solderneer

General Magic Disease

Fri, 04 Jul 2025 00:00:00 GMT

I heard this in the General Magic documentary a few weeks ago and I haven't been able to stop thinking about it since. Not because it is beautiful (it is), but because of the cautionary tale that it tells to consumer technology companies today.

Between 1990 and 1995, General Magic spent through $96 million building their vision. The founding team, Marc Porat with Bill Atkinson and Andy Hertzfeld (both architects of the original Macintosh), were technology idols who attracted Silicon Valley's brightest minds. By 1994, they'd also assembled the General Magic Alliance: Sony, AT&T, Motorola, Matsushita, and Philips betting collectively on their vision.

At General Magic, they pioneered technologies which we now take for granted: the predecessor to USB, early touchscreen interfaces and the first mobile app ecosystem. In 1994, they launched the Sony Magic Link with their fun, skeuomorphic Magic Cap OS. The tech world held its breath.

They sold just 3000 units over 6 months.

Meanwhile, two years later, Jeff Hawkins launched the Palm Pilot. A neuroscientist by training, Hawkins carried a wooden block carved to fit in his shirt pocket, to test how people actually interact with handheld devices.

While General Magic was engineering the future, Hawkins was obsessing over the present: how often would someone pull this thing out? What would they reach for it to do? His insight was that people didn't want a handheld device that was a computer, they wanted a handheld device that was an extension of their computer.

This shaped everything. The Palm Pilot became a handheld companion device which used "hotsync" to pull calendar events and contacts through a cradle from a PC. It had a 30 day battery life, and did exactly 4 things well: calendars, contacts, to-do lists and notes. By 2000, Palm had captured 76% of the PDA market with 6.7 million units sold.

The contrast is stark: Palm succeeded where General Magic failed by focusing on how people act in the present, rather than how people should act in the future.

Palm cost $299 and synced your calendar. Magic Link cost $800 and promised to change your life.

General Magic's key mistake was that they fell in love with their own mythology. They drank their own Koolaid. While both General Magic and Palm Pilot were directionally correct about people's desire to retrieve information on the go, General Magic built an impractical solution to the problem while chasing the dream of a future which wasn't to arrive until a decade later. The unfortunate part is that by chasing that dream, they wrote themselves out of it.

The minimal magical device

Fast forward to 2007. The iPhone succeeded where General Magic failed thirteen years earlier, not just because the technology was more ready, but because Jobs understood that users don't adopt paradigm shifts. They adopt better versions of existing fundamental behaviours. And not just marginally better ones, but drastically better ones.

Jobs understood that the killer app for mobile computing came down to the human desire for entertainment and connection. That's why the iPhone launched as an MP3 player, a phone and an internet communications device. The last one being something that Palm missed, which ultimately ended up in Palm getting acquired by HP at a share price 94% lower than its peak.

But here's the key: in focusing on the core features, Jobs built the minimum ✨magical✨ device. It didn't matter that the iPhone had a crappy touch screen, no Copy/Paste, no 3G, no App Store. It mattered that people could browse the web on the go, in the best way possible. When you put people's latent needs into a product, you get an iconic device that feels like magic.

General Magic Disease

What's ironic is that General Magic ignored competition (like Palm) who were creating real cultural change around them. When you build the impression of being visionary, you stop thinking "what users actually do" and start thinking about what "users should do".

I call this general magic disease, being so in love with your own narrative that you stop seeing reality. We watch the General Magic documentary today and we recount the tragedy of those who saw the future too soon. But, General Magic wasn't early. They were wrong in all the ways that mattered. Wrong about what people cared about at the time. Wrong about culture. It doesn't count to be directionally correct about the future, any sci-fi writer can do that. What counts is finding the very specific path through which technology comes to change our day-to-day lives.

And sadly, I see general magic disease all over the Valley today.

In 2024, Humane raised $230 million to liberate us from our iPhones. The Humane AI Pin projected a laser display onto your hand, responded to AI voice commands and was marketed as the panacea for screen addiction. The battery lasted for 2 hours, it heated up dangerously, and basic tasks took 10 seconds. MKBHD called it the "worst product he'd ever reviewed". And guess what, they were acquired by HP.

someone explain to me why HP acquires everyone??

Look, the directionally correct statement is that AI is changing the way we all live and work. No one gets credit for that (except OpenAI ig). The real questions are:

What do people tolerate today that they don't have to?
What human desire has always existed but was impossible to fulfil until now?
What are new values and desires which are emerging today?

Leave everything else to the sci-fi writers.

Related Notes:
good products dont change the world manufactured normalcy

On ubiquitous gaze awareness

Mon, 10 Feb 2025 00:00:00 GMT

If you could look at a light, and snap your fingers to turn it on, people would cry witchcraft. A friend sent me a link to a (pretty old) 60 GHz radar chip this week, and it reminded me of a Google project called Soli from 2016. Micro-gesture detection using radar, in both near-field and far-field forms didn't find much applications back then.

a huge missing link in human-computer interaction right now is high-fidelity gaze awareness (and put it everywhere).

But shrink the sensor into an unified SoC, with an integrated MCU, and make the package ridiculously cheap. And then chip everything - the lightbulbs, the power outlets, the TV, the doors. The Vision Pro showed everyone the power of eye tracking, and measured pupil dilation for intent signalling, and I would think that physical environments that are intimately aware of human intent would feel like an extension of our bodies.

(isn't the reason our hands feel like a part of us, that they know what we want them to do.)

