Ethan Mollick et al on AI’s Jagged Frontiers

The public release of Large Language Models (LLMs) has sparked tremendous interest in how humans will use Artificial Intelligence (AI) to accomplish a variety of tasks. In our study conducted with Boston Consulting Group, a global management consulting firm, we examine the performance implications of AI on realistic, complex, and knowledge-intensive tasks. The pre-registered experiment involved 758 consultants comprising about 7% of the individual contributor-level consultants at the company. After establishing a performance baseline on a similar task, subjects were randomly assigned to one of three conditions: no AI access, GPT-4 AI access, or GPT-4 AI access with a prompt engineering overview. We suggest that the capabilities of AI create a “jagged technological frontier” where some tasks are easily done by AI, while others, though seemingly similar in difficulty level, are outside the current capability of AI. For each one of a set of 18 realistic consulting tasks within the frontier of AI capabilities, consultants using AI were significantly more productive (they completed 12.2% more tasks on average, and completed tasks 25.1% more quickly), and produced significantly higher quality results (more than 40% higher quality compared to a control group). Consultants across the skills distribution benefited significantly from having AI augmentation, with those below the average performance threshold increasing by 43% and those above increasing by 17% compared to their own scores. For a task selected to be outside the frontier, however, consultants using AI were 19 percentage points less likely to produce correct solutions compared to those without AI. Further, our analysis shows the emergence of two distinctive patterns of successful AI use by humans along a spectrum of human- AI integration. One set of consultants acted as “Centaurs,” like the mythical half- horse/half-human creature, dividing and delegating their solution-creation activities to the AI or to themselves. Another set of consultants acted more like “Cyborgs,” completely integrating their task flow with the AI and continually interacting with the technology.

That’s the abstract of a paper written by a team of academicians based in the United States, of whom Prof. Ethan Mollick is one. The idea behind the paper is very simple: can we quantify just how much of an improvement in productivity is made possible because of using AI?

And the TL;DR is that productivity is way up. From the abstract: “consultants using AI were significantly more productive (they completed 12.2% more tasks on average, and completed tasks 25.1% more quickly), and produced significantly higher quality results (more than 40% higher quality compared to a control group)”

Some points of especial interest from my perspective:

  1. The advantages of AI are substantial, but unclear. We don’t know which tasks will be completed more efficiently (and better) by using AI, and which won’t. Worse, nobody knows for sure. It is very much a trial-and-error thing. (pg 3)
  2. This is a dynamic problem. How our interaction with AI changes, how the nature of our tasks change, and how AI gets better – all of these will vary with time. This paper will be outdated within a matter of weeks, not days – but that is a feature, not a bug. (pg 4)
  3. What was the task itself? Note that there were two different experiments, and within each experiment, there were two tasks. The first experiment was “within the frontier”, which means an experiment that was thought to be well within GPT-4’s capabilities. For each experiment, participants were “benchmarked” using an assessment task, and were then asked to work on an “experimental” task. I will always be referring to the “experimental” task:
    “In this experimental task, participants were tasked with conceptualizing a footwear idea for niche markets and delineating every step involved, from prototype description to market segmentation to entering the market. An executive from a leading global footwear company verified that the task design covered the entire process their company typically goes through, from ideation to product launch.5 Participants responded to a total of 18 tasks (or as many as they could within the given time frame). These tasks spanned various domains. Specifically, they can be categorized into four types: creativity (e.g., “Propose at least 10 ideas for a new shoe targeting an underserved market or sport.”), analytical thinking (e.g., “Segment the footwear industry market based on users.”), writing proficiency (e.g., “Draft a press release marketing copy for your product.”), and persuasiveness (e.g., “Pen an inspirational memo to employees detailing why your product would outshine competitors.”). This allowed us to collect comprehensive assessments of quality.” (pg 8 and pg 9)
  4. This was especially impressive:
    “Our results reveal significant effects, underscoring the prowess of AI even in tasks traditionally executed by highly skilled and well-compensated professionals. Not only did the use of AI lead to an increase in the number of subtasks completed by an average of 12.5%, but it also enhanced the quality of the responses by an average of more than 40%. These effects support the view that for tasks that are clearly within its frontier of capabilities, even those that historically demanded intensive human interaction, AI support provides huge performance benefits.” (pg 12)
  5. The “outside the frontier” task:
    “Participants used insights from interviews and financial data to provide recommendations for the CEO. Their recommendations were to pinpoint which brand held the most potential for growth. Additionally, participants were also expected to suggest actions to improve the chosen brand, regardless of the exact brand they had chosen” (pg 13)
  6. Even in the case of these tasks, there was improvement across the board in terms of lesser time spent, and also in terms of improvement of quality in output (pg 14 and 15)
  7. The authors found that there were two dominant approaches:
    “The first is Centaur behavior. Named after the mythical creature that is half-human and half-horse, this approach involves a similar strategic division of labor between humans and machines closely fused together.12 Users with this strategy switch between AI and human tasks, allocating responsibilities based on the strengths and capabilities of each entity. They discern which tasks are best suited for human intervention and which can be efficiently managed by AI.
    The second model we observed is Cyborg behavior. Named after hybrid human- machine beings as envisioned in science fiction literature, this approach is about intricate integration. Cyborg users don’t just delegate tasks; they intertwine their efforts with AI at the very frontier of capabilities. This strategy might manifest as alternating responsibilities at the subtask level, such as initiating a sentence for the AI to complete or working in tandem with the AI.” (pg 16)
  8. And finally, their concluding paragraph:
    “Finally, we note that our findings offer multiple avenues for interpretation when considering the future implications of human/AI collaboration. Firstly, our results lend support to the optimism about AI capabilities for important high-end knowledge work tasks such as fast idea generation, writing, persuasion, strategic analysis, and creative product innovation. In our study, since AI proved surprisingly capable, it was difficult to design a task in this experiment outside the AI’s frontier where humans with high human capital doing their job would consistently outperform AI. However, navigating AI’s jagged capabilities frontier remains challenging. Even for experienced professionals engaged in tasks akin to some of their daily responsibilities, this demarcation is not always evident. As the boundaries of AI capabilities continue to expand, often exponentially, it becomes incumbent upon human professionals to recalibrate their understanding of the frontier and for organizations to prepare for a new world of work combining humans and AI. Overall, AI seems poised to significantly impact human cognition and problem-solving ability. Similarly to how the internet and web browsers dramatically reduced the marginal cost of information sharing, AI may also be lowering the costs associated with human thinking and reasoning, with potentially broad and transformative effects”

