Learning with Co-pilots
“It’s only when the tide goes out that you learn who has been swimming naked.” — Warren Buffett.
After sixteen months of bear market, we are seeing the dominoes fall, from SBF to SVB, and more likely to come. While the end is not in sight, a new bull market is already starting. Generative AI is skyrocketing with hype, enthusiasm, and fear all happening at once. New AI tools are hitting the market daily and venture investors are making a comeback with big bets in the hundreds of millions. Bill Gates wrote “The Age of AI has begun”, while GSV advisor and AI expert Elijah Mayfield told us, “we are in a true Model T era for machine learning, and we’ve all spent the last 40 years building horse-drawn carriages.”
To emphasize the magnitude, just last week, 100 new Generative AI apps were released! AI tools are going viral, and launching them is now as easy as posting on TikTok. This begs the question: Who will be differentiated in an ocean of bots?
Learning is also getting disrupted and accelerated, with large parts of the process automated; whether writing essays, solving equations, or learning to code, we are now flooded with AI tools that complete tasks on command. It’s like we went from learning to drive a stick shift to just saying, “KITT, I need you.”
Microsoft announcing 365 Copilot and Google releasing Bard show that a lot more automation is upon us — from creating presentations, to building models, to writing emails, everything can now be done with a simple instruction. We now have an accelerator at our fingertips that’s connecting a ton more dots of information instantly.
The scale and speed are perhaps best displayed by OpenAI’s web traffic, which went from 0 in October to 1.1 billion in February. More importantly, 65% of users are below the age of 34, and ~30% are under 25, implying significant adoption among students. And btw, this is a global trend and the U.S. is only at 16% of global traffic.
Source: Similarweb
In a recent ABC interview, OpenAI’s CEO Sam Altman and CTO Mira Murati, the brainchild behind ChatGPT, said this is the greatest technology that mankind has ever seen. Both agree there is a risk the technology can become harmful and even reaching “a point of no return”, but hope we won’t get there. The future choices that leaders at OpenAI, Alphabet and elsewhere will make, will be quite important for us humans. And certainly, governments will play a key role in deciding what should be allowed and what not. In an interview with Walter Isaacson, former Google CEO Eric Schmidt talked about the problems and risks that Social Media created and said that we need to make sure that Generative AI doesn’t make the same mistakes.
Meanwhile in education, surveys show that already 50% of teachers use ChatGPT for course creation, and more than 50% of students use it for homework. A recent MIT study also found that ChatGPT users are completing tasks 37% faster with higher output quality, which is no surprise.
Given the fast pace of change, we can already see how learning is changing. Writing, calculating, reading, coding, are all getting automated, while tutoring is becoming hyper-personalized with co-pilots tailoring the content to the individual learner. In the future, the role of teachers will likely be more important given the complexities and risks of AI and how they could impact kids progress and development. Teachers will have to teach fundamental skills such as critical thinking, creativity, collaboration, etc. — Michael Moe’s 7 C’s framework — and also teach the fundamentals of Math, English, Physics, etc. Given that all students will have learning co-pilots in their hands, they will need to be taught how and when to use them, and how to interact with them in order to learn efficiently. For a long time, companies have been trying to tailor learning to the individual but it’s only now that generative AI and LLMs are connecting a ton more dots and predicting learners gaps instantly.
When Tik Tok (at the time called Musical.ly) took Gen Zers by storm, it helped design a new category of creators. Kids were given a fun tool for expression and creativity with dance, lip-synching, and stickers all in one app. Because of the cool experience and the powerful tools, TikTok became the “learning playground" for a new generation of designers and influencers. In a similar vein, Generative AI is now the big muscle that’s accelerating learning and productivity, and the emergence of co-pilots is going to shape the new generation of students and workers.
We are entering the early innings of what will be existing times, with Generative AI creating a whole new playing field for companies that are serving learners. In K12 and Higher Ed, students will gain deep personalization that will lead to better outcomes, while teachers will gain powerful tools for teaching, content creation, etc. In the workforce, co-pilots will take over repetitive and mundane tasks and free up time for critical thinking and problem solving. The fast pace of change will also increase the need for up-skilling as larger parts of the workforce will need to be retrained.
Some companies are already experiencing acceleration with Generative AI. Replit, a leader in coding, recently launched Ghostwriter Chat which is an assistant that autocompletes code. Now you can let it do the work and you tweak the results. In fact, you can talk to Ghostwriter and tell it to create an app, and just in seconds magic happens. We are basically moving away from needing to learn code to needing to learn rhetoric. And as for companies, why hire engineers if AI will do the work at Hello… It’s like we were learning to play tennis and suddenly got a partner whose name was Roger.
Today, Replit has more than 20 million developers, up ~70% from a year ago, and is accelerating as more and more non-techies start to join. This week, Replit announced a partnership with Google that will help accelerate the creation of Generative AI apps and will supercharge everything.