In 10 years, technical skill won’t be the moat, it’ll be about knowing what to build and getting people to care. The real advantage will come from social capital and strategic technical insight, the ability to see what matters and make others believe in it.

This massively favours people who can command attention (influence is by nature power-law distributed). But somewhat conversely, more people than ever will have the ability to build great things. As intelligence is commodified, social capital won't be necessary to hire the best and brightest people to bring social capital and access to intelligence are tied together, but that’s not going to last. The future belongs to hyper-niche innovation from anonymous builders—people who don’t need status, permission, or a seat at the table to make an impact.

~ as an aside, a tweet from @keshavchan about the importance of storytelling.

romanticising life though:

Clay

Fri, 13 Sep 2024 00:00:00 GMT

The origin story

We've been so deep in the trenches these past 6 months, that we haven't updated everyone as much as we should have. So, we wanted to share the crazy journey we've had so far and clue you into what's next at Clay. But first, a quick recap of where we left off :)

It started back in September 2023, when we took a day-long walk at Regent's Park. We talked about the future of computing, imagining how voice could become the way we interact with technology. But we also dug deep into personal topics - our goals, morality, and the legacy we wanted to leave on the world.

By the end of that walk, we had two convictions:

Voice-powered interactions will make our relationship with technology seamless, natural, and magical.
Innovation is needed to make voice interaction a practical part of our daily lives.

That's how Clay (back then, we called it Epiph) was born. We set out to make voice-based computing private, discreet, and practical, focusing on capturing ultra-quiet whispers as a new way to interact with computers. With that, we started building like crazy - making multiple demos, testing different bio-signals, designing electrode multiplexing systems and industrial design concepts.

In that time, we were incredibly lucky to have Jeremy, Qing Ru and Trent as amazing supporters who believed in our vision of private, discreet interactions that created "AI in your head". And with their help, we found ourselves presenting at the Foresight Vision Weekend (Foresight Institute) in October 2023 and winning the UCL AI Foundry pitch event in March 2024!

Through what feels like a stroke of luck, the Vision Weekend was also where we met Danielle and, later on, Harry from 1517 fund, who ended up putting in the largest angel check of $100,000 into Clay, six months ago.

That's where we left off, before we descended deep into the trenches for 6 months of prototyping, conducting research, talking to users, and pushing the bounds of what's possible with voice.

Build Time!

After raising the $100K from 1517, we hit a roadblock - spring break of April 2024. With our UCL attendance below 20%, we needed to pause for three weeks, cram a year's worth of content, and sit for our third-year exams.

But, with a mission ahead and a tight timeline, we couldn't stop for long. Our goal was to validate our technology, convert our proof-of-concept into a wearable prototype, and launch pre-orders by September. The vision? A voice-powered OS that lets you query information and take notes, coupled with ultra-quiet whisper recognition - creating the experience of "AI-in-your-head."

We had six months, and we were going to need some help.

So we put toge/var/folders/gz/pgrxg4n51svgsyw6gfphm49c0000gn/T/TemporaryItems/NSIRD_screencaptureui_VRCUtl/Screenshot 2026-01-11 at 11.46.58 AM.pngther a dream team:

Jim Bremner: ex-Head of AI at a facial biometrics scale-up, DJ on the side, and a wizard in ML engineering. Jim worked on our whisper recognition and sensor fusion models.

Daniel Lim Hai: A brilliant course mate from UCL, skilled in 3D design with OpenSCAD and electronics engineering. He worked on the development of our hardware and form factor.

Isaac Tay: Shan's childhood friend, who's done ML since he was 14. Isaac flew in to London from Singapore and lived on our air mattress, just because he was excited about the vision. He worked on both our voice-OS and our whisper recognition models.

Berkay Bilik: A maths undergraduate at UCL with a talent for writing clean, beautifully modular code. He was perfect to work on our voice-OS.

We turned Shan's small London room into our galactic HQ. We covered the walls with whiteboard paper, set up five tables, and huddled together every Friday for an internal Demo Day, to showcase our weekly progress.

In just a month and a half, we:

Collected over 50 hours of paired data from our vibration sensors and air microphones.
Achieved a word recognition accuracy of over 90%, pushing the boundaries of low-volume whisper detection. It looks like telepathy, check it out in the comments!
Developed three iterations of the hardware, each smaller and more integrated.
Built an agent-based information retrieval and note-taking backend using Burr
Explored 80+ form studies, working toward a sleeker, smaller form factor.

The pace was exhilarating, and the tech felt magical. But challenges began to crop up, as they always do.

Crashing out

We met a billionaire.

At an event in Zurich, we had the incredible opportunity to speak with Tom Preston-Werner, co-founder of GitHub, whose insights were invaluable. We also connected with a former Apple Product Manager who emphasized that building a great product isn't about flashy tech but asking the right questions. A week later, while giving an update to 1517, we also realized we were missing key answers and lacked a solid go-to-market plan, highlighting critical gaps in our approach.

There were huge gaps and this is what we realised:

We were building in isolation, missing out on feedback from our target audience.
A solid go-to-market plan and a clear distribution strategy is as important as building a good product.
Our focus on tech and vision led us to overlook critical aspects, particularly understanding our users' workflows and needs.

So we began customer discovery.

We sneaked into WeWorks across the city to interview people(we realised no one likes donuts).
We cold-called and emailed enterprises about their workflows (and even walked into real estate offices in West Hampstead).
We reached out to and talked to everyone we could, sent a bunch of surveys to fill and even unsuccessfully tried social media.

