The Disruptive Potential of GenAI for Emerging Markets | Rem Koning | Leading with AI
What this covers
Speaker: Rem Koning
How will GenAI transform entrepreneurship and business in emerging markets? In this workshop, Harvard Business School Professor Rembrand Koning shares cutting-edge research on how AI is starting to transform and disrupt entrepreneurship, innovation, and business across the globe. His talk walks through experiments, case studies, and new frameworks that shed light on how you can leverage AI's potential to drive productivity and growth in emerging markets.
Source description (no synthesized summary yet).
Generative AI has significant disruptive potential for emerging markets, but effects are heterogeneous: high-performing entrepreneurs benefit substantially (10-20% revenue/profit gains) while low-performing entrepreneurs paradoxically perform worse (-10%), because they apply AI to different problem types (growth vs. survival) requiring different solution characteristics.
- Low performers use AI for existential problems (competition, bankruptcy) where generic advice about financing is unhelpful; high performers use it for growth optimization where specific tactical advice drives results
- The divergent effects contradict patterns seen in developed markets (US/UK/Europe) where AI helps lower performers most, suggesting emerging market context fundamentally changes who benefits
- Success requires matching problem type to AI capability: growth problems with capital and demand can be solved with strategic/tactical advice, but survival problems with structural constraints cannot
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Current AI models risk reinforcing existing biases and inequities in two ways: (1) training data comes primarily from developed world sources, so the AI reflects developed-world contexts and norms, and (2) the people designing the algorithms are predominantly from developed countries, limiting perspectives embedded in system design.
“It has the potential to be reinforcing existing biases and also um and inequities. In the sense that most of the data right now that's coming from all the AI models is coming from uh you know, the more developed world. Yeah. And a very small portion of that is also where where the the world in general is Yeah. is very different, looks very different. And also, um people who do the coding and algorithms as well. Yeah.”
Introduction of cell phones to Kerala fishermen created market convergence where all fishermen converged on a single market price through text messaging about prices, leading to more efficient markets, expanded operations, and better supply chains for restaurants—demonstrating that simple information availability can have powerful economic effects.
“So, this is over time. And here were prices. And what you see is that each line represents a fisherman. And what you see in the market is that these fishermen, there's four of them here, none of them know the prices for fish. They go fishing every day and supply has to meet demand and they're deeply confused about what I sell for... Suddenly, cell phone towers open. And they all get cell phones... What happens to prices? All of them converge on a single market price. Because you can look at the text messages. They're texting each other the price. Market operates more efficiently.”
Women are significantly less likely to use ChatGPT and other generative AI tools compared to men, across multiple countries and data sources. This gap persists even when controlling for profession—software developers who are women are 20-50% less likely to use AI tools than male software developers with the same job title in Denmark. This suggests a barrier beyond professional necessity.
“One of the things we do see about gender, which is true in this paper, true in a paper out of Denmark, true in the US, if you go on similarweb.com, you can look at the genders uh of who visits a website. Women are much much much less likely to use ChatGPT and other generative AI tools. And this holds even if you look within a profession. So you may say, 'Oh, clearly there's an explanation.' And the explanation is because there's more male engineers, engineers, software developers use these AI tools more. So, I can zoom in on software developers. And I can take a man and woman software developer with the same job title in Denmark, exactly the same on like all characteristics, women are still 20 to 50% less likely to use AI tools.”
Small and medium businesses produce a substantial share of global economic output, often struggle with productivity, and improving their productivity via AI tools could generate massive gains in personal income, economic growth, and ability to afford education—justifying continued focus on SMB productivity tools even while other domains (health, education, finance) may offer more exponential returns.
“I would say given the share that small and medium businesses produce of the global economy and the fact that we know they often struggle with their productivity, if we can improve their productivity, massive gains in terms of people's income, in terms of growth, in terms of the ability to afford to send your kids to school.”
