Our ancestors 100k years ago had the same cognitive capacity as us (evolution works slowly) and yet all modern technology and knowledge has only emerged in the last 1000 years or so.

Why such a sudden jump?

It’s not because our individual intelligence improved, but that we assembled learning flywheels over time (writing, books, schools, colleges, scientific method) and those caused each individual to be compound over the previous generation leading to the culture explosion that were going through.

A lot of AI/LLM skeptics are missing the fact that it’s the learning flywheels that matter (once you have a threshold level of cognitive capacity, which we seem to have right now even with open source models).

And apparently these learning flywheels for LLMs are already in motion – you have a model that does OK on high school math, and you can put that model in a loop to eventually do graduate level math.

These learning flywheels are going to get assembled over time and then I expect with AI systems we will see what we saw with humans – not much happening for a while and then everything suddenly changing at once.

Join 200k followers
Follow @paraschopra

The post Learning flywheels are all you need appeared first on Inverted Passion.

This essay is part of the series in which I talk about my learnings and insights building a habit coaching app (Nintee) in 2024. It didn’t ultimately work out because an app has marginal influence in a human’s life (v/s that of friends, family, culture and immediate environment). Most apps that work in the category operate like gyms (charge upfront when the motivation is high, and be okay with high churn). I had raised VC funding for it and later it became clear to me that this wouldn’t be a VC scale business, so I shut it down and returned the remaining funding. Hope the insights learned along the way would turn out to be valuable to others.

This series comprises of the following essays:

• Science of habit building: how habits are formed and broken
• Making a product that Marl loves: why well-intentioned apps ultimately become attention-seeking and gamified
• The two views of rationality (this one): what is true v/s what is useful
• How does behavior change happen: frameworks and mental models for human behavior change
• How to coach someone: 21 points to keep in mind while coaching someone

Subscribe to the blog if you want to be notified of future essays.

Following is an internal note that I had prepared while researching self-help industry and what can we learn from it for Nintee. I’m reproducing my notes verbatim. You’ll find phrases such as “we should do this” which indicate my mindset at that time.

Self-help has its own share of criticisms. It’s important to analyze them and see what our stance is. (Obviously, we will never do anything that directly increases suffering – it’s our mission to reduce it).

Lack of scientific evidence

Critics claim that many self-help methods lack scientific backing and rely more on anecdotal evidence or personal testimonials.

It’s important to distinguish between what’s scientifically true (traditional idea of rationality) and what’s empirically useful (the deeper view).

Many “unscientific” techniques are placebos and, like placebos, they actually help people. We need to be aligned what is empirically known to help people, instead of what is scientifically known to be true.

Overemphasis or quick fixes

Some argue that self-help books and programs often promise immediate results, leading to unrealistic expectations and disappointment.

This is a valid criticism but as long as we align ourselves with what’s empirically known to be useful, we will ensure that we don’t make a claim that cannot hold in practice. Also, one of the core components of our product is habit / goal tracker which orients towards actually acting on the advice.

Commercialization and profit-driven motives 

The self-help industry is seen by some as more interested in making money than genuinely helping individuals, leading to a proliferation of questionable products and services.

There’s nothing wrong with commercialization (even therapists charge for their time). But it’s possible that perverse incentives can be set up in self-help where trust and authority is exploited for profit gain at the expense of the individual.

I think we can avoid such perverse incentives by having a simple subscription model which gives access to entire platform, so that we aren’t charging anything extra for “special” products or features (which can lead to a possibility of our advice twisting towards our profit, and not towards what’s helpful.

Oversimplification and one-size-fits-all approach

Critics argue that self-help materials often oversimplify complex issues and offer generic solutions that may not be applicable to everyone’s unique circumstances.

Obviously, this is our differentiator. We’re going to provide personalized self-help.

Blaming the individual

Some argue that self-help materials put excessive pressure on individuals to solely take responsibility for their problems, without acknowledging systemic issues or external factors beyond their control.

Our approach to self-help is that of positive-reinforcement. We would strictly NEVER blame the individual, but rather explain what circumstances could have led to the undesirable condition the individual finds herself in and how to get out of it.