Perplexity.ai is a search tool with a “truth-first” approach built on GPT. Responses come with fact checks and quality sources, and the experience is engaging. The company launched less than 4 months ago and has already received more than 13 million monthly web visits. Meanwhile, Caktus.ai is a content generation tool that helps students with writing, coding, and studying, and has a stronger emphasis on STEM subjects. While it’s received good reviews, It’s not a freemium model which could pose a challenge given its audience.
Quillbot, which analyzes writing and generates suggestions, is gaining popularity with students and workers as shown by its website traffic being up ~2x from a year ago. Meanwhile, Grammarly also announced its co-pilot, GrammarlyGO, which will use context like voice, style, and purpose to determine what it suggests.
In language learning, Duolingo has done impressively well in an environment where others have stalled. CEO Luis van Ahn integrated AI and personalization from early on, and is now going into other areas such as reading and math. This should help the company capture learners earlier and keep them engaged for longer. With an expanding LTV and a low CAC (marketing spent is only 18% of revenue), the business model becomes increasingly more attractive. Not surprisingly, DUOL stock is trading at 9.1x forward revenue, which is a meaningful premium to most tech companies today. (See Jorge Mazal’s recent post on how Duolingo cracked engagement and reignited user growth.)
Another impressive language learning company is Speak. Last year, their app surged to the top of the rankings in Korea, the first market they launched in, and now it is surging to the top in Japan, their second market. Speak is an AI tutor that holds open-ended conversations while giving feedback on grammar, pronunciation, and vocabulary. Unlike Duolingo, which focuses on gamification, Speak gets you to speak the language. It’s always fascinating how students spend years learning a language but barely speak a sentence when needed. That’s exactly what Speak is solving, as it is forcing verbal conversations in the learning process. Already, more than 100K subscribers are generating double digits of millions in revenue, and that’s just in Korea… The product is differentiated, personalization is supercharged by GPT-4, and the company is led by two impressive founders, Connor Zwick and Andrew Hsu.
More companies are now announcing integrations with GPT-4 including Quizlet, which introduced its AI tutor Q-Chat, and Khan Academy, which introduced its tutoring co-pilot Khanmigo. Generative AI is essentially the new electricity with everyone tapping into LLMs. Accordingly, differentiation is becoming more critical and success will be determined by a combination of a superb user experience, information accuracy, and learner outcomes. A tool that writes your essay won’t help you learn and therefore won’t succeed.
At the venture stage, generative AI is already a hot commodity, but in the public markets, investors have yet to react to the potential effects. It’s still unclear who’ll get accelerated versus decimated. After 16 months of bear market, we thought we were maybe out of the woods, but SVB and the subsequent dominoes suggest otherwise. With US inflation still high at 6% and the interest rate creeping up to 5%, we remain in an environment that only rewards growth with profitability.
Looking at the 13 public Edtech companies, there is a nearly perfect correlation of 0.9 between the companies forward revenue multiples and their “Rule of” number, which combines revenue growth with EBITDA margin. For example, Duolingo’s E’23 revenue growth is 35% and its expected EBITDA margin is 12%, so it is a “Rule of 47” company and is trading at 9.1x revenue. Meanwhile, Udemy’s expected revenue growth is 16% and EBITDA margin is -4%, so it is a Rule of 13 company and is trading at 1.2x revenue. Plotting all Edtech companies on the graph shows a clear distinction between those making a Rule of 30 or higher, with strong multiples, and the others who are essentially in the penalty box.
As public companies are making the switch to profitability, it’s taking a toll on their spent and on growth. Based on the individual outlooks for 2023, the average EBITDA margin is expected to increase from 15% to 18% while the average revenue growth is coming down from 32% to 17% — will likely end up higher as companies guide low to “beat and raise” as the year progresses. All 13 companies project EBITDA margin expansion while only two expect top line acceleration.
While the IPO window is locked, the pipeline of late stage companies is building up. In the GSV 150 list of the fastest growing and most impactful private companies, we have 27 companies with $250M+ in annual revenue. With Generative AI as a new catalyst for many, we expect some serious candidates for the public market.
It will be critical to focus on the RoE (Return on Education) that companies deliver. Whether in a B2C or B2B model, those who add the highest value to learners will achieve the best retention and the longest LTV. If learners receive tangible value and have reasons to come back again and again, they will drive network effects and create organic growth. Companies that can constantly deliver the highest value to learners as measured by how learners are thriving in school or in life, will be the ones racing to the stars.
For now, all the excitement is in the early stage rounds of Generative AI with mega rounds making the headlines; Character.ai raising $200M for its conversation generating technology, Inflection raising $675M for its AI-powered personal assistants, Humane raising $100M for the development of AI-powered hardware, and Cohere raising hundreds of millions at a $6 billion valuation to give businesses access to NLP. (To keep up with everything in AI, sign up for Claire Zau’s weekly newsletter.)