The six weeks between mid-July and the end of August were, without a doubt, the most exhausting period we've experienced. It was a relentless cycle of ideation, invalidation, and more ideation. We were pushing hard to find the right fit for our product, but looking back, it took a serious toll on us. The burnout was real, and at times, it felt hopeless - like we were chasing something just out of reach.

As August came to a close, so did our time with the amazing team we'd built over the summer. With the season ending and the lease on our flat expiring, we decided it was time to take a step back and re-assess. We wanted to reflect on what we truly wanted to create. After countless ideas and iterations, we found ourselves questioning whether we were headed in the right direction or if we were pivoting into something entirely new - and whether that was really the path we wanted to pursue.

Raghav & Sudharshan

Dream Song

Sun, 28 Jul 2024 00:00:00 GMT

in the quiet, my heart whispers dreams
softly (afraid to dream out loud)
but my heart’s song sings

dreaming out loud
is bravest, true
and in the gentle moments of night
my heart (quietly) tells me
to trust only the journey which is mine,
drape myself in dream song
for in hope there will be, fear
and loving wholly is partly unknowing

whispered dreams wither
while dreams out loud (echo and)
conspire with the universe
to reality the whispered dreams
out loud that were said

Consciousness as Consensus

Mon, 31 Jul 2023 00:00:00 GMT

Thinking about consciousness on a Monday night while sitting in bed is more enjoyable than you might think. I've read a number of books on the topic and I do recommend Godel, Escher, Bach, How The Mind Works and Other Minds as good places to start.

One of my pet peeves is that none of these books properly consider the societal aspect of consciousness, they mostly consider consciousness as an individual phenomenon. It seems plausible to me that consciousness is a derivative from the social importance of differentiating the self from the group. After all, the species that we consider to be somewhat conscious, dolphins, whales, apes are all social animals

Without a strong conception of self it becomes impossible to keep track of other cooperative or competitive players, remember faces, reciprocate favors and perform all sorts of other critical social functions. While non-social animals may still have consciousness, I have a sense that there is a selective pressure for strong consciousness in highly social animals.

Another curiosity about consciousness is that it's non-falsifiable. I know that I am conscious, but I cannot know if anyone else is. I am conscious therefore I am. But then, why do we all agree on the consciousness of each other, there cannot be any credible proof! Partly, I think this is what makes the invention of human rights (and animal rights even more so) such a pleasant surprise.

My theory is that consciousness is a social imaginary more so than the characteristic of an individual. We are conscious because we believe in the consciousness of each other, and our behavior towards each other reflects that belief. If dogs can convince all of us that they are indeed conscious, then they would be conscious just as we are. The same argument holds for artificial intelligence as well.

~ Shan

Consciousness is only applied externally from inside. It’s a social imaginary

The Machiavelli Effect

Mon, 31 Jul 2023 00:00:00 GMT

In one of my new favorite books of all time, Where's My Flying Car, J. Storrs Hall introduces the Machiavelli Effect.

Many people simultaneously fear the new and different, and they also want to protect whatever status quo they benefit from (or at least feel comfortable with).

In the book, he lays out an agreement for why there's a pervasive sense of stagnation in the developed world and puts risk-averse technology regulation at the heart of what's to blame. In summary, the people who've won in the current technological paradigm have the motive and the means (money, law, power) to maintain the status quo. And the people supporting a new paradigm, are always playing a lukewarm defense as they might or might not win in their new paradigm.

This is also coupled with Venkatesh Rao's point on how good products dont change the world. People are always slow to adopt technologies, but sometimes technologies can be so new that we use policy to destroy what we perceive to be a threat to our reality.

We look at the present through a rear-view mirror. We march backwards into the future.

This ends up triggering a meta baptist and bootleggers problem. The Baptists are the people fearful of the specific brand of technological change. The Bootleggers are those who profit from the status quo. I say meta because this might apply with generality to all new technological shifts.

The only way to beat this effect is to be highly intentional in the design of a technology. The best radical technologies today are designed look like incremental improvements or toys, until they're not. This form of design is less likely to face regulatory pushback, and can slowly shift the Overton window of the population until widespread adoption is achieved.

Maybe distributed ledgers are in the tail end of that toy phase now? Of course, intentional design is necessary but not sufficient, we also need to play the game well :) The challenge for blockchain tech is a delicate balancing act of avoiding the most harmful regulation while embracing regulation which genuinely avoids harm.

This seems to also be the opportunity for people working on other frontiers to watch and take notes. Biotech, neurotech, nanotech, and all other forms of paradigm shifts might face equivalent or worse challenges. And there's no better time to learn than now.

~ Shan

Baptists and Bootleggers

Sun, 30 Jul 2023 00:00:00 GMT

On January 16th 1919, the 18th Amendment to the Constitution prohibited the production and sale of alcohol in the United States for almost a decade and a half. Prohibition was supported by two groups that could not be more diametrically opposed. On one side were the Baptists and other evangelical Christians who saw Prohibition as ridding society of ills such as alcoholism and domestic violence. On the other side were the Bootleggers, whose moonshine prices were driven up by the regulation against alcohol.

This creates a paradoxical effect where coalitions with opposed interests find a common point of agreement in regulatory policy. The Baptists take the moral upper hand and help with enforcement, while the Bootleggers make profitable monopoly deals with the regulatory authorities.

I thought this was a fascinating concept by Bruce Yandle, and I find it surprisingly useful in analyzing lots of tech news. Consider:

In AI Safety, Effective Altruists playing The Baptists, and OpenAI playing The Bootleggers.
In nuclear power, Environmentalists playing The Baptists and Fossil Fuel Companies playing the Bootleggers.
more to be added...

