Investing @dragonfly There are no solutions, there are only trade-offs
Joined April 2023
- Tweets1.6K
- Following1.4K
- Followers2.9K
- Likes4.8K
AZ retweeted
finally breaking my twitter silence - I've joined @dragonfly as a partner in nyc
excited to work with this unreal squad:
@HadickM @hosseeb @tomhschmidt @contrarycap5 @0xelton @0xsudogm
backing the future of money, markets, and intelligence
AZ retweeted
We are releasing Dyna-2.1, the first Physical Agent that achieves reliable super long-horizon whole-body autonomy. It combines our brand-new semi-humanoid hardware with an agentic system built around Dyna-2 to handle ultra-long real-world workflows.
Here is an uncut footage of Dyna-2.1 completing an entire hour-long laundry room workflow, just like a human does.
AZ retweeted
There are a number of incredible low-margin consumer services that have massively grown their operating profit (and their valuation) through an in-product advertising business:
- Amazon: ~$80B
- Uber: ~$2B
- Doordash: ~$1B
- Instacart: ~$1B
It'll be interesting to see if this creates a modern form of the innovator's dilemma, i.e., enabling agentic use (which is unequivocally pro-customer) cuts deeply to their bottom line, creating opportunity for those that don't rely on ads for their operating profit (e.g., Shopify).
AZ retweeted
If you add up the equity appreciation in the last 2 yrs for SPCX, OAI, Ant it’s effectively the largest “stimulus” in US history, and it’s all concentrated in the top 1%. Very little benefit to the other 99%
We can apply the same logic to the market. For example…
1/ NEW: Josh Kushner is now worth an estimated $16.7 billion—more than triple his fortune a year ago.
The 41-year-old Thrive Capital founder rode bets on OpenAI, SpaceX and more to an eleven-digit fortune. Now he’s looking to buy a piece of the Lakers.
forbes.com/sites/giacomotogn…
AZ retweeted
Couple random thoughts...
Frontier labs appear to be 6-7 months ahead with internal models, and are reportedly seeing signs of accelerating their own cycle times, potentially pulling further away.
Open weights models are becoming highly capable and the range of tasks they can reliably accomplish is rapidly increasing the surface areas within enterprises that they can be used for
The vast majority of token spend is accruing to frontier model providers
The price per token and price per task for an increasing number of use cases is plummeting
AI potentially represents a China shock for white collar employment, except instead of concentrated losses in a handful of industries and geos its widespread throughout most companies and touches almost every geographic region.
Technological diffusion is about more than capability and cost. The spread of FDEs shows the challenge of scaling implementation.
Gross margins on inference are much healthier than people expected 12-18 months ago.
The AI industry is fundamentally making long term fixed dollar commitments and funding them via selling tokens in the spot market. Yes at certain levels of the stack certain parties are duration matched via contracted revenue, but the industry in aggregate is selling spot tokens against long term fixed costs.
The balance sheets of the core funding sources for the AI buildout remain healthy, if less so than a year or two ago. And increasingly third party capital is being mobilized in quantities never before seen.
Because the buildout is now financed out of risk capital and not internal funds, it is subject to the vicissitudes of human emotions, not simply math. A surplus of capital today at attractive prices does not imply an availability of any capital at any price tomorrow. This means there is path dependency risk from reflexivity and exogenous shocks.
AZ retweeted
Tokenization has been a hard-to-invest-in narrative because it intrinsically implies 2 competing forces:
1. Value creation at the offchain layer by cutting costs and friction
2. Those same cost cuts erode the margin the onchain layer could capture
Tokenization can win while the onchain layer barely accrues any value.
AZ retweeted
every project focused on acquiring users outside the CT crypto bubble is basically in up only mode right now.
slowly for a long time, then all at once.
lesson in there.
AZ retweeted
Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws:
• world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours,
• this human data scaling law implied a scaling law on never seen robot data,
• both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge
🧵
AZ retweeted
In some ways markets will trade alongside investor perception and consensus in the short and medium term.
If you take this to a logical end you can trade momentum by understanding how AI generated outputs will push certain investors.
You can also understand how AI generated outputs will create false narratives and counter trade them.
As always, pick your poison.
AZ retweeted
I rarely consume podcasts. I’ve stubbornly remained in the text world only to this point. But this was an interesting listen.
My takeaways are…
1) as I and others have repeatedly emphasized lately… above-comfort/baseline net worth matters less and less. There is nothing that Elon or Bezos can sensorily/physically experience that I cannot.
2) your contribution to “the timeline” is more important than how much money you have. The ability to project political power or social influence (in the direction you desire) is decoupling from financial asset accumulation.
3) I’ve written about primary, secondary, tertiary, quaternary economic activities and so on… we’ve entered quinary with the rise of content creation. My argument/observation has been that there is just not that much for humans to do the more economic development and technological progress takes place. We invent a machine that does the job of 4 men… 1 man runs the machine… but now the economy needs to absorb the other 3 by creating a new role/occupation. An optimist will say this is progress and humanity is better for it. A Luddite of some erudition will argue that the 3 new jobs created are more abstract and psychologically dangerous for the group over time because it moves further from “the real world” or the soil which is what we’re built to do. Fast forward a couple centuries and we have “value creation” driven by a few humans controlling the machines and then the mass hordes repurposed into content readers or content creators aka “posters” among which only the most proactive or unstable rise to prominence and readership to influence “the timeline.”
We are in the dystopia already.
My second conversation with @jeremygiffon. His first episode became one of the most popular we've ever done.
