🌟 Hey there! This is a subscriber-only edition of our premium newsletter designed to make you a better investor and technologist. Members get access to the strategies, tactics, and wisdom of exceptional investors and founders.

Friends, 

My correspondence with Reid Hoffman has been one of my most enjoyable sources of new information, original frameworks, and enduring wisdom on the craft of investing and technology’s frontier of the year. I’m excited to continue the conversation in 2024. 

In this edition of “Letters to a Young Investor,” Reid and I trade emails on “Signals and Mirages” in venture capital. We discuss the chatbot hype cycle of 2016, why ChatGPT was different, open-source foundation models, and AI’s promising future.

We’re sharing a generous preview with all readers, but to access the full conversation – and many others – become a member of our premium newsletter, Generalist+. For just $22 per month, you’ll unlock unique series like this one and gain exclusive access to strategies and insights built to make you a better investor and thinker. 

(If you’re already a member, thank you for supporting us!)

*In case you missed them, here are the first and second correspondences between Reid and me.

Mario’s letter

Subject: Signals and mirages
From: Mario Gabriele
To: Reid Hoffman
Date: Friday, December 22 2023 at 4:50 PM ET

Hi Reid, 

A hearty winter hello from the New Jersey shoreline. I’m celebrating the holidays here with family, enjoying long, sparse beach walks, reading by the fire, and watching the occasional storm swell the surrounding marshlands. Wherever you are at the moment, I hope you have time for some rest, recuperation, and games :) 

I’ve been turning over a question on some of my shoreside walks. The trite way to phrase it is to say something like: how can investors separate signal from noise? That captures the thrust of what I’m thinking about but misses some color. It strikes me that much of the time, the difficulty for an investor is not wading through “noise” – which feels trivial and easily ignorable – but trying to determine whether the persuasive reality presented to you is, in fact, a mirage. It may look full, rich, and real – but time will show you it isn’t. 

To take my language out of the aether and into the tangible, maybe we can turn to a practical example. Given your expertise, I thought it might be especially fun to talk about this in the context of artificial intelligence. 

My first taste of venture capital arrived in the summer of 2016 when I was lucky to intern at Red Sea Ventures, a firm in New York City. One of the things I remember from that pitch-frenzied summer was how scaldingly hot chatbots were. They seemed to be showing up in everything! Social experiences, financial services, weather apps, enterprise products – everything was being reimagined through a conversational interface powered by relatively basic AI. For a time, technology’s future looked like a series of DMs, replicated across a thousand walled gardens.

With the benefit of hindsight, this “chatbot moment” was a mirage. It presaged something interesting – more powerful AI – but didn’t produce meaningful startups. Instead, chatbots were featurized and added to existing products without attaining the ubiquity they threatened to. 

I imagine you felt this mini-revolution many times more keenly than an intern in a Flatiron WeWork did. What do you remember of it? Did it feel like a mirage to you then, or did it persuade you? Across your investing career, which mirages seemed most promising as they were gaining prominence? Equally, which were simple for you to swerve? To the extent you’ve built a process to identify interesting trends and test their fidelity, I’d be grateful to hear about it.

Compared to the chatbot buzz of the mid-2010s, the current AI revolution feels radically different. Naturally, though, many aspects may not pan out as anticipated. Indeed, there’s considerable discussion among venture capitalists about where the value of this technology will accrue. Will it be the application layer or the foundation models? Will startups benefit, or will the big simply get bigger? Each potential opportunity could prove to be a mirage. 

I’d be interested to hear how you and the team at Greylock have set about analyzing this renaissance. From what I can tell, the firm seems to see commercial promise in several areas. You and Saam were kind enough to share your “Copilot for everything” thesis in our “What to Watch in AI” series. Saam later highlighted Abnormal Security as a promising company in the space. Both appear to be app layer plays. Meanwhile, you co-founded Inflection AI with Mustafa Suleyman; Inflection is bringing personal AI to the world, powered by its own foundation models. (I have had some great chats with Pi!) 

How did you and your colleagues evaluate the different aspects of this movement? What gives you the conviction that these are not mirages? Is that even the right question? 

Ultimately, I’d be glad to hear how you sift through different, plausible futures and find the thread you believe the world will follow. As always, thank you for generously sharing these stories with me! It has been one of the great gifts of my year. 