~ Shan

Wireheaded

Fri, 28 Jul 2023 00:00:00 GMT

This note is the twenty-ninth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

I've been coming across the concept of wireheading a lot in my reading. In Superintelligence, Nick Bostrum uses it to refer to the idea of artificial intelligence systems which hack their own reward pathways, in effect lazily short-circuiting their intended behavior.

Generally, wireheading is when any agent (artificial or otherwise) artificially induces and gets addicted to pleasure, in the human case typically due to some brain stimulating machine. This implies that a wireheaded agent is impaired in their ability to exert their agency in dimensions other than maintaining the wirehead. tldr, a kind of electronic drug that is impossible to resist.

Something that's scaring me is our openness to a future where we're all wireheaded. Take social media for example, what are the engagement-driven recommendation algorithms but a weak form of wireheading. Or Netflix which seems hell-bent on getting us hooked on as many shows as we can fit in 24 hours. All while Amazon's next-day delivery keeps our dopamine hits coming and our hands twitching for the next purchase.

In big-tech FAANG, I'd argue Apple is the only company which isn't built on weak wireheading. And of aspiring consumer tech startups, it seems a good proportion of products are built on viral growth in the style of Nir Eyal's Hooked.

This doesn't sit right for me. I'm a big believer in technology which amplifies human agency (technomorality) and it feels like there's less of that going around. Instead the constant entertainment environment feels like the path towards whatever the humans in WALL-E were up to.

This doesn't have to be the way. We can still shift gears and collectively focus on building technologies which empower us, rather than those which keep us placated. I think the future we were promised depends on it.

~ Shan

Why AI isn't a tool

Fri, 14 Jul 2023 00:00:00 GMT

This note is the twenty-eighth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

People talk about how artificial intelligence is a tool. It seems to me that these people who claim so, are either ignorant of intelligence's potential or are specifically referring to artificial intelligence in its current form today. I think artificial intelligence, in its general sense, cannot be a tool.

Let's start in the nature of intelligence. Tools are designed to fulfill a purpose, a hammer hits a nail and a computer computes computations. Tools are mechanical in the sense that they follow defined steps to a predestined outcome. If the desired outcome changes, the tools do not adapt to those changes.

Weak AI in its current form, fits this description. It tends to be good at one thing but terrible at everything else, just like a conventional tool. But general intelligence, is not a series of steps, but a way to generate a series of steps. General intelligence is the ability to plan and execute steps in pursuit of any arbitrary goal.

If intelligence was a tool, it's a hammer to which everything can be a nail. Intelligence is a rising tide that lifts all boats, people who are more intelligent tend to also be more healthy, more wealthy and just better at all sorts of things. While tools are for fulfilling a purpose, intelligence seems to be for fulfilling any purpose.

This is why I think none of us can afford to ignore artificial intelligence, whether we care about the technology or not. Already our newest tool AIs can make their own decisions in multi-step problems, these are the first tools in history to have agency.

This means massive upside for us, as I've said in super obsidian, and massive risk as well, like I said in fear irrelevance.

At night, I dream of what life might look like when computers go from being tools, to being agents. It looks likely that we might need answers to those questions in our lifetimes.

~ Shan

People you want to work for

Thu, 13 Jul 2023 00:00:00 GMT

This note is the twenty-seventh letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

Today, I listened to a fascinating podcast episode between Mark Zuckerberg and Lex Fridman. A lot of the conversation was around personal LLMs, and Meta's vision for open-source models, but one quote in particular stood out to me.

When Lex asked Mark, "How do you choose who you hire?", Mark responded almost instantly. "You just hire people you wouldn't mind working for."

And that's genius. You wouldn't want to work for someone who wasn't nice, ambitious, did good work, inspiring etc. The filter of do-you-mind-working-for-them, is a neat way of wrapping up that bundle of concepts in a way that can be very quickly evaluated.

Anyway, that's it for now, just amazement at the surprising effectiveness of simple heuristics. Have a good day :)

~ Shan

Finite and Infinite Games

Wed, 12 Jul 2023 00:00:00 GMT

This note is the twenty-sixth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

James Carse writes very simply, there are at least two kinds of games. Finite games are games which are played with the goal of winning. And, infinite games are games played with the goal of continuing the play.

Finite games are played with the intent to end the game, but infinite games are played to avoid the ending of the game. A finite game may not contain an infinite game, but an infinite game may contain many finite games.

And that's the summary of the book, Finite and Infinite Games, almost everything else is commentary. You might already know how I feel about play, in play fully I wrote about nurturing the time to play. But the line I've drawn has always been between telic and paratelic activities, things in the world which are serious and things in the world which are playful.

This book made me realize the extent to which most human activities can be conceptualized as a game. Our careers, our property, even our states can be thought to be finite games for control over power, recognition, wealth and other social imaginaries.

James reiterates that we are complicit in propagating the games we choose to play: He who must play, cannot play. It is interesting that many of the decisions we "must" make to compete (notice the language we use) in the world, are in fact choices which affirm the very decisions we feel compelled to make. Noticing our act of agency in making those decisions, is a powerful way to take back authority over which games we deem meaningful.

At a societal level, James highlights that the rules of finite games rest on mutual agreement between the players on what constitutes winning. In other words, winning in finite games is only useful insofar as the game is recognized by its audience as valid and fair. A finite game exists not in reality but in the minds of its observers, and what we desire isn't to win but to be remembered as winners.

This is related with some of my earlier writings on scripted societies. In that note, I said:

A large part of a country (or a community's) culture is learning the set of scripts that govern it.