The speaker created a minimal viable version ('minimum viable AI') as a deliberate design choice, deliberately testing effectiveness before scaling, as a responsible approach to deployment that avoids accidentally harming businesses through poorly-tested tools.
“I think this is why you need to experiment cuz you don't want to roll this out on Facebook and like wipe out a huge number of businesses accidentally.”
There is active work on language data infrastructure, particularly for smaller languages in Africa (thousands of languages on the continent), requiring subsidization and coordination to ensure representation in LLMs.
“I'm seeing a lot of work around language data. So for smaller languages, there's thousands of languages spoken on the continent of Africa. What what what do we need to do around making sure that those get included in these LLM systems.”
GPT-4 was trained on massive internet data including World Bank reports, HBR articles, newspaper articles from Kenya, and other global sources. This training data means the model has surprisingly good knowledge of emerging market contexts like Kenya, knowing terms like 'matatu terminus' (bus terminus) and being able to answer questions about Kenyan dairy prices, though with some data drift since training.
“So, this is what I love about generative AI. If we were in the pre-generative AI world, your question would be like table stakes. We are doing the dumbest thing you can think of. I would like to just like point out. We're taking GPT-4 which is trained on the internet, basically everything. So, it has every World Bank report ever. It's got a bunch of HBRs. It has cases despite them saying it doesn't have cases. We tested this out. Right? It has newspapers in Kenya are going to be in the data.”
Nubank, based in Brazil, is the largest online bank in the world. This example demonstrates that developing markets can build global-scale financial technology companies, not just adopt technology from developed countries.
“The largest uh bank in the world, online bank, is Nubank. Yeah, out of Brazil. Out of Brazil. Yeah, right.”
Policy makers should work with governments and international organizations (like the World Bank) to determine which data and services require subsidization versus which can be left to private market incentives in emerging markets.
“I'm a big free marketeer. Like let's have entrepreneurs build companies and succeed. Where are the places that governments, policy makers, I've been working with the World Bank on this, need to come in and subsidize data and where where where is it not?”
Generative AI text-based tools have disruptive potential specifically for emerging markets because mobile phone penetration (70-90% WhatsApp adoption in Kenya) is high enough to deliver information-based interventions that can radically change outcomes for entrepreneurs with limited access to expertise and advice.
“everyone in the world has a phone. In terms of I think the penetration rates of WhatsApp in Kenya, you're talking 70-90%. Right? Everybody is on WhatsApp and Meta platform to a first approximation. Right? And so, suddenly, you can now do simple things like providing information that can have huge impact.”
The speaker emphasizes that after 18 months of rapid AI deployment, anyone making confident predictions about how these tools will evolve should 'take a humble pill,' acknowledging the difficulty and importance of maintaining intellectual humility about AI trajectories.
“We are a year and a half in. Anyone who tells you how these tools are going to evolve should take a humble pill and sort of learn that these predictions are hard to know.”
One of the five key open research questions is whether high-quality contextual data is necessary for AI to be useful. Rem states: 'Some days I wake up on one side of this question and other days I wake up on the other side'—meaning the evidence is genuinely unclear about whether AI needs rich contextual information or whether it can achieve sufficient value without it.
“Does more contextual data improve the value of AI? This goes down really to this fundamental question of how important is it to have really rich data on the context or not? Maybe these AIs are kind of good enough without context. Maybe they need really rich data about the places they are. That's an open question. I don't know. Some days I wake up on one side of this question and other days I wake up on the other side.”
Leapfrogging is possible in emerging markets: countries can skip intermediate stages of technology adoption. Example: Brazil experienced high economic volatility, leading to rapid adoption of WhatsApp payments and direct-to-account payment systems (Pix), which then became entrenched. When you introduce new technology, adoption can be fast, and once people are in that market, 'there's kind of a no way back.'
