There’s an iconic scene in season one of Mike Judge‘s Silicon Valley TV show, where Christopher Evan Welch, playing the character Peter Gregory, beseeches young career professionals not to go to college.
A professor in the audience chimes in, irate at Gregory‘s attacks on his livelihood.
“The value of a college education is intangible,” he sputters.
Welch, who sadly did not survive the end of the show, (Gregory‘s character had to be written out), eyes the bearded gent and replies sardonically:
“The value of snake oil is also intangible.”
Follow me here, because this is interesting.
I’ve been writing about the tension between academic goals and the goals of innovation – between the urge to pursue higher degrees, and the value of getting out in the world and building startups.
But you can have both – case in point, a recent interview I did with Aravind Srinivas.
At the very beginning at the very beginning of his remarks, Srinivas noted how his parents are prouder of his PhD than his business track record, even though he’s the cofounder of one of the most famous model services around.
Encapsulating this idea, he boiled down a principle he has seen in his community of origin into a few simple words: “seek knowledge even more than wealth.”
Starting Out
As for his early days in tech, he talks about YouTube videos and coding competitions, and his graduate degree at Berkeley, followed by some internships.
“People around here were way more intense, hard-working, way better at actually, not just coming up with new ideas, but implementing them and getting them working, or writing all the code to scrape data sets …. so that taught me a lot, and pushed me more towards entrepreneurship.”
Re-Orienting Ambition
There’s a very interesting part of this interview where Srinivas talks about getting an interview at OpenAI and presenting some of his ideas to Ilya Sutskever.
Sutskever, he said, told him to his face that his ideas were bad. It upset him – but it set him on a different path.
I thought it was extremely interesting when Srinivas actually narrated part of the interaction and then showed how it informed the work being done over the past few years:
“(Sutskever) took me to a board and just drew two circles, one big circle, and inside it a smaller circle,” Srinivas said. “So the big circle is generative unsupervised learning, and the smaller circle is reinforcement learning. And you don’t need anything new. You just need to do this in sequence, and throw a lot of compute at it, and train on (the whole) internet, and then you build the general intelligence. … nobody had any clue at the time, but he just saw the future, and that ended up being right.”
And again, he stressed the path of practicality.
A Demo, Not a Deck
Srinivas said he hasn’t made a presentation deck in ages. Instead, he said, just show potential investors what you do, and what your products do.
“The number one thing to do is just iterate and do something,” he said. “I’ve seen many founders spend at least six months to a year in the idea maze, going around and around and not getting anywhere, not knowing what it takes to actually (show people) something, get in the hands of people, see them use it, learn from that, and then go and update your hypothesis about the world.”
Evolving Search
Referring to Google user intent programming, Srinivas talked about how search is a fundamental service, and what that has meant for the industry. He came up with some pointers: for instance. making sure news sources are prominent, and exposing the chain of thought, and urged innovators to “change the paradigm.”
Srinivas also suggested that a company might buy Twitter, for example, promoting a “healthy mix of useful content” and strengthening the community notes feature, where, he said, we might do well fact checking with software.
Later, he pointed out how Hollywood simplifies the process of innovation:
“Academic people don’t have the discipline to try small scale experiments,” he said, citing movies like “A Beautiful Mind” and “Good Will Hunting” where observers just marvel at a grand idea, scrawled across a whiteboard, in idolatrous glory. In reality, he said, it’s better to iterate – to come up with things one piece at a time.
Thoughts on the Eventual Outcome of AI
Evaluating the difference between closed models, where a few oligarchs control the output, and democratized AI, where the people have the models in their pockets, Srinivas said it’s extremely important that we end up with the latter type of arrangement.
“Open source is the only thing that can keep people in check here,” he said. “I hope we end up with the utopian outcome.”
Silicon Valley Unleashed: Art Imitates Life
Midway through the conversation, Srinivas actually referenced the Silicon Valley TV show. Further, he talked about the show’s significant impact on the tech community, which I thought was so useful in looking back at that time.
“Someone actually told me: ‘hey, look, I know you’re making jokes about this show all the time, but realize that some people might take it personally, because it’s actually pretty true,’” he said. “’It’s not just a humorous show.’ A lot of people were very depressed watching it, because it reflects Silicon Valley in almost a brutal way.”
He did note that the theme of lossless compression as a holy grail, for him, translated into the real world, too.
“The particular idea, in that TV show, that was explored was around lossless compression, which is very directly generative AI,” he said. “They also talked a lot (in the show) about how you can use neural nets. So that was actually the idea I really wanted to start up first on: lossless compression with generative models.”
So there you have it – Mike Judge’s classic take on the budding community of innovators in California influenced at least one high-level disruptor who went on to work on ground-breaking LLM projects.
As for the fictional Peter Gregory’s advice, take that into consideration – if you really value the abstract stake that learning gives for exploration, academia may be the way to go, (especially if you want your parents to be proud of you). But don’t neglect that urge to actually come up with something, and make it real.