This chart tells uite the story: (pg 28)

The appendix (pg 44 onwards) details the tasks, if you would like to go through them.

Finally, a part of the abstract that I’m still thinking about:

“Consultants across the skills distribution benefited significantly from having AI augmentation, with those below the average performance threshold increasing by 43% and those above increasing by 17% compared to their own scores. For a task selected to be outside the frontier, however, consultants using AI were 19 percentage points less likely to produce correct solutions compared to those without AI”

A lovely, thought-provoking paper. Whatever your own opinions about the impact of AI upon productivity, employment and output, a carefully designed academic study such as this is worth reading, and critiquing.

And if you are currently in college (any college), learn how to get better at working with AI!

On Note-Taking Apps

Google Keep. Microsoft OneNote. Roam. Obsidian. Notion. Readwise.

There are other apps with whom I’ve had, so to speak, even shorter relationships, but the ones above are the ones that I have really and truly tried to use on an extensive basis. Google Keep, as with so many other things Google, is excellent in some ways, but utterly hopeless in others. You’ll never guess what their latest enhancement is, for example. OneNote was very promising, but Microsoft went through a bit of a phase where they had a OneNote app for Windows, and a separate one for Office365, and it just got too confusing for words. Roam was too expensive at 15USD per month, and Obsidian had too steep a learning curve for me. And if you want to talk about steep learning curves, you should try out Notion. Gah.

The latest one that I’m trying out is Readwise, and well, it’s going… ok, I guess. And we all know what that really means, don’t we?

Long story short, none of these have really worked out for me. And that, I suspect, is the case for most of you reading this. There will be some who are true converts and zealots of any one of these, and I envy you. I really do, good for you, really! But whichever one of these you’re selling, I’m not really on the market. And no, that other new new one ain’t for me either, whichever one it is.

And that’s why this article in The Verge really resonated with me:

Note-taking, after all, does not take place in a vacuum. It takes place on your computer, next to email, and Slack, and Discord, and iMessage, and the text-based social network of your choosing. In the era of alt-tabbing between these and other apps, our ability to build knowledge and draw connections is permanently challenged by what might be our ultimately futile efforts to multitask.

As always, do go through the whole thing. It is full of fascinating snippets, including the somewhat surprising, somewhat entirely predictable finding that the average time we spend on a single screen before shifting our attention elsewhere was 2.5 minutes. If that seems too long for you, you’re right. That was in 2004. Today’s stats? 47 seconds.