In the language of finite games, I codified the identity of a group with the set of finite games that they choose to play. But James Carse draws another distinction here between culture and society. To exist according to a set of fixed games, is in and of itself a bigger game, the game of society, where the winning move is societal success.

But culture is different from society, because culture doesn't identify with a fixed set of finite games, but rather "always points towards the endlessly open", it evolves and looks to continue evolving.

A culture does not have a tradition, it is a tradition.

Culture is an infinite game. In that way culture is the home of the infinite player, who believes that only what can change can continue.

The joy of being an infinite player is to learn to start something we might not finish. But the challenge of being an infinite player is to figure out how to hold the serious in the playful, how to keep all finite games in a bigger infinite game. How can we joyfully subvert the finite games we play in society?

One thing's for sure, I'd much rather be an infinite player than a finite player. Perhaps you might feel so too.

~ Shan

Super Obsidian

Tue, 11 Jul 2023 00:00:00 GMT

This note is the twenty-fifth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

It looks like the AI powered tools are here, and the AI powered tools are here to stay. The good news is that all of us have been gifted the tools of hyper-productivity, to be able to create more impact than we could before.

The bad news is that LLM based infrastructure heavily favors cloud-based apps over local-first apps. This means that products like Notion have an intrinsic edge over products like Obsidian.

To attach a more concrete face to the problem, today I started thinking about how I might use AI to speed up my thinking and writing workflow in Obsidian. I've identified the following wishlist for a start.

Text generation - the ability to generate or continue text given a prompt
Text formatting - the ability to summarize, change tone, spell/grammar check
Related Notes - the ability to retrieve related notes given a specific note
Semantic Search - the ability to retrieve related notes given a query
Hybrid Search - the ability to combine fuzzy keyword search with semantic search
Question-Answer Search - the ability to ask questions to my entire set of notes
Text Classification - the ability to auto-tag, topic model and plot semantic graphs

In broader categories, it boils down to generation, editing, search and classification. It's an added bonus if the plugins also use local embeddings to preserve Obsidian's local-first ethos. (I explored this in self socratic dialogues!)

Then, I looked through every single AI-enabled plugin in the Obsidian Community Marketplace, to curate the most effective AI plugins I could use today. What I found wasn't super encouraging, lots of the Obsidian AI plugins are cumbersome, lack a significant set of features or just plain don't work.

Of all the ones I looked through, the only plugin I am considering using is obsidian-copilot and it is missing half of the wish list I laid out above. This is concerning because, if Obsidian fails to help people leverage the cognitive shortcuts enabled by LLMs, they will migrate to better integrated software like Notion. The increase in efficiency enabled by the features above is too significant to ignore.

Fortunately, from my experiments two weeks ago in self socratic dialogues, there doesn't seem to be any technical constraint preventing an elegant local embedding implementation for Obsidian. An embedding index running on my Macbook Pro 2016, runs fast enough for real-time semantic search, and indexes locally at about 100 files every 10 minutes.

There's definitely still space for anyone who wants to work on a be-all-end-all AI plugin for Obsidian, the space is very nascent. One of the reasons I was looking into this was because I wanted to build one myself - if you're interested in working on one, please reach out to me on twitter!

(I've attached a detailed overview of all the AI Obsidian tools I looked through to this note)

~ Shan

obsidian-ava

😢: I tried this plugin, and it seems to be broken currently, unfortunately.

This plugin is intended to be a fully functional AI assistant, baked directly into Obsidian. It's the highest ranked result, and has the greatest depth of features.

Features:

✅ Text generation
✅ Text formatting
✅ Related Notes
❌ Semantic Search
❌ Hybrid Search
✅ Question-Answer Search
❌ Text Classification

✨ Extras: Image Generation

chatgpt-md

This is mostly a wrapper around ChatGPT, and lets the user hold conversations similar to the ChatGPT UI. Not exactly what I'm looking for in an AI tool, I'd like some more specific features to the Obsidian experience.

Features:

✅ Text generation
✅ Text formatting
❌ Related Notes
❌ Semantic Search
❌ Hybrid Search
❌ Question-Answer Search
❌ Text Classification

obsidian-copilot

👍: I tried this plugin, and it works well. It does everything it claims to do, and it performs reasonably. I like it!

This is a plugin sitting in-between obsidian-ava and chatgpt-md. It uses a conversational ChatGPT-esque interface with a side-by-side window design that's very minimalist. One cool thing is that it uses a local embedding store, and only sends the required context to the LLM API of choice. This means that it is, in a sense, more privacy-preserving than the previous options.

Features:

✅ Text generation
✅ Text formatting
❌ Related Notes
❌ Semantic Search
❌ Hybrid Search
✅ Question-Answer Search: but only with a single file at a time
❌ Text Classification

✨ Extras: Local Embeddings

obsidian-ai-assistant

This plugin focuses more on the variety of integrated models rather than specific AI-assistance features for Obsidian writing. It supports text, image and audio generation, but in a fairly boilerplate manner.

Features:

✅ Text generation
✅ Text formatting: manual prompting
❌ Related Notes
❌ Semantic Search
❌ Hybrid Search
❌ Question-Answer Search
❌ Text Classification

✨ Extras: Audio Generation

auto-classifier

This is a specialized plugin that focuses on using ChatGPT to classify text into topics, to suggest frontmatter tags. Very cool stuff, would love to see more interesting versions of this with topic modeling or semantic graphs.

