is also a casino aspect to Crypto, which needs regulation and can cause SBF sized scandals. I cannot speak for other VCs, but we 100% believe not only in Crypto, but in its importance in the future of humanity. 

AI, on the other hand, is a fundamental breakthrough in solving a class of problems that we could not solve with conventional programming (due to their combinatorially explosive nature). This is every bit as exciting as the invention of the computer itself. In general, new technologies carry new risks and opportunities for criminals (e.g., fraud through AI deep fakes, security hacking, etc.). However, overall, AI will enable us to solve all kinds of important problems ranging from drug design to nuclear fusion and, like the internet or Crypto, will be worth the downsides (I hope :-)).

To find more from Ben, follow him on Twitter and check out some of his writing at a16z!