A very happy holiday, 

Mario 

Reid’s response

Subject: Signals and mirages
From: Reid Hoffman
To: Mario Gabriele
Date: Tuesday, January 9 2024 at 6:22 PM PT

Happy 2024, Mario!

Not starting the new year with small thoughts are you? :)

Your use of “mirage” is fitting and I think captures why the average venture firm typically only returns capital, let alone a risk-adjusted return. The movie Glengarry Glen Ross comes to mind— first place, a Cadillac; second place, steak knives; third place, fired 🫠.

Perhaps it’s clearest to talk about mirages as a process of separating illusions from reality. As a rough rule of thumb, let’s say I find that we can filter 70% of the “plausible realities” through rigorous analysis and diligence. After some thought and phone calls, we’ll find things like the foundational technology needed is just not ready, the GTM or another strategic choice is flawed, or don’t believe it can gain enough operating leverage (even) at scale.

That leaves 30% of “attractive, plausible realities” to work with. This category fundamentally has remaining variables that we can’t fully reason our way through: it requires estimating odds and placing a basic bet on the future. So let’s say that ⅚ of that remaining 30% will reveal themselves to have some fatal flaw that can’t be overcome.

Then we’re ~left with ⅙, or 5%, of the original “plausible realities” for our potential winners: As the non-mirages. In our first email exchange, I shared the story of Greylock’s Airbnb investment and my dialogue with David Sze. David and I both definitely saw Airbnb as part of that 30%. But for him, the accumulation of challenges from local regulatory complexity, liability overhang, unions, etc were too much, and it didn’t make it into his 5%. My theory of the game was that the founders’ plans for confronting these obstacles were credible and that they were scrappy enough to persistently iterate their way to success.

To capitalize on your word choice of ‘mirage,’ perhaps I can take this further by adding the notion of an ‘oasis.’ Airbnb is an example of what success looks like when a bet in one of those 5% hits: an oasis. As we investors wander the sands of the technology landscape, we seek to sidestep mirages, and replenish in the oases that keep us going.

My first startup, SocialNet, way back in 1997, is an example of a mirage. The vision of the internet as a vibrant medium for rich matching of roommates, dating etc was very attractive and plausible. But with hindsight, it’s clear that there were some fatal flaws that went beyond simply timing: mainly that users are expensive to acquire but will still churn in a few months. In the success state, they cancel because they got what they wanted. In failure, they blame the service (instead of how they utilized it) and cancel.

That’s not to say there haven’t been successes in this space. After all these years, Match.com remains a significant business. But ultimately, technology (mobile smartphones) led to the advent of swiping as the dominant sorting mechanism for the masses. Craigslist, with its different, barebones monetization model, was able to dominate the other types of matching for years. So it goes.

I do remember the chatbot wave in 2016; though, I’d characterize it as a surge moment, since it’s actually been a decades-long effort. 

Chatbots are powerful because language is powerful. Humans are social creatures; we’re language creatures. We rely upon language for all kinds of essential things: judging intelligence, forming tribes/companionship, and fundamentally it’s how we have coordinated to achieve our compounded, exponential progress. To interact with machines in our preferred way, the promise of them acting as agents on our behalf has been much dreamed of.

While I believe that 2024 and 2025 will be remembered as the arrival of the first truly indispensable user agents, we should remember that this dates back to at least the late 1960s when ELIZA was tried as a therapy chatbot at MIT.

Back to 2016, though. I think we were impressed because those chatbots finally were functional. In fact, one of my partners at Greylock, Seth Rosenberg, was the PM on the first commercial Messenger bot that Facebook deployed with KLM. There was a lot of excitement around how these models would transform corporate communications. It was relatively straightforward to see how things like customer service would be transformed. 

This chatbot wave preceded our investment in 2019 in Cresta. We saw the opportunity for AI to transform the way humans worked through a combination of automation and augmentation. Cresta set out to do this through creating a new expertise management platform that they’d start by deploying in enterprise contact centers as a way of riding the growth in conversation commerce.

Still, a few years later, in November 2022, when OpenAI unveiled ChatGPT, it felt like a revelation. ChatGPT’s powers to synthesize information from different areas and fluidly respond to the user were truly powerful – but the reaction was even more outsized than the utility. I believe that was because it was not just about functionality but about how crisply ChatGPT summoned our vision of an entirely new horizon opening up. It marked a significant shift in AI development and human empowerment...