Features:

❌ Text generation
❌ Text formatting
❌ Related Notes
❌ Semantic Search
❌ Hybrid Search
❌ Question-Answer Search
✅ Text Classification

obsidian-gpt

This is also a specialized plugin, which focuses on question-answer. Many of the plugins above already include the features of this plugin.

Features:

❌ Text generation
❌ Text formatting
❌ Related Notes
❌ Semantic Search
❌ Hybrid Search
✅ Question-Answer Search
❌ Text Classification

obsidian-semantic-search

😢: I tried this plugin, and it seems to be broken currently, unfortunately.

This is a specialized plugin for semantic search using OpenAI embeddings. I like the idea of this plugin because it does one thing, and it does it well. The floating modal interface is simple and should just work.

Features:

❌ Text generation
❌ Text formatting
❌ Related Notes
✅ Semantic Search
❌ Hybrid Search
❌ Question-Answer Search
❌ Text Classification

ai-mentor and obsidian-gpt-liteinquirer-plugin and obsidian-ring-a-secretary

These are very similar to obsidian-copilot, both offer a ChatGPT-esque interface. One cool thing that ai-mentor does, is that it can simulate different mentorship figures (like a scientist).

Features:

✅ Text generation
✅ Text formatting utilities (to summarize, change tone, grammar check etc.)
❌ Related Notes
❌ Semantic Search
❌ Hybrid Search
❌ Question-Answer Search
❌ Text Classification

✨ Extras: Mentorship Simulation

obsidian-brain

This is another plugin specifically for question answering. This plugin also uses an additional custom docker container which runs a seperate python server responsible for indexing embeddings for the document.

Features:

❌ Text generation
❌ Text formatting
❌ Related Notes
❌ Semantic Search
❌ Hybrid Search
✅ Question-Answer Search
❌ Text Classification

silicon

This plugin uses OpenAI text embeddings to do related notes lookup.

Features:

❌ Text generation
❌ Text formatting
✅ Related Notes
❌ Semantic Search
❌ Hybrid Search
❌ Question-Answer Search
❌ Text Classification

Some others I think are similar enough to the options above that I won't elaborate: vault-chat

Most Days 2

Mon, 10 Jul 2023 00:00:00 GMT

This note is the twenty-fourth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

Welcome to another Monday, and I wish for a beautiful week ahead in your life. This letter is going to be another progress checkin, similar to most days, where I review the goals that I had set at the start of summer. We're now about a month into summer, and there are some big changes in my summer plans that I have to account for.

Most importantly, I started a new internship with Dynacyte this month (I talk more about this in homecoming), which means I have less time to focus on my goals. In general, it's become clearer to me that I don't have bandwidth for some of my goals, so I will be stripping some of them out.

Goal 1: I am really enjoying the 104 days of summer vacation experiment, and using writing as a tool to process my thoughts. I've managed to remain consistent with it, and I'm happy with my progress so far.

Goal 2: I think I definitely won't have the bandwidth to do this, so I will be dropping this goal. Instead I'll focus all my writing energy into 104 days of summer vacation and try to get out some higher quality letters as well.

Goal 3: As I expected, the India trip threw my routine into whack with regards to the 90 day transformation project. But, I've managed to claw my routine back this week, and I will write an update on the project tomorrow.

Goal 4: Progress on the Happily Ever After website has stalled, because of my internship. I'm not dropping the goal but I think I need to pay more attention to it.

Goal 5: I haven't started on it yet, like I mentioned in most days, but I will be starting on it now that I'm back from India.

Goal 6: I'm going to drop this goal for now, because I don't have bandwidth for it. Perhaps I'll consider picking it up when some of my other goals are cleared.

Goal 7 and 8: I wrote about my thoughts on networking recently in thoughtful networking. Building on that, I want to adopt a different approach to these goals. Instead of attending networking events, I'll try to be mindful with my existing connections instead. For a start.

Goal 9: I need to apply for more grants and programs now that I am back.

Goal 10: This is going well, we have our first meeting this Wednesday!

~ Shan

Fear Irrelevance

Sun, 09 Jul 2023 00:00:00 GMT

This note is the twenty-third letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

In Player Piano, Kurt Vonnegut paints a terrifyingly real picture of industrial capitalist America gone all in on automation, a world where there is nothing for most people to do. In Player Piano's world, humanity trades off agency and work, for a centrally planned but highly convenient form of life.

The book itself follows Dr. Paul Proteus, an engineer and manager working in the last few remaining jobs and grappling with the question of human irrelevance in a world of automation. It's a theme that hits close to home, especially with the current frenzy around AI, AI safety and alignment(see: Superintelligence).

In a particularly memorable scene, a doctor named Harrison expresses a distaste for machines, and when pressed on why, he asserts, "Machines are slaves and they compete with people".

Anybody that competes with slaves becomes a slave

And in another scene near the climax of the book, a revolutionary Paul proclaims,

The main business of humanity is to do a good job of being human beings, not to serve as appendages to machines, institutions, and systems.

Player Piano depicts a world, where on the surface everything had gone right. A supercomputer precisely controlled every aspect of production, and ensured everyone had access to life's creature comforts. Manufacturing was completely automated and resulted in no deaths, injury or menial labour of any sort. And for those who wanted to work, the Army or the Reclamation and Reconstruction Corps offered some solace.

Player Piano had no AI alignment problem, no existential risk, no major suppression of freedom of speech, and no dictatorial system, there was still a functioning democracy. And despite all that, the lack of meaningful work in Kurt Vonnegut's world still left people decidedly unhappy.

Because if a machine can do everything you do, and do it better, what's the point of doing anything at all. One answer might be a greater focus on paratelic experiences, but whether social structure could be reframed into a purely paratelic endeavor is still an open question.