Our differentiation: self-help made fun, structured and personalized

Nintee will make self-help fun, structured and personalized. We promise to deliver it by making a personal coach for every human being on the planet, and packaging it in an experience that’s extremely delightful.

Why focus on personalized?

People have unique needs in growth. But most self-help is done via books or courses, which are one-to-many mediums. The only one-to-one self-help available is therapy, which is often expensive and not sought because it has a stigma attached to it.

How can we make it personalized (and why the mainstream can’t)?

We will utilize AI/LLMs to make it personalized to specific individuals. The mainstream (mostly book/course publishers) don’t have expertise for this.

What does personalization include?

• Understanding user personality and adapting approach/advice accordingly 
  • Different people are motivated by different things and have different temperament. We need to adapt accordingly.
• Serving the long-tail of user goals
  • Right from helping stop nail-biting to successful mid-career switch from a lawyer to a guitarist
• Anticipating user needs and addressing them proactively
  •  One to one focus would mean we anticipate what the user needs proactively

Why focus on structured?

Bite-sized self-help content is everywhere (on YouTube, Instagram, etc). Going through those scattered content leaves people confused. We will help them give a structured path so they can master a particular area of life skill. Books give structure but people aren’t reading them anymore.

This is similar to Duolingo. There’s lots of content on how to say hello in different languages, but for structured path to learn a language you need an app.

Why focus on fun?

Because people are more likely to engage with an experience that’s fun and delightful, v/s one that feels like work. Courses and books have a high abandonment rate because they feel like work. An experience that doesn’t feel like that has high retention (think of games, Duolingo, Instagram).

How can we make it fun (and why the mainstream can’t)?

Our hyperfocus on creating delightful experiences will create org capability over time. The mainstream (mostly book/course publishers) don’t have expertise for this, and it is difficult to hire tech-savvy designers in non-tech businesses.

How fun and personalized will help us attract customers

User experience does not have a ceiling – you can make it as good as you can. So, to attract the people interested in self-help to our app, we have to absolutely nail the UX. Our inspiration here is Duolingo which entered into an established category of language learning, and because of their hyperfocus on user experience, they won the market. 

Today, anyone who wants to casually learn self-help downloads duolingo and uses it (instead of buying a book). We want to be the same for self-help: anyone casually interested in self-help should use Nintee (instead of buying a book).

Useful thumbrule

As we develop our app, we should look at each screen and ask ourselves: “does it look fun? Is it personalized?”

Sidenote: how existing industries/players get disrupted

There are two ideas that support our approach.

Disruption from the edges

Typically, a startup focuses on satisfying the needs of customers that are typically ignored by the mainstream players. These customers often have weird/extreme needs that the mainstream cannot fulfil. A startup addresses these and wins those initial customers and slowly iterates its way to the mainstream customers who adopt the startup’s solution later. The mainstream ignores the startup initially because it looks like a toy focused on customers who are not their focus. By the time mainstream notices what happens, the startup is big enough.

A simple example of this is early cloud providers. They were satisfying the need of extreme customers who wanted on-demand software (while the rest of the world wanted on-prem software). These early cloud providers seemed like a harmless toy, but over time iterated their way to the mainstream.

The important aspect here is that a startup has to differentiate from the mainstream on clearly marked attributes, attracting customers who appreciate those attributes. There is no point competing head to head and attracting the average customer because she is already satisfied with the mainstream options. In our case, our differentiation will be on three dimensions: fun, structured and personalized.

Read more about this idea here and here (quote from this book).

Which dimensions to pick for differentiation is an important consideration. Not any two dimensions are equally important. It matters what you choose to differentiate on. If you differentiate on dimensions that humans naturally prefer to move towards, you win. Otherwise, you lose.

In our case, we have high conviction that people will prefer personalized experiences, so if we provide that, it’s a net value add. We also have high conviction on the importance of user experience and making it enjoyable and fun. This conviction comes from the fact that there’s no upper-limit of UX and people always prefer a better UX over a worse UX (in fact, Apple’s premium is entirely dependent on this).