“I think emerging markets have the ability to leapfrog certain Yes. uh movements. I'll just give you the example of Brazil, which is highly volatile. Mhm. So, usage of WhatsApp uh and now payments through Peaks, which is a direct payment into the accounts, Yeah. has skyrocketed. And, you know, because of all the changes in economic challenges, people have had to adapt very quickly. So, I My sense is that emerging markets are faster in adoption rates. Mhm. Um and once you put these people into the market, it's kind of a no way back.”
Developing the prompt that powers the business advisor required 3 months of iterative testing with entrepreneurs and local experts (including RMI lab member Brian from Kenya) to refine tone, vocabulary (removing Harvard jargon), and contextual appropriateness. The prompt works by instructing the AI to 'use simple, non-technical English' and 'avoid discussing health, religion, politics, or current events.' This demonstrates that effective AI tool design for emerging markets is fundamentally a writing/humanistic problem, not a coding problem.
“There's a very structured way it responds... 'Use simple, non-technical English. Uh my favorite one here is as a business mentor do not discuss health, religion, politics, or current events.' This prompt took us 3 months to develop.”
There is potential for the AI tool to improve significantly beyond its current V1 version by adding contextual data: GPS location information to provide regionally tailored advice, integration with Shopify or other platforms to provide store-specific data (products, sales numbers), or integration with health records and doctor data to provide medically contextualized advice. However, this raises inequality concerns because data availability varies dramatically between developed and developing markets.
“You could now start updating this with someone's GPS location. Right... I can pull down their entire Shopify store. So I can know you're a Shopify store based in South Africa, that you sell this type of clothes, what your sales are. I can start putting that into the AI... that's when I start getting a little bit like dystopian and I'm like, oh no, none of us are going to have jobs anymore. Cuz the AI starts coming up with amazing product recommendations, amazing sales strategies as you put in more of this contextual data.”
A fifth open question is whether effects of AI mediated through humans (AI-to-human-to-human loops) can persist. Rem gives the example: teachers use AI to improve their teaching, which then improves student outcomes. Can impact chain through this human intermediary, and does it compound or diminish?
“Can AI to human to human loops drive impact? So that teaching example, there's an AI that's helping a teacher and then that teacher's helping students. Can the effect persist that long? And in emerging markets, that's going to be really, really important.”
The speaker hypothesizes that financial constraints may be a binding constraint for low-performing entrepreneurs—even if they know what strategic actions to take, if they lack capital to implement them, the information is useless, suggesting the need to pair information/advice tools with financing mechanisms.
“The first one is are financial resources necessary for someone to benefit from AI? Right, information matters. Small information changes can have huge impacts, but if you don't have any money, if you can't get a loan, right, if you can't get financing, how useful is really knowing what you should do? And I think that's a big open question that we need to understand. And are there ways to pair these two together so that we can do better lending?”
The divergent effects between high and low performers stem from applying AI to fundamentally different problem types: low performers focus on problems of competition and survival (e.g., 'my business is facing drastic change and incurring losses'), while high performers focus on problems of growth (e.g., 'I have capital to expand to a new location'). The AI provides useful tactical advice for growth problems but provides unhelpful generic advice (like 'go get financing') for survival problems, causing low performers to become discouraged and make fewer business changes than the control group.
“the low performers focus on problems of competition. So, you have things like my hotel business is facing drastic change and is now incurring losses no matter what changes I try to implement. How can I bring my business back to thriving? That's hard... On the flip side, the high performers are coming in flush with cash. The high performers are focusing on problems of growth. So, they're asking things like, 'I already have 20,000 to expand my business which I would specialize in selling foodstuff. How can I think about expanding to a new location?'”
In a randomized controlled trial of 640 Kenyan entrepreneurs (311 control, 306 treatment), providing access to a GPT-4-powered WhatsApp business advisor had no statistically significant effect on revenue and profit in the full sample (approximately 5% increase but within error bars), contrary to expectations.