I think this is the key takeaway for me from this book, even if we execute technological post-scarcity right we still have some major philosophical and social questions to answer. How do we live in the world where there is no pursuit of happiness, because everyone is given everything they need?

~ Shan

To Dance with Joy

Fri, 07 Jul 2023 00:00:00 GMT

This note is the twenty-second letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

I am happy today!

First things first, my results for the second year at university came out, and I performed far beyond my wildest expectations. I owe my partner and flatmates who supported me through the exam period a massive thank you!! I'm also grateful to all of the professors, mentors and course mates who made my second year so enriching 💗.

And thanks to our newest healer in Happily Ever After, I had a tarot card reading for the first time. I would like to share it with you as well, it's a reminder to dance with joy.

I'm not going to drag this letter on, I think it's time for me to relax and celebrate. If you'd like, join me and do a lil happy dance, but otherwise have an amazing day :)

~ Shan

Fault-tolerant Biology

Thu, 06 Jul 2023 00:00:00 GMT

This note is the twenty-first letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

Isn't it incredible how resilient biology is? All you need is water, air and sunlight to grow some cells and they will grow almost anywhere on the surface of Earth. Biology is fault-tolerant engineering, I've been getting a new found appreciation for that at Dynacyte.

We were talking about artificial meats earlier, and the biological processes required to culture these meat cells. Much of artificial protein research has been done with specialized labs and expensive scientific equipment like clean rooms and industry-standard bioreactors.

The problem here is that when we try to scale this research to mass-produce artificial proteins, we also need access to lab environments similar to pharmaceutical labs. Meanwhile the competitor to artificial protein, the chicken, can be grown in any old crusty henhouse. The economics of producing artificial protein in labs is tough to overcome.

This is because artificial meats are designed in the same reference frame as regulated pharmaceuticals, the only way to circumvent it is to start from radically different assumptions. And the correct assumption to have is that the wet-ware should be as fault-tolerant, as any other chicken.

Interestingly, it seems like fault-tolerance is something biology should have had by default. But it turns out that the trade-off multicellular organisms made many millennia ago, was to exchange some of that cellular fault tolerance for a greater level of complexity and interdependence.

This is why bacteria are fault-tolerant and easy to grow, but chicken cells are not - they rely on a complex network of immune cells, scaffolding and other "public services" to survive. Kind of like us, and the trade-offs we made for complex human civilization. I reckon not many of us would survive in a forest alone either.

TLDR, a difficult problem between us and commodified synthetic biology is the ability to engineer fault-tolerant biological systems from scratch, which I don't think we've cracked quite yet.

~ Shan

Wisdom of Elders

Wed, 05 Jul 2023 00:00:00 GMT

This note is the twentieth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

I hope the week has been treating you well so far. After my first day interning at Dynacyte, there is a buzz of excitement over something new, like having activated a dormant set of neurons in my head.

Two thoughts rest in my head. First, it's gratitude for having people to look up to. I think my resistance to authority figures has prevented me from finding many mentors since I've naturally gravitated towards working alone.

In that regard the team at Dynacyte consists of experienced engineers and scientists who I am so excited to learn from and build with. There's comfort in knowing that someone can extend guidance from a much larger corpus of life experience than I have available to me.

I've gotten used to being the end of line, for the past two summers I've been leading projects where there's no one for me to ask and I'm the person everyone asks. That's a lot of responsibility which becomes clearer now that I'm in this internship. It's just so nice to be able to focus on the technical details of a project without constantly having to also worry about the bigger integrative picture.

Perhaps, founder culture has undervalued the importance of mentorship by encouraging lots of us to disregard learning and growth under the mantra of "move fast and break things". Why learn from others, when you can start your own thing and figure it out, seems to be a fairly common entrepreneur's mindset.

But there's value in spending time to learn from others. Especially in technical fields, arrogance can be dangerous and lead to terrible outcomes like the recent OceanGate submersible implosion. As young engineers, we have a responsibility to learn and understand what we're doing, not just to build blindly for the sake of speed. So there's a lot of wisdom to be gained from listening and working in environments with people who have a lot more experience than us.

~ Shan

Homecoming

Tue, 04 Jul 2023 00:00:00 GMT

This note is the nineteenth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

I am terribly jet lagged today, so this will be a short one. My 10-day trip to India is over, and I am happy to finally be back home. It's funny how going to India has this time-dilation quality to it, the ten days feels like years but it was nice to spend some time with loved ones.

I'm also excited to get back to routine, there are a number of changes coming up. Tomorrow, I am starting an internship with Dynacyte Biosciences, working on modular, scalable and small bioreactors. This wasn't in my original summer 2023 plan, which means I'm going to have to scale back some of my goals to make extra time.

Regardless, the team at Dynacyte are inspiring, and the vision of "grow all the things" is something I would love to get behind. Biology is beautiful and there's a lot to learn. 🧬💗

In my recent reflections, notably in do what contributes and grieving otherselves, I mentioned the classic 20s struggle of figuring out what I should do. Dynacyte is an experiment on that front, and an opportunity to work with feedback from an ambitious team.

On an unrelated note, I hope you never have to deal with allergies on a plane. I wouldn't wish that on my worst enemy. I will be sleeping now, contently back in my own comfortable bed. Good night.

~ Shan

Thoughtful Networking

Mon, 03 Jul 2023 00:00:00 GMT

This note is the eighteenth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

I'm thinking about how I dislike networking. To me it feels like an unauthentic front for people to pursue their own selfish goals, perhaps due to the failures on the part of people who've 'networked' with me.