Who is Nintee not for?

Nintee is for those who are casually interested in self-help or growth by learning broad, horizontal life-skills.

So, Nintee is not for those:

• People with clinical mental health issues
  • They require therapy from certified professionals
  • Nintee is for casual self-help enthusiasts, anyone for whom downstream consequences for not getting sufficient help are immense, we’re not for them
    • Just like Duolingo is for casual language learners. If your job depends on learning a language
• Who don’t seek self-help at all / non-growth minded
  • Stability seekers  (those who want to maintain status quo)
  • Risk-averse (those who dont want to go outside comfort zone)
  • Pleasure seekers (self-help requires short term sacrifice) 
  • Family-oriented (those who derive bigger joy from family/friends v/s self-growth)
• Those who are skeptical of self-help
  • Many skeptics exist who believe self-help doesn’t have value, and exists for fooling people. We can’t help such people
• Those who require in depth skills
  • Nintee is not for those who want to learn coding for career, but is for those who want to think like a programmer

Emotional v/s functional needs

Products fulfil two kinds of needs: emotional or functional. Functional needs are utilitarian (e.g. calling a taxi via Uber). In such cases, product is a means to an end. Emotional needs require fixing a bad feeling or elevating to a good one. Products that fulfil emotional needs are an end to themselves (e.g. a good music track, or a game). Many products fulfil both (e.g. Instagram can serve a functional need of being a platform for brand, but also an emotional need for social validation).

At Nintee, we will fulfil both emotional and functional needs, but be biased towards emotions. The reason for this is simple: functional needs are already fulfilled by the internet. To grow in life, you have all the information in the world. It’s just doing it is very difficult, and hence there’s a requirement for something to fulfil the emotional need of growing.

Top 3 emotional needs in the casual self-help market

These needs are ultimately derived from basic psychological needs:

• Instant gratification
• Aesthetically pleasing experience
• Relatedness / community

Top 3 functional needs

• Convenience
• Low-time commitment
• User-friendly design

How does self-help benefit

At a meta-level, here’s what self-help benefits people from:

• Self-efficacy and confidence: the confidence that they can get things done, which creates a positive feedback loop of actually doing things
• Structured tools for life skills: a way to get things done
• Accessible info: easy to consume information instead of academic jargon
• Social support: communities form around key ideas in self-help and people gain identity from it
• Experimentation ground: variety of techniques for people to experiment with to see what fits them best
• Placebo satisfaction: feeling that they’re taking effort towards improving their life is good enough for most people

Most common self-help needs

What people are looking for in self-help can be assessed by top self help books of all time. These can be categorized into following categories:

• Stress & Anxiety Management
• Self-esteem & Confidence Building
• Goal Setting & Productivity
• Relationship & Communication Skills
• Career Advancement
• Mental Well-being & Mindfulness
• Physical Health & Fitness
• Personal Finance & Wealth Management
• Spiritual Growth
• Personal Identity & Self-Discovery

(This corroborates with replies to my tweet)

What motivates people seeking self-help

It is significantly similar to what motivates humans. These motivations are simply applied in a different context

Gaining trust as a prerequisite for winning in self-help

Any time we’re talking about helping people, the first question in people’s mind would be: “why should I trust you / this advice?”

So, gaining trust becomes paramount in the coaching / self-help industry. People would be constantly asking: “can they deliver results? Do they know what they’re talking about? Should I trust them?”.

There are various ways to gain trust, and we must do whatever it takes to establish trust:

• Instant gratification that feels personalized
• Testimonials
  • Coaching requires time for results, so testimonials become paramount
• Research
  • For scientifically minded people, show what you’re talking about is backed by evidence
• Expert sources
  • Cite expert sources
• Data privacy
  • Since this is an app, we ought to be very vigilant about user data. If there’s a data mishap, we won’t recover from it
• Hallucinations
  • If AI hallucinates something that’s untrue, we lose trust
• Harmful advice
  • If AI gives harmful advice, we are in legal, business and moral issue

Aligning our product roadmap

• Goal tracking and coaching -> aligns to acting towards goals
• Learning -> aligns to new knowledge and skills
• Advice -> aligns to giving immediately useful advice

Zooming in: early audience

At the start, we’re targeting people who want to break bad habits and build good ones. We will iterate from there.