The author of the article goes on to hope (as do some of us, while others are repulsed by the thought) that AI will help us make sense of all of these links that we have been squirrelling away for years. I’m on Team Maybe about this myself. But I really do agree with this bit:

In short: it is probably a mistake, in the end, to ask software to improve our thinking. Even if you can rescue your attention from the acid bath of the internet; even if you can gather the most interesting data and observations into the app of your choosing; even if you revisit that data from time to time — this will not be enough. It might not even be worth trying.
The reason, sadly, is that thinking takes place in your brain. And thinking is an active pursuit — one that often happens when you are spending long stretches of time staring into space, then writing a bit, and then staring into space a bit more. It’s here here that the connections are made and the insights are formed. And it is a process that stubbornly resists automation.

And that is, in a way, comforting and reassuring. I haven’t failed all these awesome note-taking apps, and they haven’t failed me either. In each of these cases, it just wasn’t meant to be.

The article refers to the works of Andy Matuschak (Google him if you don’t know who he is), who says that the ultimate goal is to think effectively (amen!), and that all of us should really be thinking about two questions.

  1. What practices can help me reliably develop insights over time?
  2. How can I shepherd my attention effectively?

Don’t look to me for the answer to the second of these questions, I have no idea. If you know the answer, help a guy out, will ya? But I do have my own personal answer regarding the first of these questions.

I read a lot. Not as much as some others that I know, and I wish I read more, but I do think I read more than the average person. Some of what I read I find interesting enough to talk about with some people whose opinions I truly value. Some of these conversations end up being friendly arguments, where they challenge my view, and I challenge theirs. Then I have a cup of coffee and think about some of these arguments.

Then I write about it. And after I write about it, I send a draft of what I’ve written to them. Then, if I’m really lucky, we have another argument about the draft I’ve sent them. I think more about this second argument, and refine the draft.

How I would like to tell you that this is how every single post on EFE gets written.

The reality is that all of what I’ve described above happens for maybe one post every month. Those posts, and those arguments stay with me then for a very long time. But the vast majority of the posts you read over here are me reading something, finding it interesting enough to write about it, and well, I write it and you read it.

When the whole process described above works the way it should – that is utopia.

But between living in utopia and not writing about it at all lies a happy medium. Happy not because it is perfect, but because it is attainable. It involves at least one of all those things happening – me reading about something and then writing about it.

So my favorite note-taking app?

It happens to be a blog called EFE. The posts over here are me taking notes on something I’ve read – and that, more than anything else, helps me remember stuff better.

I’ve said this before, but I’ll say it again.


Write and put it out in the public domain for all to read. Best way to remember something, anything and most everything.

And if you can figure out a way for me to do achieve my utopian process for all the posts that I write, please do tell!

To work from home or not to work from home?

The Economist says not to work from home because “it is not more productive than being in an office, after all”.

A gradual reverse migration is under way, from Zoom to the conference room. Wall Street firms have been among the most forceful in summoning workers to their offices, but in recent months even many tech titans—Apple, Google, Meta and more—have demanded staff show up to the office at least three days a week. For work-from-home believers, it looks like the revenge of corporate curmudgeons.

They cite the case of the paper that went from showing an eight percent increase in productivity due to working from home when it was a working paper, to showing that there was a four percent reduction instead. There was no problem with the paper or its methodology, to be clear. The difference was simply because of better quality of data. There is a world of other research worth going through in The Economist article, and I would urge you to read it.

What reasons come through for the decline in productivity? Well, it’s just hard to work from a dining table! The ability to go to a co-worker’s desk to chat about work, to get some help, to resolve an issue – that is harder to do online. Why is it harder? Because “teleconferencing is a pale imitation of in-the-flesh meetings”. That’s a fancy way of saying online sucks. To use Coasean terminology, as the article in The Economist does, coordination costs matter.

Most important of all, networking becomes harder. We are, at heart, a social species, and we need proximity to other people. Not only for the psychological benefits, but also because we learn best in person. That might seem like a contradiction given yesterday’s post, but it is not. Learning in person doesn’t necessarily mean listening to someone like me drone on in a classroom!

But to me, the article came alive towards the end. As with all well written articles, this one too segues into an implicit “on-the-other-hand” section.

Perhaps the greatest virtue of remote work is that it leads to happier employees. People spend less time commuting, which from their vantage-point might feel like an increase in productivity, even if conventional measures fail to detect it. They can more easily fit in school pickups and doctor appointments, not to mention the occasional lie-in or midmorning jog. And some tasks—notably, those requiring unbroken concentration for long periods—can often be done more smoothly from home than in open-plan offices. All this explains why so many workers have become so office-shy.