“what you're going to see here, this is a treatment effects, this is in standard deviations. You can think of it as roughly percentage increases in revenue and profits. Right... what we find, a little unexpected, which is that in the full sample of entrepreneurs, there's no effect. These are statistical error bars. So, what this is telling you is there's a 5% increase in performance, but that statistically could be zero.”
Writing ability is the critical skill for prompt engineering and generative AI customization, not traditional computer programming, making humanists and writers more valuable than computer scientists for this work.
“You're programming through writing. Humanists are going to win the day. Right, English majors are going to have their return and make more money than the computer scientists...writing this and getting the tone right and getting it to respond in the right way took 3 months. We've iterated on it since, but it's all writing. There's no computer programming.”
AI text generation has disruptive potential that is not necessarily good. While most users have good intentions and outcomes, a meaningful portion are bad actors (scammers) who use large language models to manipulate people, steal money, create deepfakes, or interfere with politics.
“I would argue that probably the vast majority of users in emerging markets are good. They have good intentions, and the outcomes are good. But, there is a meaningful portion who have bad intention, scammers who use large language models to manipulate people to steal large sums of money. Yeah. And they do so pretty effectively because they don't have to learn new languages. They just have to put in the input, and they're able to manipulate people thousands of miles away. And then you have deep fakes and politics is something that Yeah. So, that's what I was makes me nervous.”
Many entrepreneurs in the study do not have alternative sources of trusted business advice. When surveyed, entrepreneurs say they either live in suburban Nairobi or rural areas, don't trust people around them, or feel embarrassed to ask questions. This explains willingness to use an AI advisor: it provides privacy and removes social risk from asking what they perceive as naive questions.
“when we survey them, say they don't have people they can go to. They either live in a suburb of Nairobi or somewhere more rural. They don't trust the people that that are around them. Maybe they feel embarrassed. One of the things we're seeing a lot with this Gen AI stuff is people ask questions where they're really embarrassed. Right. Um and so, you know, maybe this is just sort of privacy. Okay, I can trust this to a robot.”
The study recruited 640 entrepreneurs on average 26 years old, running a mix of businesses (cyber shops, fast food joints, poultry farms) across Kenya. The diversity of business types is important because the AI advisor can provide domain expertise across multiple sectors (food, agriculture, IT, transportation), whereas a human consultant would typically specialize in one domain.
“we recruited them through Facebook... on average 26 years old. They run a mix of cyber shops, fast food joints, poultry farms all across Kenya. Huge heterogeneity of these entrepreneurs as well. Right?... I mentioned this cuz I think this is really important and one of the powerful from aspects of these tools, which is that I am sure I could find an expert in food in Nairobi who could help consult these shops. But, if I also wanted an expert on agriculture, clothing, information technology, and transportation, that's hard to find. These AI tools know all of that.”
Daniel Jordan at Columbia University is using LLMs to help teachers teach better rather than directly training students. The approach avoids giving young children language models, instead improving teacher effectiveness so they can better teach students. This is scaling expertise through human intermediaries.
“Finally, on schooling, this is a a cool paper by Daniel Jordan who's at Columbia University. They're using LLMs to help teachers teach better. So you're not going to give a six-year-old a I give my six-year-old a phone when I need to do something else and they play on your phone. Right, but you're not going to give them an LLM. Right, so if you want to help improve early childhood education, right, what you're going to want to do is train the teachers to do a better job training the students.”
Emerging markets could experience 'creative destruction' as AI increases firm productivity. Some firms will close, creating inequality in the short term. However, the trade-off question is whether productivity benefits outweigh these costs, and whether displaced workers can be rehired elsewhere.
“on the one hand, right? I do a lot of work in emerging markets. One of the things you see is this missing middle. There's a lot of missing mid-size firms. You have lots and lots of really small firms. If AI is going to drive productivity of firms, we are going to see creative destruction. You are going to see some inequality happen. Now, the question is is the trade-off worth it for the productivity benefits? Can we find those people whose firms closed other jobs?”