We will utilize the science of habit building for nailing this aspect.

Join 200k followers
Follow @paraschopra

The post The two views of rationality appeared first on Inverted Passion.

I recently built an open source Chrome extension that scrapes titles of your conversations with Claude and ChatGPT and then asks an AI to imagine a visual that best captures your current state of mind.

So imagine you’re talking about statistics / correlation and when you open up a new tab, you see this:

The project is open source, so you can try it yourself: https://github.com/paraschopra/murmuration

You’ll need an OpenRouter key and note that it may cost ~$5/mo for generating 3 visuals/day with Sonnet 4.6.

Here’s how it works:

• Whenever you visit ChatGPT or Claude, it scrapes the titles of conversations
• It takes all such topics and randomly selects a few topics to generate a visual
• A model on OpenRouter takes the topics and outputs a self-contained html/css/js visual

I’m absolutely mesmerized with this stuff!

The fun part is that you’d never know what you’ll see next!

I got the idea for this while visiting an immersive art exhibit on a Saturday evening and then on Sunday morning, it came to life. It’s truly astonishing how far these models have come!

Join 200k followers
Follow @paraschopra

The post What does your AI dream of? appeared first on Inverted Passion.

This essay is part of the series in which I talk about my learnings and insights building a habit coaching app (Nintee) in 2024. It didn’t ultimately work out because an app has marginal influence in a human’s life (v/s that of friends, family, culture and immediate environment). Most apps that work in the category operate like gyms (charge upfront when the motivation is high, and be okay with high churn). I had raised VC funding for it and later it became clear to me that this wouldn’t be a VC scale business, so I shut it down and returned the remaining funding. Hope the insights learned along the way would turn out to be valuable to others.

This series comprises of the following essays:

• Science of habit building: how habits are formed and broken
• Making a product that Marl loves (this one): why well-intentioned apps ultimately become attention-seeking and gamified
• The two views of rationality: what is true v/s what is useful
• How does behavior change happen: frameworks and mental models for human behavior change
• How to coach someone: 21 points to keep in mind while coaching someone

Subscribe to the blog if you want to be notified of future essays.

We started building Nintee first as an AI-driven weight loss coach and then as a habit coach. We wanted to change the world by helping people achieve their goals in a methodical fashion. The plan was simple: internalize the science of habit building into an app and give people an AI coach for motivation.

As we surveyed the market, we didn’t find even a single a competitor. Positive news for us? Why couldn’t we find successful habit coaching apps? Were there any successful apps which helped someone become a better person, or pick up a skill? The only name for a successful skill-building app that came up was Duolingo. They made $700mn+ of revenue in 2024, so clearly this model could work, right?

As we studied Duolingo, it became clear that they weren’t a language learning app but rather a game. Reddit is full of stories where people say they’re addicted to Duolingo but can’t speak the language they’re learning. Why would Duolingo be optimizing for game-like experience and not for actually teaching a language.

The answer lies in the fact that the average marginal user of mobile apps is minimally committed, highly distracted, constantly craving dopamine hits and always ready to ditch apps after a microsecond of boredom. For investor-backed startups and public companies like Duolingo, this average marginal user is the one they end up targeting because that’s where the eternal growth comes from.

So, it was clear to us that for Nintee, a VC-backed startup, to succeed, we had to build a game for Marl where he feels his time a better time spent that other casual games or social media. Because if we can’t compete against cute cat photos, we’re toast.

We decided we had to build apps that give Marl meaningful fun – in retrospect, we should have realized that this was an oxymoron but we still had (naive) hopes that you could make experiences that are both valuable and fun. In retrospect, it became clear that the reason edutainment as a category never took off is because it is neither entertainment nor fun.

But, for now, let’s dissect Marl.

Following is my (verbatim) notes from the time period when we were researching who this Marl is and what does he need.

Who is Marl?

