Return

Return to the Self and Other Interests

for i in x: 
  I've been wandering a lot.
  It's all somewhat connected, at the end of the day.
    return

The Things You Return To

There’s this essay by a guy named Sarvasv I skimmed called What You Return To.

The author is getting older, finding himself in an eternal loop, and reverting to the same old habits, people, and interests. Every time he goes too far off trail, he finds himself snapping back in an almost inescapable loop back to his true self.

This is something that has been more and more clear to me as I age. The things you accept as who you are. The things you can’t really run away from. But also the strengths that you can win with. 

I subliminally and explicitly built Delta ∆ for this reason. It stemmed from me manually extracting the “things I returned to” from my essays. These things, I found, were helpful for me to understand myself. 

Some of these things are easy to interpret and sometimes a cigar is just a cigar. But a lot of the time, it’s really difficult to disentangle why we are naturally gravitated towards certain interests, work, aesthetics, etc. It’s at its core something so raw, so human, and instinctual (LLMs have the same problem).

I’ve Been Wandering

Come to think about it, I’ve also been pretty promiscuous about my work, albeit intentionally. This is my form of multi-start optimization as a way to prevent falling into a local minimum. Creativity and determinism are incompatible. Wandering into the unknown is a way to uncover unknown-unknowns. Periods of creativity before execution. Calm before the storm. But at this point, I’m done. I’m done moving. I’m done changing. I’m ready to stick it here and do me for some time. 

So let’s recount the last three or so months. I’ve lived in North Carolina, Chicago, and Berkeley. I drove across the nation, seeing friends in Atlanta, Kansas City, and sleeping in the occasional Parking Lot (Thanks University of Utah Hospital!)

Shaper Capital

I learned a lot from Shaper Capital. It was the best job I’ve ever had.

“Shaper was built on the philosophy of Shaperism, the belief that individuals can shape the world for the better. Too many people are too quick to accept perceived constraints and don’t deeply believe they can change the world; they view outcomes as luck or destiny. Yet much of the world we live in was built by shapers who believed in exceptionalism and believed that they could change the future — and worked relentlessly to shape it.” 

I worked mostly across a few of the portfolio companies. I got to see the mechanics of companies built up from 0 employees, acquisitions, financings, products designed, cultures built, sales ramped, and other lessons.

Of the lessons I learned, there are several that I think about a lot.

1. The one that is the most important to me is that I left a better person. The people I worked with proved that you can win and be nice simultaneously, which is inspiring. Shaper purposefully hired people that are “smart, nice, and get things done.”

2. Velocity. Compounding wins and speed in software breed success. Hustle as a Strategy.

3. Culture is really important and potentially the most important feature of a company. Most things are downstream of company culture and culture is downstream from the team.

4. I learned a lot about software and the middleware playbook. I didn’t know a lot about middleware, but I left having a strong interest in the role it will play in the future we are entering.

5. I learned about the “Venture Creation Fund” model, reminiscent of my time in biotech and comparable firms like ARCH, a model that remains interesting to me (likely more to come). 

Middleware for AI

I’ve been interested in AI since sophomore year of college. Although I was way too naive to understand what was truly coming. I had some pretty fundamental experiences that showed me that AI is just sexy statistical modeling on top of data. Data is extremely fragmented in the world. By unlocking these silos of data, you power a more intelligent world. The idea of data currencies specifically captivated me, which I will get into later.

My Time in NC, Chi, and Berkeley (SEE APPENDIX)

Recent Projects and Interests

Like Sarvasv, in some ways I’ve gone off trail recently. In other ways, I’ve returned to my natural state. I’ve looked at each independently, but through this exercise I’d like to decompose each, understand what it is about it I find interesting, and then find where they overlap.

Data Currencies

Reflecting upon my time at Shaper, one of the features of a middleware business that was the most powerful to me was the data currency angle. 

See the below excerpt from one of Travis’ essays.

• Datavant — the Datavant key is used between parties in healthcare as a common patient identifier.

• LiveRamp — when two parties in the marketing industry want to exchange data about a consumer, they use LiveRamp IDs as the unique identifier.

Data currencies create powerful network effects and build the infrastructure for individual software point solutions that will define the age of AI. You can imagine that data is the roads that the cars (AI) drive on. Cars grow old, there’s a lot of competition/limited moats. You want to be the person who owns the roads and takes tolls. 

Necessary Conditions for a Data Currency

Why doesn’t every data business use a data currency? There must be some necessary features in an industry, the companies, and the data that lend well into a data currency model. If we can properly pin down those characteristics, maybe we can search for opportunities.