Conversations are beautiful, deep ways to share meaning with each other but networking boils it down to a transactional set of small-talk protocols. "How can I help you, but more importantly how can you help me?".

This pre-conception is one of the reasons I keep to smaller social circles with high-touch and meaningful friendships. The downside, is that I fall out of touch with people who are very cool but I don't have the bandwidth to form high-touch friendships with. After all, there is an upper limit to the number of quality friendships I can cognitively maintain.

Is there a way to be kind, and thoughtful with people without full friendships?

Serendipitously, I remembered something I'd seen in the rebootHQ discord server called Clay. Clay is a CRM designed to help people mindfully cultivate relationships, I think about it as a cognitive tool to increase the number of meaningful connections a person can juggle.

I set my account up a couple of days ago and I think it is a massive unlock for people who struggle with maintaining social relationships. Using Clay with the right mindset (being genuinely curious about the lives of amazing people who've crossed paths with me) is already helping me build meaningful relationships with people without the pressure of full friendships.

So if you're someone who wants to make meaningful connections, hates small-talk networking, but can't sustain any more friendships, here's what works for me now.

Find people who you're genuinely curious about, it's very important that you care about their journey. Filter out people who might just not be a good fit right now.
Focus on what you can give rather than what you can take.
Use Clay to help remember, and remind you to nurture these relationships that matter.
Be intentional and thoughtful, the goal is to be kind and support the people you've crossed paths with.

~ Shan

Meditations on Effective Altruism

Sun, 02 Jul 2023 00:00:00 GMT

This note is the seventeenth letter in the 104 days of summer vacation series. You can also follow the full twitter thread here, and leave any thoughts and comments that might come up!

Doing good is good, it makes us happy to be of service to others. So, it could be small or big but I hope that you get the opportunity to help someone else today. The principle of doing good is also the core of Effective Altruism, I wrote about some of their career advice yesterday in do what contributes.

In essence, Effective Altruism constructs an optimization problem out of the idea of maximizing the amount of good contributed to the world. Using metrics like QALYs (Quality Adjusted Life Years), the EA community advocates for a rational approach to select and participate in activities that are of highest value to the world.

This sounds perfectly reasonable. When the EA movement first started, it focused on improving living conditions in developing countries (mosquito nets being a prime example), where the impact of an additional dollar of investment has much greater returns as compared to other first-world problems.

Additionally, effective altruists like to work on neglected problems which cause great suffering, where an additional unit of work is more valuable than when allocated to an alternative cause which already has plenty of manpower. This rational, metric-driven approach leads directly to EA's brand as the evidence-based approach to doing good. The science of living a good life.

Things start to get concerning when we introduce the ideology of longtermism- a belief that the number of current humans is miniscule compared to all the future humans that might live if we did everything right. Longtermism says that we have a moral responsibility to protect and influence the long-term future. In recent years, EA ideologies and longtermism have become somewhat synonymous, with many EAs pivoting from helping developing nations to working on existential risks.

Since this pivot, there have been major changes in Effective Altruism discourse, like this move away from climate change because it doesn't threaten civilisation. Many EAs now focus on issues like pandemic prevention, biosecurity, and AI alignment due to the potential for civilization ending (x-risk) events.

This feels wrong. Not that AI alignment isn't important, but bearing the responsibility for of all future humanity can lead to reprehensible moral outcomes. Hal Triedman in Ineffective Altruism cites this interesting thought experiment:

a situation where the head of the CIA explains to the US president that they have credible evidence that somewhere in Germany, there is a lunatic who is working on a doomsday weapon and intends to use it to wipe out humanity, and that this lunatic has a one-in-a-million chance of succeeding. They have no further information on the identity or whereabouts of this lunatic. If the president has taken Bostrom’s argument to heart, and if he knows how to do the arithmetic, he may conclude that it is worthwhile conducting a full-scale nuclear assault on Germany to kill every single person within its borders.

I think the moral problem manifests when the tools of rationality are applied to the belief of longtermism. Taken alone, the EA approach to rational goodness and the optimistic belief of longtermism are both appealing to me. Put together, the idea that we should disregard impacts on the next 100 years because we are but infinitesimal, inconsequential specks in human history seems ridiculous. In some sense, EA methods applied to longtermism conclude that our primary moral responsibility is to not mess anything up.

Referring to Hal Triedman again,

Ineffective altruism eschews metrics, because “What does doing good look like?” should be a continuously-posed question rather than an optimization problem.

But this rebuttal doesn't satisfy me, it falls right back into the traps of scope neglect, which is the cognitive bias that rationality was supposed to get us out of. And yet in the words of Philosopher Karl Popper as quoted by Hal Triedman himself:

We must not argue that a certain social situation is a mere means to an end on the grounds that it is merely a transient historical situation. For all situations are transient. Similarly we must not argue that the misery of one generation may be considered as a mere means to the end of securing the lasting happiness of some later generation or generations; and this argument is improved neither by a high degree of promised happiness nor by a large number of generations profiting by it. All generations are transient. All have an equal right to be considered, but our immediate duties are undoubtedly to the present generation and to the next.

If rational longtermism leads to outcomes that I find morally incorrect, there are only three explanations for that.

The rational inference procedure is wrong (or ill suited for moral questions)
The prior assumptions of longtermism is wrong
The outcome is correct and I need to overcome my scope neglect and bite the bullet

I don't know yet about (1) and (2), but I'm not confident enough to accept (3) just yet.

~ Shan