Marl has:

• Short attention spans
• Need for instant gratification / dopamine hits
• Aversion to effort of any kind

Note that we are all Marl for most of the day and are largely driven by our evolved animal instincts. Why? Because using our brain is costly/effortful, and we only do so for highly-motivating projects (like job where we get paid).

What about non-Marls? Those who behave logically, rationally and are ready to put effort

Of course, nerds with obsessions exist, but they’re a tiny market. To make a big consumer business, either you:

• Make a high-priced product and be able to sell to nerds
• Or, make a mass-market product that appeals to Marl

In our case (personal development on mobile), the former isn’t possible AND the opportunity is not clear too.

So, we chose to serve to Marl and make a product that is broadly appealing.

Notes from the article Tyranny of The Marginal User

Article

• “Nearly all popular consumer software has been trending towards minimal user agency”
• “Since most software products charge a flat per-user fee (often zero, because ads), and economic incentives operate on the margin, a company with a billion-user product doesn’t actually care about its billion existing users. It cares about the marginal user – the billion-plus-first user – and it focuses all its energy on making sure that marginal user doesn’t stop using the app.”
• “The first thing you need to know about Marl is that he has the attention span of a goldfish on acid.”
• “Marl’s tolerance for user interface complexity is zero. As far as you can tell he only has one working thumb, and the only thing that thumb can do is flick upwards in a repetitive, zombielike scrolling motion. “
• “Marl will never click through any of your hamburger menus, never change any setting to a non-default”
• “Insufficiently stimulated, Marl throws a fit and swipes over to TikTok, never to return to your app”
• “Marl can also be a state of mind. We’ve all been Marl at one time or another – half consciously scrolling in bed, in line at the airport with the announcements blaring, reflexively opening our phones to distract ourselves from a painful memory.”

HN comments:

• “There will always be a way to find more Marls to add to your user pool because Marl is the basest human need for a steady effortless dopamine drip. Just about everyone has some amount of time that they spend as Marl, so there is an almost limitless pool of Marl time to pull new users from.”
• “If you are making a product for everyone, Marl is the only persona that is in everyone, so you should probably target Marl.”
• Why build systems that encourage this?
  • Because there is an endless amount of human time spent like this, but we all have limited attention for other things. If you try to appeal to someone’s better parts, then there is a limit to the amount of attention that they can apply to your product. If you try to appeal to someone’s base need for dopamine, then the limit of attention is much higher
• This is something important that people do not seem to grasp. Intelligence is a really high dimensional thing. Even the ones that are highly intelligent in some dimensions are dumb as rock in vast majority of other dimensions. So we all are basically morons with some occasional flashes of intelligence in some individuals.
• In my experience, making stuff that provides no intrinsic value like video games, only value of meaning, my games have only gotten better by catering to those impulses, which are really not that negative or stupid.
  • Really, who needs tutorials? Why make stuff that needs a tutorial? You can have complex games without tutorials and FTUEs. You don’t need so much UI. It’s not about scrolling so much as it is that so many apps obscure, rather than transmit, anything meaningful, through really obnoxious UI.
• Now, let’s find any single company/product that actually is successful by encouraging us to become better versions of ourselves. [link]
• Marl isn’t the “marginal” user, Marl is the “average” user. If the average user actually cared about deep and meaningful content, then any A/B test that throws her under the bus in order to please Marl will show bad data, and the proposed change would be killed.
  • The average user doesn’t want deep and meaningful content. The average user is Marl. That is why every product, no matter how noble it starts off, eventually degenerate into Marl-fodder. Because that’s where the money is. The only way to escape this is to take on a huge pay cut and work at a company that doesn’t care about growing profits. Go ahead, you first.
• “if you walked up to Marl, built trust with him, and asked him whether he wanted more meaningful content in his life (for a definition of meaningful which made sense to him) I think he would say yes.”
  • If you walked up to me, built trust with me, and asked me if I wanted more exercise in my life I’d say yes. And yet.
• I’m mixed on the shorts. I like it when they do a “you fix this by pushing this button here” in 15 seconds instead of it being 8+ minutes so they can get mid roll ads.