Travis makes several important points in Introducing Shaper Capital. I want to focus on 1 and 2 as I think through this.

1. Businesses that solve data fragmentation challenges are needed in every industry.

2. Middleware businesses thrive when: i) data is extremely valuable, ii) data is extremely fragmented, and iii) applications using data are extremely fragmented.

3. In an age of AI, all three of these conditions are more true than ever — creating more and more value for companies that can help organizations navigate this complexity.

4. I expect nearly every industry will generate a $1+ billion middleware company this decade (and some substantially bigger).

Timing

What was it about LiveRamp, Datavant? Why did they happen when they did? Where is the Datavant for law? Was the industry just not working on it? Is the industry not ready? Is this an inefficient market opportunity?

LiveRamp took off in the early 2010s, probably while I was picking my nose in middle school. Isn’t that approximately when digital advertising took off?

Datavant took off seemingly immediately after its founding in 2017, coinciding with the explosion of health data, regulations on patient privacy, and use of real world evidence. But could Datavant have worked in the 2000s as well? I’m not sure.

These are industries that were once “off web”, that experienced a very real shift, bringing data online to map to a single person. That data was valuable and connecting it made it more valuable. Then, applications came online to build on the data, which made the data more valuable. AI makes the data even more valuable.

Key Question: What industries have recently “come online,” and can we develop a thesis around which is well-positioned for the next technological shift?

Type of Data

Notably for these two instances, the data was mapped per person. Why is that? Is that just a coincidence? Many rows of data tied to one key seems structurally clean. Ad companies target people. Healthcare companies need the full information per person. For healthcare, the anonymization aspect likely lends well into the model.

It may not need to be a key per person but likely some important singular thing that is tracked. A good heuristic here would be helpful at some point. 

It may also be true that the industry has to be ready for this shift and that they must get value from this connected data and be incentivized to push for it. For healthcare and ad data, this is clear. Better insights to target consumers better and better serve patients.

Key Question: What data characteristics lend well into the data currency model?

Delta ∆

Writing is a projection of the mind. Since LMs now understand text, they understand us. I won’t go deep here as it was the subject of my last essay. Text is a much deeper form of data that reveals millions of tangled semantic relationships, mapping abstract concepts in neural nets in a VERY real way.

I have a general riff that people are too entranced with next-word prediction in LLMs, and if you disentangle GenAI in next-word prediction and the semantic understanding that powers it, the latter is significantly more important. When you take two, three, n… blocks of text and can disentangle them into the most fundamental differences mathematically with vector representations, you have a lot of opportunity.

The initial use case here is obvious: personality tests that go further than just bucketing you into 16 categories and being able to help you understand who you are and how that has changed over time.

Application to Law

“Jurisprudence is the philosophy and theory of law. It is concerned primarily with both what law is and what it ought to be.”

So much of law is semantic. It is a tug of war between two opposing sides. Law is moral and law is not black and white at the highest levels. In fact, that is a feature, not a bug. Our moral fabric is human and it changes over time.

Last Sunday, I randomly joined a Stanford Law School hackathon, where people came from all around the world in teams of 4-5 lawyers and engineers to present what ideas they’ve been working on. I came with nothing but the above intuition and ended up winning the Dispute Resolution Track with the below pitch.

Mariko AI

Linked Pitch Deck.

Through the hackathon, I spoke with an amazing California Judge, a few CEOs of legal-tech companies, and the head of AI at one of the big law firms. The perspectives I gained affirmed my instincts on the potential for AIs massive impact on how law is practiced and grounded my interests in practical applications which legal professionals would find value.

Problem: Law is an industry built around disagreement. This brings out the worst in people. Simple disagreements between parties in mediations can escalate, leading to millions in avoidable legal costs.

Solution: Mariko is a background filter on communication that detects if an email or zoom conversation is escalating and steers the person communicating back on track. 

The tool could ingest the formal case briefings as context to see how opposing counsel sees an issue and if your response falls too far out of bounds.

Between Delta ∆ and Mariko AI, the common thread discovered through customer research is that people fail to see how others' perspectives are different from inside their own mind.

Ultimately, it is solving for human blind spots. We live in our own minds.

If you can differentiate two perspectives, you can then do 3,4,n to the scale of cultures and generations and quantify the google trends of ideas and the overton windows shifts, such as with how ideas have changed for the last 200 years in the House of Hansards Archive.

RAG

The way data is stored is changing in AI. I don’t think that it is super interesting as a better search method and there are people way smarter than I working on it, but I think it changes the game in important ways.

