EIEIO...Murder or Manslaughter?
“Good is Dead, Excellent isn’t Enough, Outstanding is the Future” – Tony Robbins
“I’m still confused as to how a non-profit to which I donated ~$100M somehow became a $30B market cap for-profit. If this is legal, why doesn’t everyone do it?” – Elon Musk
“We were wrong. Flat out, we were wrong.” – Ilya Sutskever, OpenAI’s Chief Scientist
In October 2016, the New Yorker wrote a profile of Sam Altman titled “Sam Altman’s Manifest Destiny.” The subtitle is just as telling today as it was then: “is the head of Y Combinator fixing the world, or trying to take over Silicon Valley?” We’re still trying to figure out the answer.
OpenAI’s GPT-4 announcement took the Tech World by storm this week. But it’s what OpenAI didn’t announce this week that’s grabbing our attention. As Shuki Cohen, Head of Data at AI21 Labs put it, OpenAI is *officially* not “opening” AI.
Was it premeditated murder or involuntary manslaughter?
The first hint is that OpenAI didn’t mention any of the researchers who contributed to the paper. This is one of the first significant AI research papers that did not mention the authors.
This may seem like a small gesture that it’s the OpenAI TEAM that comes first…but it’s bigger than that. This isn’t Penn State and Notre Dame not putting the names on the back of their jerseys. It's more like Alabama or Georgia winning the CFB…and then refusing to release the names of their players.
Given that the authors of the eight seminal papers on LLMs have already raised billions of dollars for their own companies, these names are just as valuable as the top athletes in the NCAA (or the NFL for that matter).
Robert Smith famously said that the most scarce commodity on the planet is software engineers. The engineers behind GPT-4 are a very valuable connection to have on LinkedIn.
The second (and more troublesome) hint is that OpenAI refused to release anything about the contents of their training dataset. OpenAI’s 98 page paper introducing GPT-4 disclosed nothing about their model, training methods, compute, or hardware.
While there’s a tradeoff between “competitive and safety considerations” versus the “scientific value of further transparency,” we’d be remiss if we didn’t mention the third factor that OpenAI is prioritizing: enterprise value.
OpenAI is making two statements by refusing to disclose their training data.
The first statement is that OpenAI may be a Mission Corp, but it’s still going to compete like a for-profit business. While open source has been the status quo of Silicon Valley for decades, Sam Altman is taking the same position on open source that Steve Jobs took before him.
The second statement is that by being silent about where they train their model, OpenAI is being loud and clear about where they get their data: by stealing it from all over the web.
When Reddit COO Jen Wong was asked about the threat of ChatGPT this week at SXSW, she put it bluntly: “where do you think they get their answers from?”
The question for OpenAI is if they’ll be able to make the jump from piracy to legitimacy. We know who’s on Team OpenAI , but the counterforce is rapidly putting together its coalition. We’ve seen this story many times, with one end of the barbell being Napster with the other being YouTube.
Napster went out of business in just over a year after a fierce battle with the RIAA. Meanwhile YouTube – thanks to the help of Google – became the second most visited website in the world.
In 1966, Stanford philosophy professor Patrick Suppes predicted that one day, computer technology would evolve so that “millions of schoolchildren” would have access to a personal tutor. He said the conditions would be just like the young Prince Alexander the Great being tutored by Aristotle.
57 years later, with the introduction of ChatGPT, (and as of Tuesday, GPT-4) there is the potential for every student to have 24/7 access to this tutor in their pocket. This tutor has infinite energy, patience, and data access…and can now understand images. It can literally turn an idea on a napkin into a website.
The ratio of students to teachers in K–12 education in America is 15:1. Think of a classroom in a standard US public school classroom - of those 15 kids, how many are engaging with the teacher? How many have expensive tutors walking them through the homework at home every week? How many have the confidence and self esteem to raise their hand and ask questions in front of their peers when they don’t understand something?