The average user used to be an 18-40 year old person with post-secondary education, in a mid to high income country, using a laptop. Now, The average user is a 10-65 year old person using a smartphone.

Notes from Nikita Beir (for social apps mostly)

• The people and content on an app always trump slick design & novel interactions. So focus more on getting network effects and solving the “cold start.” 
• Habit formation requires recurring organic exposure on other networks. Said another way: after people install your app, they need to see your content elsewhere to remind them that your app exists (e.g., Instagram photos on Facebook, TikTok videos on Instagram).
• People download apps to solve core human needs (1) finding love, (2) making or saving money, and (3) play. People rarely take time out of their day for anything else.
• The number of social products that took off among older audiences can be counted on 1 finger. Our habits become immutable as we exit our formative years.
• One of the most important differentiators between great consumer founders and the rest is a willingness to acknowledge a shameful truth about people—and then being comfortable attaching your name to that product. 

• Think of consumers as lizard brains: they mindlessly tap on rectangles on a screen for basic needs—like making money or finding a date
• A dumb PM will stare at an onboarding funnel for hours, fabricate problems, create months of work for engineers, and ultimately improve conversion by 5%. A smart PM will look at funnel in a minute, make a button bigger, and improve conversion by 20%.
• We tend to overcomplicate what it means to have great product sense. To put it simply: you can look at a screen and intuitively predict the percent of users who will convert to the next screen within a 10% margin of error.
• There’s some weird cognitive bias where product creators vastly overestimate how likely people are to do or pay attention to something based on how much the product creator wants them to, and they vastly underestimate how lazy people are.
• They also don’t realize that yea I could keep going through thier onboarding flow, but timewise it’s competing with me looking at more dancing/cat videos on Tik Tok, and that is some seriously compelling content to compete with!

• When designing a consumer product, you should consider every tap by a user to be a miracle. The motivation to stop using a new app will always be stronger than to use it To signup & complete a profile on Gas, we got it down to 15 taps total—with no keyboard required at any stage

• Needing a user to type with a keyboard at any stage to become activated is asking for failure.

Join 200k followers
Follow @paraschopra

The post Making a product that Marl loves appeared first on Inverted Passion.

What I have now:

• Everything organized by sprints in ./sprints/v1, v2 etc folder

• Custom command /prd to help me brainstrom requirements for a sprint and break it down into atomic tasks (which should take 5-10 mins each)

• Custom command /dev to pick highest priority task in a prd and follow test driven development to implement the tasks

• Custom command /walkthrough to write a sprint review report that details what code was produced so I can read and understand exactly what the code does

• I run Claude Code in docker with bypass permissions so it can run overnight without asking me anything. I use the excellent cco [1] for that!

• I made a Codex skill in claude so it can consult Codex for plans, tests, code review, etc, <- amazing what Codex catches that Claude doesn't

• Specify in /dev that testing has to happen via browser screenshots (so claude installs playwright and uses headless chromium for end to end testing)

Oh yes, btw, it helps that I have Claude Max $200/mo plan so basically I run everything on Opus and never worry about limits.

Now I feel I have a setup that automates 95% of whatever I want to build. I just brainstrom per sprint, and everything else happens automatically while I read research or live my life.

The only question now is – what to build? what to explore? everything seems possible.

[1] CCO https://github.com/nikvdp by @arghzero <- shoutout to him, he has added features very quickly to the repo. Super-responsive!

Join 200k followers
Follow @paraschopra

The post My Claude Code workflow appeared first on Inverted Passion.

Evolution doesn’t like to waste energy. Hence all communication between people is a coordination tool where all parties are interested in getting their job done, but not wanting to invest more energy than it’s necessary to do so.

So if someone uses the word “God” or “Love”, the job is done if it elicits the emotions, actions and associations roughly associated with what the speaker intended so our search for what those words “truly” mean is just misguided. Meaning is in what the exchange does in a particular context. By themselves, words are empty.

A lot of philosophy is misguided in that sense where we end up attributing more depth to words than they contain.

This suggests a guiding principle: focus on what words do and not what they mean.