Especially in the legal context, where you can’t rely on a probabilistic vector prediction. You need documents and clauses based in fact and historically documented truth.

So the result is companies building internal document systems such as this. However, privacy is extremely important. And this doesn’t solve it. It actually makes it very difficult to see who has access to a particular document.

What you should expect is each law firm having a documented internal RAG system, but needing to share these documents externally. According to the head of AI & innovation at one of the largest law firms that I spoke with, doing this in a compliant way is a big challenge.

The New Age of Software - Defined by Intelligence

"If the 1980s were about quality and the 1990s were about re-engineering, then the 2000s will be about velocity." 

- Bill Gates, 1999 [LINK]

He also called this tech in a company “Using a Digital Nervous System.” He was extremely correct. Why? What is the prescription in 2025?

Ahhh, that makes sense. Gates called the internet the digital nervous system of a company. 

In 2025, we are building the brain of a company.

"Anthropic CEO says future of AI is a hive-mind with a corporate structure"

It’s All Somewhat Interconnected, at the End of the Day

AI changes rapidly, and that changes the middleware playing field, but people will need their data connected regardless. My instincts tell me there’s a way to combine my interests in data currencies, vector embeddings, middleware, and machine learning into a way for companies to connect their data inwards.

Essentially, these internal systems are all creating “single-player mode” data aggregation approaches. However, they will still need to be connected, at the end of the day for the true value of AI to flourish. RAG will not just be valuable for search. It will be valuable for evaluating meaning between vector embeddings.

This will not just be a problem for a single industry. It will be inter-industry and it will require extreme privacy protection and access gating.

(Still looking for concrete ways to solve a problem by building something here.)

–APPENDIX–

Chicago

I’ve spent some time in Chicago! I’ve been far from home for so long. Most people are still themselves, and it’s always great to have people you can pick back up like nothing’s changed.

Chicago is a beautiful city. It is almost too good of a city. All of my friends are there. The summer is great. It’s on the water. It’s well spaced out. People are nice. People see you as an end in yourself. 

There are a few reasons that I am unfortunately incompatible with Chicago right now. 

1. Chicago has no ambition. It is the average of everything. It is literally the “mid-west.” Chicago feels like where people go at 22 years old to settle down.

2. Chicago has no technology. Which is something I am unfortunately obsessed with. Everyone is working in finance, industrials, marketing 

North Carolina

I wasn’t only in NC for work. I honestly liked it. I lived in Cary, NC, which is still hilarious to me given I grew up in Cary, IL and always only heard of this place. I should have branched out more. But at the same time, everyone was 40 with kids, so not sure they would have wanted to talk to me anyways. People from Cary called it “Container Area for Relocated Yankees.”

There was Raleigh and Durham, but I never really got around to them except for a few visits. Lastly, it gets too damn hot.

Berkeley (and SF by association)

I’ve only been here for a few weeks now, but The Bay is actually more extreme than I thought it would be. A friend calls it chaotic and creative. That’s probably pretty close.

Whereas people in New York work really hard for jobs they don’t care about to accumulate material possessions and are more “work to live,” Bay people work less but really care about their work and many don’t care much for material possessions. People in NYC were very transactional. Here, less so.

People are really smart here, but at the same time have no common sense and some people are really odd. Growing up in the midwest suburbs, I felt like it was all a giant homogeneous urban sprawl and everyone was a suburban clone without much distinction. The Bay is the most heterogeneous place I’ve ever been. It’s still mimetic in many ways, but people here are defined by their differences, and they push those differences to the extreme.

I’ve always loved ideas. I loved going to random speakers in university, and they have a lot of stuff like that here. It’s actually the only place I’ve ever been where I’ve fit in in that way. One of the first days I was here, I showed up at a robotics/hardware meetup and an event where.

This openness and acceptance on a mass scale leads to all sorts of new, innovative things, but also a lot of delusion, and I spend a lot of my time trying to filter the signal from the noise. Another side effect of the openness is you can just do a lot of things and just walk into places and people don’t really ever question it.

I was at a house (supposedly) funded by Sam Bankman-Fried, I’ve been attending lectures at the Simons Institute for the Theory of Computing, and I will be at Stanford Law School tomorrow where anyone can register. There ended up being a homeless guy sleeping on the second floor of Simons, but he was the smartest hobo ever and showed the campus police his alibi that he had registered for the seminar haha.

Things like the above combined with giving 20 year old nerds $10s of millions of dollars turns the West Coast into a funky place.