“What are you optimizing for?” remains an underrated question! If you’ll allow me to cite an example from my own life: I’ve been working from home for the last year and a half, approximately. I stay on Baner Road in Pune, and not having to battle University signal everyday is something that saves me time, gives me more energy and enthusiasm to work, and frees up time to do other stuff (work/exercise/napping/whatever). I have the pleasure of picking up my daughter from her bus-stop every day. I get to go for morning and evening walks with my dog. Sure I miss the conversations with some of my colleagues, and god knows I miss being able to interact with my students on campus. But hey, opportunity costs are everywhere, no?

As the article goes on to say in its concluding paragraph, hybrid weeks are here to stay. Sure it is not as productive as working in an office, but woking from home is more soul satisfying. And so the answer to the question “work from home or not” is, well, both.

The truth, and stop me if you’ve heard this one before, lies somewhere in the middle.

Lenny’s Newsletter on Reigniting Duolingo’s User Growth

I’ve written about Duolingo before, and I have no doubt that I will write about it again.

Why? Because I am a very lazy person, and I appreciate all the help I can get when it comes to building good habits. Writing daily on this blog is a good habit – it is, alas, not one that I have perfected yet. Taking my dog out for a walk everyday is a good habit – a necessary one for the dog, and so also for me now. Practicing a language (currently Italian) daily on Duolingo is a good habit, and while I’m not at 1435 days yet, I’m a little more than halfway there, and hey, that’s progress!

Exercising daily is a habit I’ve tried to build and failed at (one day, one day). Eating healthy on a daily basis is a habit I don’t want to build, but eating mostly healthy on a weekly basis is something I’ve more or less succeeded at – and that’s good enough for me. The point is that given that I’m so lazy, building up a habit is the only way to stick at doing something. And anything that helps me enjoy getting into a habit is, to me, a fascinating thing to study.

Which is why I have no doubt I will write about Duolingo again.

Lenny’s newsletter is worth reading in any case, but his latest post is really very, very good. It’s not written by him – this one is a guest post by Jorge Mazal – but that’s all the more reason to read it. If you’re interested in learning about metrics, user retention, driving growth, this article is self-recommending – and that would be a good reason to read it carefully.

But even if you are not interested in any of those things, it still makes sense to read it. My framing of my own incentive while reading it was “Can this article teach me how to gamify my life?”, and from this perspective, it is an eminently readable article.

I had a very interesting conversation with a friend this past Sunday, and his take on habit formation and productivity techniques was that this has perhaps been taken a little bit too far in today’s day and age. I actually agree with him on that point – we try to wring every little bit out of every little hack, to our overall detriment. But that being said, I think it makes sense to take a look at our own lives and ask to what extent we could make our lives a little bit better along dimensions of our choosing. To each one of us goes the right to choose which dimensions, and to each one of us goes the right to choose how to improve our life along those dimensions, and finally, to each one of us goes the right to choose the magnitude of improvements.

But once you’ve answered those questions – which dimensions, how to improve, and to what extent – you could do with help regarding tips and tricks re: Making It Happen. And that’s where this article is worth reading.

My key takeaways:

  1. Gamification matters, and it helps. Try gamifying those aspects of your life that you want to get better at.
  2. A blind CTRL-C CTRL-V of gamification done well elsewhere is a pretty poor way to go about it. Think carefully about which incentives matter to you, and design your gamification strategy accordingly.
  3. These three questions are a good way to frame this:
    Why is this feature working in this product? | Why might this feature succeed or fail in my context? | What changes do I need to make to make this feature succeed for me?
  4. Compounding the benefits of a habit is an excellent, always underrated idea.
  5. We like to win. Set up a competition for yourself, and make sure there are tangible rewards (and punishments!)
  6. Reminders help, but don’t end up irritating yourself out of a habit.
  7. Streaks are a great way to compete with yourself (related to pt. 5), and preserving that which you have is a great motivator (the endowment effect matters)
  8. The social aspect matters – get other people to join you on your journey (hello to all of my friends on Duolingo, and thank you!)

Learning in the Age of AI

How should one think about learning in the age of AI?

That is, if you are a student in a class today, how can you use AI to make your experience of being a student better?