For example, when someone says “that’s not fair!”, don’t ask what fairness truly is. Ask: what is this utterance doing? (Signaling displeasure? Requesting redistribution? Trying to game the system? Pointing to a flaw in current scheme of things?)

Join 200k followers
Follow @paraschopra

The post Words are minimum-viable coordination tools appeared first on Inverted Passion.

why would anyone change their behaviour to accommodate your product?

And the only correct answer to this question is that they’re *already* exhibiting the behaviour and your product will simply help them be >2x more efficient on a dimension they cares about.

Behaviour change is tough. Internalize that nobody changes their behaviour for marginal or incremental benefits.

And nobody certainly cares about your idea.

What everyone cares about is themselves and things that bring efficiency in how they’re already living or are strongly motivated towards (but that shows up in their attempts to be that way).

Join 200k followers
Follow @paraschopra

The post Nobody cares about your idea. appeared first on Inverted Passion.

Some people want to pause all AI development because of this risk of human disempowerment (+ also possible extinction risk). But stopping tech development also means we give up on benefits and abundance that could come along.

Is there a middle ground? Can we keep reaping the benefits of AI without rendering humans obsolete.

I think so. Here’s how we can probably achieve it.

The issue we’re dealing with isn’t AI development, but the development of general intelligence. So, governments should start measuring AIs on a scale of generality.

Allow and encourage narrow AI that augments people and leads to medical breakthroughs (like AlphaFold) but disallow general AI that scores above a certain threshold.

We already have validated measures of adult cognitive dimensions, so we can simply use them to test new releases. In fact, a recent paper [1] measures AGI on these dimensions so governments should simply use the same ones. And because the field is evolving so fast, probably govts should review and iterate the measure every quarter.

We already have standards for car engines to prevent pollution, why not have standards for AI releases as well?

By the way, my proposal doesn’t limit research into what general intelligence is. I think humans are inherently curious and we shouldn’t limit investigation. I’m merely proposing to prevent deployment of a general intelligence.

As a counter to my proposal, an e/acc pilled person may argue that innovation is good and for certain problems we do need general intelligence. I agree, but for both counts we do have humans to fill the generality gap.

What we want is better tools, not a replacement for ourselves.

[1] A Definition of AGI

https://arxiv.org/abs/2510.18212

Join 200k followers
Follow @paraschopra

The post A proposal to prevent job losses from AGI appeared first on Inverted Passion.

This essay is part of the series in which I talk about my learnings and insights building a habit coaching app (Nintee) in 2024. It didn’t ultimately work out because an app has marginal influence in a human’s life (v/s that of friends, family, culture and immediate environment). Most apps that work in the category operate like gyms (charge upfront when the motivation is high, and be okay with high churn). I had raised VC funding for it and later it became clear to me that this wouldn’t be a VC scale business, so I shut it down and returned the remaining funding. Hope the insights learned along the way would turn out to be valuable to others.

This series comprises of the following essays:

• Science of habit building (this one): how habits are formed and broken
• Making a product that Marl loves: why well-intentioned apps ultimately become attention-seeking and gamified
• The two views of rationality: what is true v/s what is useful
• How does behavior change happen: frameworks and mental models for human behavior change
• How to coach someone: 21 points to keep in mind while coaching someone

Subscribe to the blog if you want to be notified of future essays.

How habits are built or broken is no longer a mystery. Psychologists and neuroscientists have figured out key mechanisms behind habits, and in this note, I want to document whatever I have learned so far (after reading 4 books on habit building and many research papers). 

What are habits? 