The speaker implemented peer-to-peer feedback sessions in his HBS entrepreneurship courses where students teach each other to give and receive advice, finding it 'incredibly powerful' for learning, and scaled this model to hundreds of entrepreneurs in India and Togo, but found it impossible to scale beyond a few hundred people while maintaining quality.
“I teach entrepreneurship here at Harvard Business School. I do something that's a little sacrilegious. Instead of doing all case discussions, about a quarter of my sessions are peer-to-peer feedback sessions. I teach the students how to give and get advice from one another. Incredibly important for helping them learn. The problem is I can do that at HBS with 40, with 80, with 100 students. I can do it in India with a couple hundred entrepreneurs. I can do it in Togo with 300 entrepreneurs. But if you want me to go do that with a thousand, two thousand entrepreneurs, I can't.”
In Peru, organizations are rolling out generative AI to provide pregnancy health guidance via conversational interface, including advice on when to see a doctor and trust-building, with extensive piloting and randomized controlled trial testing to ensure efficacy before scaling, acknowledging that providing bad medical advice to expectant mothers is a severe failure mode.
“they're rolling out generative AI for basically anyone who's pregnant in Peru. And it's going to give you advice. It's going to talk to you. When do you go to the doctor? Build trust. They've done a ton of piloting. They're running a big randomized control trial to test if this is actually effective. Cuz the worst thing you could imagine is we've got some AI giving terrible advice Right. to expecting moms.”
The question of whether a WhatsApp interface is 'almost always good enough' for emerging market AI applications is a key frontier. Rem suggests: 'I think the place right now to start is basically can I just run it over WhatsApp? I don't need to do any fancy front end engineering. I don't need to do any design. Just is that enough to be able to drive the impact that we want to see?'
“Is a WhatsApp interface almost always good enough? Right, like can we just kind of do everything over text message? I want to leave that with you because I think as you're developing your own AI stuff and thinking about where it can have impact, I think the place right now to start is basically can I just run it over WhatsApp? I don't need to do any fancy front end engineering. I don't need to do any design. Just is that enough to be able to drive the impact that we want to see?”
Sarvam AI, based in India, is building generative AI tools for all 22 scheduled languages of India. Rem predicts: 'In two years, for any language with over 500,000 speakers, we're going to have basically perfect generative AI for all those languages.' Recent cost reductions (10x drop for Japanese) suggest private market incentives will solve many language problems.
“This is Sarvam AI based out of India. Uh we talked a little bit about language here. I think this is really exciting. They are basically building for the 22 scheduled languages in India AI tools to allow you to converse in any of them. I'm going to put my money down in two years for any language with over 500,000 speakers, we're going to have basically perfect generative AI for all those languages.”
Mino Health.ai, funded by the Gates Foundation, is building foundation models for health on the African continent, starting with radiology and expanding to help doctors make better decisions and take better medical notes. This represents applying AI to healthcare in emerging markets with adaptation to local context.
“This is a phenomenal example. Mino Health.ai funded by the Gates Foundation. They want to build foundation models for health on the African continent. Right, so they're starting with radiology, but they're doing a bunch of work as well around how do we help doctors make better decisions, take better medical notes, right? And really adapting it to that setting.”
Data drift has occurred since the AI model's training: prices for Friesian dairy cows have changed following Kenya's currency devaluation, meaning the advice given reflects outdated price information. This is a technical limitation of static models applied to dynamic emerging market contexts.
“The prices for these cows are a little bit high, but before recent devaluation, pretty good prices, right? So, when the model was trained, it had pretty good data on that. Since then, there's been drift.”
