EdTech Thoughts Weekly Update - 6/13
Beachfront property on Chegg Island
Hello!
I hope that many of you enjoyed Sunday’s edition of Weekend Reading. As a reminder, this Weekly Update is for paid subscribers only - see below for how to upgrade.
This newsletter covers my resilient belief in Chegg’s ability to navigate the ChatGPT threat, Stepful’s $7.5M funding round to train more junior healthcare workers, an argument for more marketing majors to go into Tiktok farming, Byju’s choice to get litigious, and Khanmigo’s place in the EdTech ecosystem.
A note to paid subscribers: I experimented with a new format this week. Now that the free newsletter provides a broad cross-section of news, I’d like to use this space to go deeper on a smaller number of topics. The goal remains the same - to keep you informed on the most important stories in education - but I’m hoping that a longer-form space helps both of us develop stronger opinions on the topics covered.
With that, on to the news!
1. ChatGPT needs better training data, who is going to pay?
Based on the company’s stock price, it is quite possible that I am the last remaining resident on Chegg Island that is not employed by the company. If we are to believe the good people of Wired, The Information, The Wall Street Journal, and Bloomberg, I may have bought property on Rodanthe.
This article helps illustrate why I am putting storm shutters on the house rather than abandoning the property. To summarize the author’s core premise, today’s large language models are trained on poor data. Specifically, “for [pre-GPT4] versions [OpenAI] used an online corpus of thousands of self-published books, many of them skewed toward romance and vampire fiction.” OpenAI’s training is not anamolous.
Vampire fiction is great for achieving language fidelity, but not great for establishing…facts.
I suspect that OpenAI is working on how to bridge this gap. Reddit’s poorly-received pivot to charging for API access is (probably) one of OpenAI’s first steps in this direction. The upside of being able to charge LLM providers like OpenAI market rates for access to their data outweighed Reddit’s user concerns.
Reddit is the center of internet culture, which enables it to charge substantial rates for LLMs to train on their data. Chegg’s negotiating position is not quite so plum. But it is not hopeless. It is still in OpenAI’s interest to answer student questions effectively and the breadth and depth of Chegg’s dataset remains the best in EdTech. If execution follows logic, an OpenAI partnership *should* reinforce of Chegg’s position in the market. We will see if the company can adapt quickly enough to survive the AI storm.
2. Stepful raises $7.5M
Successfully filling junior healthcare positions at scale is the holy grail of workforce development right now. No fewer than 5 startups have raised venture funding in the past ~4 years and almost every major EdTech company with a workforce division has a healthcare play - not to mention all the universities and states that have their own stakes in the training process.
New York-based Stepful is Reach Capital’s bet on a breakout hit in the space.
The non-intuitive thing about the healthcare workforce market is that monetization of learners is usually not the issue. Employers are so desperate for quality talent that they are often willing to foot the cost of training + a healthy margin for the training provider. Even where employers are not an option, there are a multitude of financing options that make these training programs relatively accessible to learners.
The bigger issue is finding learners and designing a program that consistently graduates them. Stepful is still too nascent a company to declare whether they’ve “cracked the code” or not, but the team’s experience bears watching. The company’s CEO comes from Uber’s driver recruitment division, which gave him experience with a user demographic that tends to look similar to those recruited into many junior healthcare roles. They do not yet have a learning/instruction leader, but others on the team add experience in the recruitment space and technical expertise.
3. The Ag influencer goes to college
The EdTech community spends a lot of time talking about “experiential learning.” I am a fan of this! But the examples usually involve pushing a learner from one career trajectory to a tech-focused one. That is not a bad thing, but it is somewhat limiting.
What I love about this story is that the protagonist, 20 year-old farmer and Washington State University marketing major Kaitlyn Thorton, is building an exciting career path that leverages tech, but is not dependent on it. She is excited to use her marketing skills to build her family’s farm for the next generation rather than working a generic entry-level marketing job.
I’d love to see more vocational work < > tech crossover stories like this. The subjects may not all have 350K+ followers on Tiktok, but bringing more technology into vocational businesses is a path that is more lucrative than it is given credit for.
Now that college career centers are being funded to match the conversation about post-graduate employment outcomes, they should spend just as much time encouraging these types of diamond-in-the-rough job opportunities as they do sourcing high-volume corporate jobs.
4. Byju’s chooses not to make a $40M loan payment; sues investors
This is a morbidly fun example of “when you owe the bank $10,000, that is your problem. When you owe the bank $1.2B, that is the bank’s problem.” Given all the Byju’s financing noise this spring/summer, the company’s choice not to make the payment was not shocking, but it does commit the company to a chaotic path forward.
When a company stops paying their debt, you might expect a relatively orderly process where the terms get re-negotiated. The company’s debt investors would step in and figure out how they might be made close to whole by forcing the company to find the cash, sell assets, or take/require loan collateral like equity in the company. In this case, I suspect the asset/equity in question is Byju’s (profitable) subsidiary Aakash.
The wrinkle here is that Byju’s debt investors are not just negotiating against the company and a normal collection of investors, they are negotiating against the state of Qatar. That means it is entirely possible that the company is able to get out of what has become a particularly onerous loan - made when rates were lower, but, presumably, with a floating interest rate - without too much immediate damage. However, I have to believe that doing so significantly reduces the company’s funding options outside of the Qatari soverign wealth fund for the foreseeable future.
5. Most clicked link from Weekend Reading: Khanmigo For All
“It’ll enable every student in the United States, and eventually on the planet, to effectively have a world-class personal tutor”
It feels sort of impossible to read that line and not think of Neal Stephenson’s Diamond Age, which gives narrative life to the concept of a personalized AI-tutor.1 Forgive the inelegant language, but the concept is just really cool.
The organization with the early lead on bringing this concept to reality is the non-profit Khan Academy via their recently-released Khanmigo. Of specific intrigue to the EdTech community, Khanmigo is being designed with real go-to-market intent - charging $20/month to consumer users. (I presume that the price will be lower if/when the company sells in bulk to districts.)
What would it mean for a non-profit to control the default user experience for a significant number of learners?
This reads like a loaded question, harkening back to the debate over for-profit universities, but I mean it much more tactically. For the past ~20 years, private equity has dominated ownership of large EdTech companies (the major publishers, software infrastructure providers, and universities). That is not a generically good thing, but it is relevant that PE firms - both directly and through their subsidiaries - provide the vast majority of exit options in EdTech.
That doesn’t necessarily change if one of the major players is a non-profit, but it would mean that one of the seats at the table is occupied by a stakeholder with meaningfully different financial incentives.
Bonus news: New rules on State Authorization and Reciprocity Agreements (SARA) | in ED Gainful Employment regulation
An emerging topic that hasn’t gotten much coverage outside of the above links from Phil Hill and Whiteboard. A helpful TLDR tweet thread of the change’s impact.
Question of the Week
Stealing the question of the week for product feedback! As noted above, I took a pretty different tack in today’s paid newsletter - did it resonate? Please feel free to reply to this email and/or call/text me with thoughts. As always, votes remain anonymous.
Results of last week’s poll: I am surprised by this answer! You could convince me that Maximal Learning has the highest potential upside of the three, but we have no idea what business models are going to work for adoption of personalized technologies. Antimatter (standard B2B SAAS) and Data Masters (bringing an established training model to a new geography) feel like they have far simpler - and therefore faster - go-to-market paths.
A warning to those who are not Stephenson/sci-fi readers - the book is excellent buy ZANY. It requires you to really commit to the concept to get something out of it.