The best teachers and tutors in the country are those who take the time to connect with their students on a personal level and understand what learning style suits them best on an individual basis. There is a small subset of great teachers who over the course of the school year begin to glean what teaching styles portions of their classes are more receptive to compared to others. If a student is lucky, the teacher will have figured out what teaching style works best for them by the end of the first semester.
With ChatGPT and GPT-4, this mobile supertutor analyzes students’ learning styles in real-time, and can offer content in endless formats. GPT-4 can teach students using memes, conversations with historical figures (Julius Caesar teaching Political Science, Walter White teaching Chemistry), and can even explain anything in basketball terms.
MasterClass built a $3 billion dollar business by enabling students to learn from the world’s best; now, anyone can have a conversation with them using GPT-4.
While GPT-4 has gotten headlines for getting answers, its impact can be just as large as a tool for asking questions. While teachers say that there’s “no such thing as a stupid question,” most students are held back from asking questions because of fear of judgment from their peers.
Office hours were rarely effective for me…not to mention students who are working, taking night classes, or have children. I could not set the time, they could be all the way across campus, and there was often a line to ask the TA specific questions. I sometimes left more confused and frustrated than when I arrived.
However, as Dylan Field shared in a recent interview with Elad Gil, these new tools are "like a professor that you can go to office hours for at any time."
Generative AI-powered tutors will augment the way teachers teach and students learn in countless ways. A teacher can sort a semester’s worth of assignments by each individual student, and GPT-4 can provide detailed analysis of each student’s strengths, weaknesses, and recommend personalized interventions that point the teacher toward specific resources that will improve student outcomes.
Schools and school districts that have enacted knee-jerk bans on this technology are fighting gravity. Critical thinking and creativity are required to put this innovative technology to work. Without it, no one will identify use cases or solve real problems. Who cares if they no longer have to remember all of the grammar rules of the English language?
As UT Professor Steven Mintz puts it, “much as Google devalued the steel-trap memory, electronic calculators speeded up complex calculations, Wikipedia displaced the printed encyclopedia, and online databases diminished the importance of a vast physical library, so, too, platforms like ChatGPT will profoundly alter the most prized skills.”
So what does this mean for the future of learning and education?
On Youtube, you can already learn nearly anything…instantly. If the best creator in ANY category knew that they could get distribution and get paid for their services, they would. Anyone can learn from the best teacher in the world in any given field, and the teachers will make great content because they will be rewarded – not just with clicks, but with cash.
So what will this do to campuses and higher education? Will students have any reason to pack up and ship off to live at school with a group of strangers for four years when they can access so much from their own home?
Just like we’ve seen with the Future of Work, we’re predicting a “barbell strategy” will win out: the best universities will evolve to maintain their brand equity and social currency, while the rest of “learning” will be digital-first platforms offering learning in a faster, better, and cheaper way.
Whether on campus or online, the best universities allow students to surround themselves with similarly driven and, like-minded individuals, creating a strong sense of community. As Dylan Field put it, "universities are a social club for life.”
Some of the best learning people do in college happens during late night conversations with their dorm mates, or during spontaneous chats in line at the campus coffee shop. People find jobs, spouses, and lifelong friends on college campuses.
So many elements of college life have no replacements. Lots of students get the bulk value of their college experience from the fraternities/sororities they took part in, the teams they played for, and the clubs they joined.
Because of this, the world’s most elite universities with the strongest reputations and alumni networks are more likely to survive (and thrive). There’s a reason that the brands of the oldest universities have lasted hundreds of years longer than the oldest businesses in the United States.
However, lower-tier and regional universities may have to be repurposed due to the coming AI education wave. If it is not materially beneficial to be on campus collaborating and networking in person, people will likely not spend time and money living on campus.
Just like the Future of Work changed how we use offices, turning specific campuses into hybrid colleges/community innovation and upskilling hubs is a great use of space, staff, and equipment…that would still provide massive value.