In the study, entrepreneurs asked high-quality, strategic business questions of the sort that would place in the top decile at Harvard Business School, such as: 'I have a fast food joint adjacent to a matatu terminus with stiff competition—how can I overcome this?' and 'I wash cars and motorbikes alone and find it tiresome; how do I hire staff and monitor them to maximize profit?' These questions demonstrate that entrepreneurs in emerging markets have genuine, sophisticated business challenges, not trivial ones.
“can you calculate for me an estimate budget for four Friesian dairy cows and half piece of land and possible risk that I might encounter. The second one, I have a fast food joint that I run adjacent to a matatu terminus. There are several such food joints in that particular place, and it's fair to say competition is stiff. How can I overcome the competitive business environment and be ahead of the rest of the pack? I love this one. I've taught strategy at HBS for many years. This is like a strategy final exam question.”
Entrepreneurs in the treatment group showed high engagement with the AI advisor over WhatsApp: some explicitly requested more reminder text messages because they forgot the tool was available ('Could you send me more texts cuz I forget to talk to the AI? It gets pushed down on my WhatsApp'), and approximately a dozen users sent 'happy new year' messages to their chatbot at the turn of 2024 ('It has been amazing working with you').
“We hit 2024. Lovely happy New Year everybody, this is great. We have about a dozen of the users of this say happy New Year to their chatbot. And they're like, 'It has been amazing working with you.'”
Grab, a Singapore-based platform serving Southeast Asia, operates with hundreds of thousands of small restaurants using its delivery service. The company is excited about scaling the business advisor tool to help merchants improve profiles, optimize menus, and improve businesses—providing to long-tail merchants (too small for account managers) the kind of expert support previously available only to key accounts.
“We're actually working with Grab based out of Singapore which serves all of Southeast Asia basically. Um they work with hundreds of thousands of small restaurants. Right, their delivery you know, you go on, you order your food like DoorDash or Grubhub or whatever it is. Hundreds of thousands of small restaurants. They're really excited about the tool we built. So we've been working with them to basically say, could you build GenAI to help these merchants improve their profiles, improve their business, optimize their menu, get the advice that some merchants have right now because they have a key account manager. But most of these businesses are too small. They're in the long tail.”
All participants in the study were informed that they were talking to an AI, provided consent to participate in the experiment, and knew data would be collected for research purposes, with careful ethical protocols outlined in a three-page ethics appendix of the paper.
“They all know they're talking to an AI. No one's being deceived. They all consent to be part of the experiment, to know that we're going to be collecting the data for what we do.”
The study did not find meaningful demographic differences between low and high performers across industries, gender, or age. The difference between groups was not demographic but behavioral and problem-focused.
“across industries basically the same, across genders basically the same, across age basically the same. So it looks very similar.”
There are 'way more' HBS students unable to write effectively (evidenced by nearly a thousand MBA essays the speaker has graded over his career), suggesting that writing quality is a significant limitation constraining who can benefit from AI writing tools, not just a technical problem.
“Writing's not that easy, right? It's actually quite hard as I evidence from my nearly a thousand MBA students I've taught in my career. I've had to grade every one of those essays. And oof. Words for you.”
The AI advisor provided phenomenally good answers to entrepreneur questions, including detailed responses on dairy cow purchase budgets (covering costs, veterinary care, land maintenance, and risks from disease and price volatility) that Brian from Kenya verified as accurate for current prices, and strategic advice on competition that includes Porter's Five Forces-style differentiation analysis with specific tactical options (loyalty programs, community involvement, menu differentiation).
“when I look at the uh answer, it's actually phenomenally good. Brian Arra uh uh uh went out and checked uh in Kenya. We shot to some folks. The prices for these cows are a little bit high, but before recent devaluation, pretty good prices, right?... It talks about the cost of the cow, feeding the cow, veterinary care, land prep and maintenance, outlines risks from disease to feed price fluctuations, market price volatility, weather conditions. It's a pretty exhaustive list of what you'd want if you were thinking about buying some cows for your business, right?”