  1. Use AI to create work, but learn how to work with it to make it better: In my experience of having spoken with people about AI, it has been a bit of a bi-modal distribution. There are folks (and I’m very much one of them) who think of ChatGPT as a fantastic tool whose potential to be useful is only going to grow over time. And there are folks who triumphantly assert that AI simply isn’t good enough, citing examples of hallucinations, not-good-enough answers or sub-standard essays. All of these arguments are good arguments against AI, but the last one in particular can be easily overcome by providing better prompts, and by suggesting improvements. “Write a seven paragraph essay on India’s economic reforms of 1991” is a barely acceptable prompt to give it, for example. Mention specific people, events and dates that you might want it to mention in the essay, ask it to revise certain paragraphs in the essay, ask it to write “like” a certain person, mention the conclusion you would like it to reach – spend time with it to make it better.
    All of my suggestions – and this is important! – require the student to know enough about the topic to be able to make these suggestions. You need to think about the prompt, you need to critically evaluate the first-pass answer, and you need to know enough to suggest suitable improvements. AI can take away the drudgery associated with polishing an essay, but it will still (so far) require you to know what you’re talking about. A student’s life is much more interesting today, rather than easier.
  2. Ask it to teach you stuff you didn’t understand: Small class sizes aren’t really a feature of most Indian colleges, in my experience. The idea that you will have five to ten students in class, and will therefore be able to have meaningful, extensive discussions about your doubts in class is a far fetched one in most Indian colleges. So treat AI as a very helpful research assistant who will be able to explain to you your doubts about a particular topic. This can very quickly become too addictive a practice, because the AI will be able to carry out a much more detailed conversation about literally any topic you can think of than most (all?) of your peers. Converse with humans about your conversations with AI, and figure out a ratio that works for you. But corner solutions (of both kinds) are almost certainly sub-optimal.
  3. Check it’s “facts”: You will run into trouble if you accept it’s output as the gospel truth. It asserts facts that simply don’t exist, it will cite papers that it has made up on the spot and it will confidently tell you about books that were never written by people who’ve never existed. It is not about to replace search engines – in fact, search engines have become more useful since the launch of ChatGPT, not less.
  4. Use specialized AI tools: Of which there are hundreds, if not thousands. You can use AI to cite papers (, to design presentations (, create simple animations (look it up) and so much more besides. Don’t restrict yourself to any one tool, and learn how to get better at improving all aspects of your workflow.
  5. Document your work with AI, and make it public: Create a very public repository of work that you have created with AI, and share how you’ve become better at working with AI. Your career depends on your ability to do this, and on your ability to teach other people to do this – so the more the evidence regarding this is in your favor, the stronger your argument for your own career. Begin early, and don’t be shy about showing the world what you’ve done, and how good a worker you are with AI by your side.

Duolingo, Gamification and Habit Formation

I got “promoted” on Duolingo recently, and today’s blogpost is about how weirdly happy I feel about it.

I am an incredibly lazy person. We all are to varying degrees, I suppose, but I’m convinced that I do putting off and procrastrination better than most. There are a few things that I do with enthusiasm and something approaching regularity (writing here being one of them) but with most things in life, tomorrow is a better day for me than today.

There is a very short list of things I am compulsively addicted to doing on a daily basis, There’s Wordle, for example. Reading blogs, for another. The NYT mini crossword, and some other stuff. But there is a clear winner on this list: Duolingo.

On Duolingo I have a 753 day streak, and counting. That is, I have practised on the Duolingo app for 753 days and counting. And it’s not because I am awesome at showing up regularly – it is because Duolingo has incentivized me to show up regularly, and here’s how they do it.

First, peer pressure. Duolingo allows you to follow people in your contact list who are also on Duolingo, and it’s a two way street. That is, they can follow you too:

AshishKulk is my user ID on Duolingo, and please feel free to “add” me to your network if you are so inclined. I can always do with more peer pressure! Knowing that my friends are practising more than I am is a great incentive to try and keep up – which, of course, is what peer pressure is all about.

The score-keeping mechanism in Duolingo is XP. The app tells you how much XP you “acquired” on a daily, weekly, monthly and all-time basis.

Each of these frequencies is gamed differently. For the all time streak, you can check where you rank among your friends (I’m third, if you’re wondering). For the monthly score, Duolingo hands out “badges” if you earn a certain number of XP in a month:

For weekly streaks, you get promoted to different “leagues” based on how many points you score on a weekly basis. It is a double edged sword: you also get “demoted” if you don’t practise enough in a week. The leagues start and finish every Sunday afternoon India time, and I’m typing this out on a Sunday morning – wish me luck!

And the daily basis is perhaps the best gamification of them all, because you end up in a contest with yourself. How long can you keep your daily streak going? Like I said, mine is at 750 odd days, and I got “promoted” for it:

There are also daily quests, friend quests, stories involving characters that build a more personalized, relatable learning experience, and recently, learning paths that use spaced repetition to make sure that weaker concepts are revised and firmed up over time. You can “buy” streak freezes to “protect” your streak in the Duolingo store. These artefacts can be “purchased” using “gems”, and that’s yet another gamificaiton story.

Long story short, the Duolingo team goes out of its way to try and keep you hooked on to learning, and I’m here to tell you that it has definitely worked in my case.