There’s the question of what this evolution will look like in practice. The downside of everyone having a personalized tutor is that students still benefit from structure and community. Telling a 16-year old, “you now have all of the information in the world at your fingertips and an LLM to explain it to you, go nuts” is not an effective way to breed impactful members of society.
However this evolution plays out, there will be new forms of credentialing, and different credentials will become far more valuable than others. Showing what you know will be crucial in a Knowledge Economy where knowledge is a currency.
People who “talk the talk” on their LinkedIn and resumes will lose out to individuals who “walk the walk” with knowledge portfolios that exist on an immutable ledger. There’s more incentive to be a lifelong learner when that portfolio is the key to earning potential and societal impact…not a degree that’s on your wall from thirty years ago.
Volatility, instability and fragility were all part of the Market dance last week. The rescue of depositors from Silicon Valley Bank’s demise were enough to send tech soaring last week with NASDAQ advancing 4.4%, followed by the S & P 500 gaining 1.4% and the Dow dropping .1%.
On the winners side, big tech names such as Microsoft and Alphabet increased over 12% for the week. META, despite (or probably because of) taking a several billion dollar restructuring charge and firing another 10K people, on top of the 11K people fired earlier. Some simple math, if each person was making $150K on average, that will save META over $3 billion annually.
On the losers side, even with its bank brethren pitching in $30 billion in deposits, First Republic fell 70% for the week, and 33% on Friday alone. Credit Suisse has been on the ropes for months and received a $50 billion lifeboat from the Swiss National Bank but still fell 14% for the week. Major bank consolidation is likely to be the next chapter and Barron’s cover story over the weekend was titled “Buy the Big Banks.”
Given all of this, it wasn’t shocking that Gold made a 11 month high. Additionally, and initially counter intuitively, Bitcoin roared up 32% for the week (we remain positively inclined toward Bitcoin).
These are times that test one's resolve but are also the periods where opportunity is greatest. We will focus on what we know and remind ourselves that volatility is the friend of the long term investor.
GSV’s Four I’s of Investor Sentiment
GSV tracks four primary indicators of investor sentiment: inflows and outflows of mutual funds and ETFs, IPO activity, interest rates, and inflation. Here’s how these four signals performed last week:
#1: Inflows and Outflows for Mutual Funds & ETFs
#2: IPO Market
The US IPO market remains quiet, but things are hot under the sun in the UAE. At least ten companies across various industries are expected to list in Abu Dhabi this year, while Presight AI’s IPO is 136x oversubscribed with $25.8 billion of total gross demand.
Source: Renaissance Capital
#3: Interest Rates
This week, Goldman Sachs economists said that they no longer expect the Fed to hike rates next week. That’s still a contrarian view: as of midday Thursday, markets were pricing in an 83% of the Fed continuing with another 25 bps hike.
Inflation has been easing for eight months versus prices a year ago, but it’s still a long way from the Fed’s 2% annual inflation target. Meanwhile, inflation in Argentina broke the 100% barrier for the first time in 30 years.
Videos of the Week
Chart of the Week
Chuckles of the Week
Entrepreneurship: $25,000 — value of a startup in 24 hours created by HustleGPT, an entrepreneurial AI (Source)
Innovation: $50,000— amount the CEO of Match Group has personally spent on Clash of Clans in the last 3 months (Source)
Education: $6,065 – average annual tuition of more than 200 New York City private schools (Source)
Impact: 39% – amount of Americans who suffered from a sleep disorder in the last 12 months (Source)
Opportunity: $36,000 – amount that a $100,000 salary feels like in New York City (Source)
Connecting the Dots & EIEIO…
Old MacDonald had a farm, EIEIO. New MacDonald has a Startup…. EIEIO: Entrepreneurship, Innovation, Education, Impact and Opportunity. Accordingly, we focus on these key areas of the future.
One of the core goals of GSV is to connect the dots around EIEIO and provide perspective on where things are going and why. If you like this, please forward to your friends. Onward!
Make Your Dash Count!