The speaker predicts within 10 years, every doctor everywhere could have access to an AI coach helping them make better decisions, and this doesn't seem 'out of the realm of possibility' given current technological trajectories.
“I think the idea that every doctor everywhere in the world has a coach that can help them make better decisions, right? In 10 years doesn't seem out of the realm of possibility.”
Large language models like ChatGPT, despite their sophistication, are fundamentally just text generation systems—a word document on steroids—that have captured global attention disproportionate to their technical novelty.
“It's a text box. Right? I got some friends who worked on designing this thing. I think it's a pretty design, but it's a text box. It's like a Word document on steroids. Right? And it says, 'How can I help you today?' and then you type things in and it returns text. That's all this machine does.”
The speaker expects that with continued refinement of the prompt and more data over another year of work, he can improve the negative effects observed for low-performing entrepreneurs and 'get that up' (improve their treatment effect from negative toward positive).
“I don't think that negative is fixed. We did this within less than a year. Give me another year to refine that prompt and get more data on these people. I think we can get that get that get that up.”
Babangona, a company started by HBS alum Kola and his wife Masha, works with over 100,000 farmers in northeastern Nigeria (a region where Boko Haram operates) and wants to scale to 10 million farmers. Many of these farmers are illiterate, which raises questions about text-based AI utility in this context.
“here are three photos um uh from different markets where we've been working on in the lab. So you're seeing all the way on the left a guy with a phone, uh works for a company called Babangona, actually started by uh HBS alum, uh Kola and uh his wife Masha. They started Babangona. They work with over 100,000 farmers in northeastern Nigeria. This is where Boko Haram operates. 100,000 farmers. They want to scale to 10 million farmers. Right, many of these farmers are illiterate, Which opens up questions about whether this text is useful.”
One example of a creative/specific piece of advice the AI gave was suggesting a free candy giveaway at a food store next to a school to attract children, which the entrepreneur reported working well. This demonstrates the AI can generate location-specific and contextually-appropriate business tactics when provided with context.
“My favorite one was there was a guy who uh mentioned that his store was next to a school. And so, he said, 'Oh, I have a strategy for you. Give away a free piece of candy.' Right? Cuz then all the kids after school are going to come to your store. And then you'll be able to drive sales cuz they'll love the candy. And that seemed to work quite well for this guy. Now, whether that's ethical or not, that's a separate conversation. But brilliant marketing strategy, like nowhere.”
The study's control group received a PDF 'course note' similar to what HBS distributes to students. This is an appropriate placebo because empirically, when HBS distributes course notes, students do not read them—they read cases (because they know they will be cold-called on cases), and they ignore module notes and PDFs.
“311 in a control group got the equivalent of a PDF like note you would distribute to HBS students, or like a course note. And I can tell you for teaching here for eight years, you distribute a course note to the students, nobody reads it. They'll read the cases, they know they're going to get cold called. Course note, module note, nobody reads PDFs to a first approximation.”
The 10% decrease in revenue/profits for low-performing entrepreneurs who received AI advice (vs. control) was surprising and unexpected. Rem states: 'all that work and nothing?' and describes the team hitting 'the accelerator' in May 2023 and working nonstop until November before discovering this negative result.
“Cuz you know, we worked on this, we hit the accelerator in May of 2023 and worked on this nonstop till November. I'm like, 'Oh my gosh, all that work and nothing?' Yeah, the way we're in this, we're over here. No one just added, right, like maybe it's just useless. Right, like all the examples, if you go to the the the the Open AI...”
There is a 'peak bubble' in AI investment and attention. Rem received an email offering $20 million for generative AI research funding, and jokes that it might be from a scammer. He states: 'Definitely a peak bubble. Like let me just say that.'
“I got a an email last night that says I've got $20 million to spend. I'd love to invest it in the research we're doing on generative AI. Now, I hope that is real. Maybe it's a scammer. I don't know. Definitely a peak bubble. Like let me just say that.”