And that’s the point of this post, really. I think Duolingo to be an exemplar when it comes to gamification, but the meta-point here is an obvious one: how do we all go about gamifying activities in our life that we wish would turn into habits? How about a Duolingo experience for exercise? Meditation? Learning cooking? Financial habits?

There are those among us who can build out habits in their lives without gamification, of course, and I envy them for it. But for those of us who are paashas of procrastination, such a tool can help us get better at showing up more regularly.

I speak from personal experience when I say this, though: Duolingo has done it better, and been more effective, than any other habit forming app that I have tried.

Now, excuse me while I go and try to stay in the Diamond League!

Quick Thoughts on Google Chat

I’ve been a fan of Google ever since I saw for myself how much better the search engine (how quaint, no?) was compared to the alternatives, and I’m old enough to remember what a revelation 1GB of storage was for inboxes. Chrome in 2008 was a game changer, I’m an unabashed Android fan, and I spend more than half my life in Google Drive.

I’ll never, ever, ever forgive them for their cold blooded murder of Google Reader, but let’s not get hung up on that for now. Feedly is here and it works just fine.

But what was a hobby (learning more about how cool Google can be) suddenly became an utter necessity when the pandemic took over our lives last year. Working remotely has been a challenge for all of us, and utilizing all of Google’s features was no longer a luxury, but a necessity.

Figuring out how to get your colleagues (and in my case, our students) to learn how to use all of Google’s features has been both a challenge and a pleasure, and most of us at the Gokhale Institute are now fairly comfortable with the following tools/apps: GMail, Google Calendar, Google Classroom and Google Drive.

What especially helped was their decision to launch the sidebar on the right, in GMail, that allowed for most (but not all!) of these tools to be accessible from within just the one tab.

One feature in particular that we’ve made fairly heavy use of has been a separate tab for Google Chat (go to Most of us know Google Chat as that little box on the left in our GMail tabs, but the separate stand-alone tab is much better. You could have chat rooms (about which more in a bit). But most importantly, a separate tab made more sense because visually, chatting was easier in a separate tab rather than those little pop-up windows in GMail.

That apart, the ability to use “bots”, such as Polly for conducting polls and the Meeting bot for setting up meetings1 has been really helpful this past year.

But yesterday, they announced some serious updates to all of these features. Dieter Bohn has a quick explainer at The Verge, but as is usual with Google, the full feature set will be “coming soon”. But here are my quick reflections on whatever it is that we’re able to to do right now. Note that I work in a university, not a conventional office. YMMV, as they say:

  1. Starting projects with colleagues/students is much better in a chat room in Google Chat than via email. The discussion happens much more quickly, responses are searchable, and threaded discussions make it much more convenient.
  2. There are three tabs available up top in all chatrooms: the actual chat itself, files and tasks. Files shared in the chat room are now available to see at any point of time, and now they even open up right there, in the chat window. Much more convenient. Note that seeing comments etc requires the document to be opened up in a separate window/tab. Tasks is basically Google Tasks (a tool which almost nobody uses), but assigned to work for the group that is in that particular chat room. Tasks, used as a group, is much better than Tasks in GMail. A richer feature set here would be awesome, but that’s another blogpost by itself.
  3. Add in the Polly and Meeting bots to your chat rooms (and please let me know if you know of other good bots to deploy)
  4. Stuff I wish they’d add: the ability to pick a message and reply specifically to it (as in Whatsapp) is sorely missed. Conversations would be so much more streamlined if this was around.
  5. Chatrooms are searchable by person and by date, among other things. The trouble is that most people won’t know that this is possible, and Chat doesn’t (yet) have the drop-down menu in search like GMail. Most folks don’t know about the drop-down menu in GMail search, but that’s another story.
  6. Google Chat now has the same bar to the right that GMail does: Calendar, Tasks and Keep show up over there. Education specific request: throw in Classroom there too?
  7. While we’re at it, why can’t all Classrooms automatically have chat rooms created? Why can’t files shared on Classroom automatically sync with this chat room? Why can’t assignments given in Google Classroom automatically sync as tasks in these chatrooms? This would help so much!
  8. Setting up a calendar appointment, or starting a Google Meet call is possible from within the little box you use to type messages in Google Chat. When you set up a calendar invite, it automatically invites all participants in that chat, which is great.
  9. My own personal workflow involves Feedly, Roam, GChat, GDocs, GDrive, GCal. Hopefully, API’s will allow one to add in Roam and Feedly on to the sidebar in the near future. If that becomes possible, I’m happy to live entirely inside Google Chat when I’m working, with minor excursions into the Twitter tab every now and then. From a purely selfish perspective, maybe Google can buy out Feedly and Roam (hint, hint)? Keep as a note-taking tool just isn’t good enough!
  10. Finally, any educational institute anywhere: if you need help learning about this, or setting it up, or just a call where you want to see how we use these tools at the Gokhale Institute, I’m just a shout away. Happy to help, any time 🙂
  1. it is a life changer once you get the hang of it, trust me. It uses NLP, and you can type stuff like “set up a meeting with xyz at ten am tomorrow morning” and it does the rest. Yes, really. It is an old feature, used to be available in Google Calendar years ago, but is now sadly missing from there[]

Correlation, Causation, Coffee…

… and so much else besides!