The definition is actually beautifully simple: 

Repeated rewarding actions performed under stable contexts 

Let’s unpack the definition: 

• Repeated: habits are formed when the same action is performed multiple times. There’s no magical number though, habit formation can take anywhere from a couple of weeks to couple of months, or even more. 
• Rewarding: habits are much more likely to form if there’s a reward given immediately after the action is performed. (Think of the fresh, minty taste of toothpaste reinforcing the behavior of brushing) 
• Stable contexts: it’s important that context under which the action is performed remains the same, for a behavior to become an automatic habit. Think of watching TV while eating lunch. The cue becomes food, and habit becomes watching TV and reward is, of course, the entertainment. The cue-response is fundamental to habit formation. This is why people advocate habit stacking, where a new habit is stacked after a pre-existing habits. It also explains why we adopt new habits when there is a major change in life (like moving cities or changing jobs) 

How to influence habit formation 

Research suggests the following elements aid in forming new habits or breaking existing ones: 

• Self-vigilance: Until a habit becomes automatic, continual reminders to one-self to perform an activity 
• Change the environment: make good habits easier to perform (more frequent exposure to cue), while making bad habits harder to perform (less frequent exposure to cue) 
• Specific implementation intentions (for strengthening cue-action loop): deciding if then-else kind of scenarios upfront as specifically and clearly as possible (e.g. when I see the elevator door, I will take the stairs instead; or when I pick up the cigarette, I will immediately break it into pieces and do push-ups)
  • Implementation intentions pass the control of behavior to the environment.
  • implementation intentions promote goal attainment by helping people get started. 
• Repeating the activity the same way each day: it’s important to keep the context from day-to-day as similar as possible, if a behavior is to be converted into a habit
• Focusing on consistency of repetitions over doing it perfectly: aiming to make any amount of progress but doing it consistency, instead of aiming to do it perfectly
• Realistic goals first and then increasing difficulty: deciding to walk 2000 steps daily first, then aiming to increase it gradually to 10000 steps. This is because unrealistic goals fail, cause dissatisfaction and hence make it unlikely for behavior to be repeated 
• Rewards and positive reinforcement: upon each repetition, if there’s a reward (of any kind), it makes the behavior much more likely to be repeated (and hence, much more likely to be converted into a habit) 

Breaking bad habits 

As far as bad habits are concerned, it’s commonly mistaken that the goal is to eliminate them. Instead of eliminating the bad habit, a better goal is to break automaticity of it. The underlying behaviours behind bad habits are often enjoyable (eating sugary food, for example) so eliminating them completely is often very difficult (and probably counter-productive). 

Our goal then should not be to eliminate such behaviors but to promote their mindful and intentional execution. 

Nevertheless, here’s how we break the automaticity of bad behaviors: 

• Self-monitoring: knowing under what circumstances does one perform the habit that one wants to break, and watching oneself carefully (thinking, “don’t do it”) 
• Mindfulness: being more mindful while performing a bad habit so you can consciously break the automaticity of habit 
• Coping planning: Deciding on the lines of.. “If a circumstance arise which pushes me to do X, I will… ” (do something else) 
• Replacement habits: Deciding on the lines.. “If I think about doing X, I will… ” (do a replacement rewarding habit) 
• Change the environment: make good habits easier to perform (more frequent exposure to cue), while making bad habits harder to perform (less frequent exposure to cue) 

Rewards and positive reinforcement: upon each successful execution of replacement habit (or null habit), if there’s a reward (of any kind), it makes the replacement behavior much more likely to be repeated (and hence, much more likely to be converted into a habit)

Join 200k followers
Follow @paraschopra

The post Science of habit building appeared first on Inverted Passion.

It’ll be inconceivable to us that there was a time when if you had a question, you did not get an intelligent instant answer for it.

Some more interesting predictions that came out from our conversation:

– Maybe we will stop having multiple softwares; just one agent that spins up personalized software on demand

– Maybe soon we will have so much automation that our AI agents will predict what we want to do next, leaving us to wonder what’s our role when we open up our work laptops

– Maybe we will have ambient, proactive agents who will monitor our meetings and before they end, actually show us results of actionables

To note – my friend is a CTO and just a few months back, his day job was writing code day in and day out.

Now he barely opens the IDE!

My friend obviously isn’t an exception; even Anthropic team is using Claude Code to write newer versions of Claude Code.

How crazy our trajectory has been in our lifetimes – from waiting to get a landline phone to having a genius in our pocket!

And it’ll only get wilder

Join 200k followers
Follow @paraschopra

The post Will we forget what pre-LLM era looked like? appeared first on Inverted Passion.