Alexey Guzey’s newsletter is a treasure trove of interesting things he finds on Twitter, and in Guzey’s case, interesting is an understatement.

But even by his high standards, the article I am sharing with you today is something else altogether.

Said article begins the same way most articles I have shared here:

“The break point in America is exactly 1973,” says economist Tyler Cowen, “and we don’t know why this is the case.” One possible culprit is the 1973 oil embargo, because many of these trends have to do with energy. But Cowen doesn’t think this holds water. “Since that time, the price of oil in real terms has fallen a great deal,” he says, “and productivity has not bounded back.”
Another possible culprit is the US going off the gold standard in 1971, part of the set of measures known as the Nixon shock (also the name of our new Heavy Metal band). This makes some sense because many of these trends have to do with the economy. But it’s not clear if this is a good explanation either, as many of these trends seem to be global, and most of the world is not on the US dollar.

But it then takes on a life of its own. And if this excerpt doesn’t make you curious to read more, nothing ever will.

Bier of course was a surgeon, and so when it was his turn to give Hildebrandt the injection, he performed it flawlessly. Soon Hildebrandt was very anaesthetized. To test it, reports Regional Anaesthesia, “Bier pinched Hildebrandt with his fingernails, hit his legs with a hammer, stubbed out a burning cigar on him, pulled out his pubic hair, and then firmly squeezed his testicles,” all to no effect. In a different account, this last step was described as “strong pressure and traction to the testicles”. They also pushed a large needle “in down to the thighbone without causing the slightest pain”, and tried “strong pinching of the nipples”, which could hardly be felt. They were thrilled. With apparently no bad blood over this series of trials, the two gentlemen celebrated that evening with wine and cigars, and woke up the next morning with the world’s biggest pair of headaches, which confined them to bed for 4 and 9 days, respectively.

The whole article is impossibly fascinating, and is peppered with Today I Learnt moments. Along with the surgeon above, Tesla (as in the scientist, not the firm), Robert Louis Stevenson, Freud, and the Beatles also make guest appearances – as do two Popes.

Please, do read.

On X-Inefficiency

Yesterday, I wrote this in my summary of Bloom and co-authors’ paper on productivity in India:

Economists tend to not buy into this because they assume that profit maximization implies cost minimization
So in other words, if firms are not minimizing costs by adopting good management practices, it is because “wages are so low that repairing defects is cheap. Hence, their management practices are not bad, but the optimal response to low wages.”

… which brought to mind of the topic of X-inefficiency, for the second time this year. The first was when Tyler Cowen wrote about it in January. Here’s Wikipedia:

X-inefficiency is the divergence of a firm’s observed behavior in practice, influenced by a lack of competitive pressure, from efficient behavior assumed or implied by economic theory. The concept of X-inefficiency was introduced by Harvey Leibenstein

X-inefficiency, in essence, is the idea that the economic theory idea about efficient firms in efficient markets is perhaps a little overblown. Here’s a quote from the paper itself:

The simple fact is that neither individuals nor firms work as hard, nor do they search for information as effectively, as they could. The importance of motivation and its association with degree of effort and search arises because the relation between inputs and outputs is not a determinate one. There are four reasons why given inputs cannot be transformed into predetermined outputs: (a) contracts for labor are incomplete, (b) not all factors of production are marketed, (c) the production function is not completely specified or known, and (d) interdependence and uncertainty lead competing firms to cooperate tacitly with each other in some respects, and to imitate each other with respect to technique, to some degree.

Leibenstein, Harvey. “Allocative Efficiency vs. ‘X-Efficiency.’” The American Economic Review, vol. 56, no. 3, 1966, pp. 392–415. JSTOR,

By the way, the entire paper is worth reading, because it contains multiple delightful nuggets. The Hawthorne effect, which I mentioned in yesterday’s blogpost makes an appearance, and it also helps one understand why microeconomic textbooks are a very poor way to learn about the real world. Consider this delightful quote, for example:

One idea that emerges from this study is that firms and economies do not operate on an outer-bound production possibility surface consistent with their resources. Rather they actually work on a production surface that is well within that outer bound.

Leibenstein, Harvey. “Allocative Efficiency vs. ‘X-Efficiency.’” The American Economic Review, vol. 56, no. 3, 1966, pp. 392–415. JSTOR,

OK, so people and firms are both not as efficient as econ textbooks make them out to be. This is not, to put it politely, headline material in the non-econ world. What might be potential solutions?

In situations where competitive pressure is light, many people will trade the disutility of greater effort, of search, and the control of other peoples’ activities for the utility of feeling less pressure and of better interpersonal relations. But in situations where competitive pressures are high, and hence the costs of such trades are also high, they will exchange less of the disutility
of effort for the utility of freedom from pressure, etc


In English, this means the following:

  • Government offices are unlikely to be as productive as private sector offices
  • Surround yourself with folks who are go-getter types
  • And this is my take: figure out for yourself a good boss/manager/mentor who will push you, but in a non-zero sum way

This last part is all but impossible, but oh-so-important.

In any case: x-inefficiencies. An underrated topic from micro!

Notes from Does Management Matter? Evidence from India, by Bloom et al

  • Yesterday, I had linked to a paper by Bloom et al, and said that it would be a good place to start reading about productivity, particularly from an Indian point of view. Here are my notes from the paper:

  • As per Hsieh and Klenow the ratio of TFP in Indian and Chinese firms is 5(!) between the 90th and the 10th percentile
  • The quality of management, and therefore management practices, is one explanatory factor
  • Economists tend to not buy into this because they assume that profit maximization implies cost minimization
  • So in other words, if firms are not minimizing costs by adopting good management practices, it is because “wages are so low that repairing defects is cheap. Hence, their management practices are not bad, but the optimal response to low wages.”
  • In this paper, large multiplant textile firms were split into treatment and control groups. The treatment groups were given management consulting from a top consulting group, the control groups weren’t.
  • The result: “We estimate that within the first year productivity increased by 17%; based on these changes we impute that annual profitability increased by over $300,000. These better-managed firms also appeared to grow faster, with suggestive evidence that better management allowed them to delegate more and open more production plants in the three years following the start of the experiment. These firms also spread these management improvements from their treatment plants to other plants they owned, providing revealed preference evidence on their beneficial impact.”
  • So why wasn’t this being done already?
    • No need, because benchmarking was with local competition, who weren’t doing it anyway
    • Simple lack of awareness
    • A naïve belief that nothing would change by adopting these practices
  • But even within local competition, why did firms not exit?
    • Competitive pressures were heavily restricted
      • High import tariffs
      • No entry of firms by lack of external finance
      • Number of male family members
      • Lack of trust of professional managers (family owned businesses)
  • TFP in India is about 40% that of the USA, as per Caselli 2011
  • “Indian firms tend not to collect and analyze data systematically in their factories, they tend not to set and monitor clear targets for performance, and they do not explicitly link pay or promotion with performance. The scores for Brazil and China in the third panel, with an average of 2.67, are similar, suggesting that the management of Indian firms is broadly representative of large firms in emerging economies.”
  • The interventions comprised of improvements in:
    • Factory operations
    • Quality control
    • Inventory
    • Human Resource Management
    • Sales and order management
  • This was done by implementing the following steps:
    • A diagnostic phase
    • An implementation phase (this was for only the treatment group, obviously)
    • A measurement phase
  • The authors carefully consider whether the Hawthorne effect was at play, and reject the possibility.
  • ” In every firm in our sample, before the treatment, only members of the owning family had positions with any real decision-making power over finance, purchasing, operations, or employment. Non-family members were given only lower-level managerial positions with authority only over basic day-to-day activities. The principal reason seems to be that family members did not trust non-family members. For example, they were concerned if they let their plant managers procure yarn they may do so at inflated rates from friends and receive kickbacks.”
  • “A key reason for this inability to decentralize appears to be the weak rule of law in India. Even if directors found managers stealing, their ability to successfully prosecute them and recover the assets is likely minimal because of the inefficiency of Indian courts”
  • “Hence, the equilibrium appears to be that with Indian wage rates being extremely low, firms can survive with poor management practices. Because spans of control are constrained, productive firms are limited from expanding, so reallocation does not drive out badly run firms. Because entry is limited, new firms do not enter rapidly. The situation approximates a Melitz (2003)–style model with firms experiencing high decreasing returns to scale due to Lucas (1978) span of control constraints, high entry costs, and low initial productivity draws (because good management practices are not widespread).”
  • There are three reasons for inefficiency:
    • motivation problem
    • inspiration problem
    • perception problem
  • I need to read Lucas (1978) and Melitz (2003) next!