Since then he's become a friend I talk to every day, so this is a taste of one of those conversations.
We discuss:
- The billion dollar PDF
- Why billionaires have become subservient to the "poaster" class
- The philosophers who secretly shaped Silicon Valley
- Lessons from the last 18 months in private markets
- East v. West coast finance
- Buffett + beating the market
- and much more
Enjoy!
0:00 Intro
5:50 The Billion Dollar PDF
11:31 Algorithms and Power Laws
20:28 Peak Guy
31:19 Opting Out of the Timeline
36:14 AI and White-Collar Jobs
43:31 The Next Era of Finance
53:56 The New Economics of Software
1:03:22 Underwriting Emerging Managers
1:18:17 Silicon Valley’s Hidden Philosophy
AZ retweeted
I’m capitalist af but for once I’m with Jason on this. System is broken.
Yes kids nowadays are disproportionately coddled pussies…but there are real structural issues and the fact that a kid born in America has a better opportunity set than anyone born in any country in history doesn’t mean they are immune to basic human tendencies to compare themselves relative to the people around them and thanks to a (somewhat) broken system structure people born twenty years before them have managed to accumulate access to finite goods that have been artificially supply constrained.
Crashing housing would be a net societal good. It just is not politically feasible until enough boomers die. And the immediate impact to consumer demand would likely be catastrophic as demonstrated by Chinas ongoing malaise.
The job market is lopsided with older gens refusing to retire from cushy jobs that leverage constant influx of younger labor.
Education has become over saturated and lower ROIC than before but given how rich as a country the U.S. is the median W2 knowledge work wage is still superior to most of the self-directed entrepreneurial options a non-talented low-agency graduate might be offered.
We also just rug-pulled the highest agency gen-Z who dedicated the last decade towards breaking into tech only to tell them coding is no longer a valuable skill thanks to AI.
This all will change fairly soon. But to tell the kids there is nothing wrong is a lie and the wrong approach.
The first question we should be asking is how do we cultivate a new cohort of mass employers who are comfortable building new cities for young employees like Starbase Texas to solve supply-demand from first principles.
The second question is how can we solve supply with AI and robotics and then message it properly so that younger gens are simultaneously optimistic but also not complacent.
1. Stop building homes — create an unnecessary boomer bump (bonus: keep real estate taxes for homes purchased decades ago unfairly low)
2. Brainwash kids that college degrees always pay off — raise the price by 5-10x over two generations
3. Freeze the minimum wage for decades
4. Let the teachers unions control primary education for their own benefit while test scores plummet
5. Stack the tax system to benefit the rich
6. Blame the youth for being lazy 🙄 !
SCENE! 🎬
AZ retweeted
The smartphone really revolutionized the waiting room. Oh I get 20 minutes to look at my phone? Don’t mind if I do. The only problem is sometimes when I’m looking at my phone at home my home starts to feel like a waiting room and I realize it is and I’m just waiting to die
AZ retweeted
There are two broad ways this can work:
1. You divide this 5% over all US households, handing each a direct stake.
2. You give the stake directly to the government.
(1) is fine. (2) is probably ruinous, akin to inviting rats to live and reproduce in the walls of your house.
AZ retweeted
It’s remarkable how much of life is about finding something to work on that gives you energy and surrounding yourself with people you love and then ruthlessly and militantly blocking out every single thing that gets in the way of maximizing your time spent on those two things
AZ retweeted
the anthropic co-founder jack clark advice that stuck with me:
read the primary material. not the summary. not what the ai said about it. the actual thing.
form your own opinion first. then ask the model. never the other way around.
keep practices in your life where it’s just you against the world ~ a sport, an instrument, reading, building something with your hands. spaces where the algorithm can’t mediate what you learn about yourself.
and don’t defer to AI even when it’s usually right. especially then, actually. that’s precisely when the habit forms.
the people who won’t get eaten by this moment are the ones who stayed hard to replace. not because they avoided the tools but because they kept the parts of thinking that make the tools worth using.
AZ retweeted
i don’t like the attempt from some insiders to paint the zcash bug disclosure as “heroic”
it’s not heroic. it’s a systemic problem
yes they paid a smart researcher to find bugs and he did. that’s great
but the issue isn’t that the bug existed
the issue is that zcash, unlike almost any other cryptocurrency, enables a unique class of bugs, where if they’re exploited no one would know
this unique class still exists. the fact that they fixed this specific bug is immaterial. mythos could find 8 others. and then mythos 2
bugs can exist in all cryptocurrencies. they can exist in btc or eth or sol. but if someone exploits them we will IMMEDIATELY KNOW and limit the damage
with zcash, when someone secretly finds a bug and exploits it, WE WILL NOT KNOW
imagine if the kelpdao hack was exploited but it somehow magically wasn’t visible onchain. the attacker stole 300m but no one would know. and aave and other apps would assume everything is fine not knowing that they secretly have a hole in the balance sheet. how much wider would the damage have spread by the time we found out?
this is the real issue. not a specific bug. the systemic vulnerability: if a hack happens, we won’t know until much later
and a lot of zcashers have been trying to downplay the severity of this fundamental issue for years (including yesterday when this was first disclosed but played down)
regardless of this specific bug, i don’t think zcash is a safe place to store meaningful wealth long-term, until the design fundamentally changes
i wish zcashers were more open about that. that would’ve been heroic
AZ retweeted
Maxi/mini takes on AI:
Maxi-
1) Cybersecurity will be an enormous consumer of tokens. All public-facing code needs regular, in-depth pen testing.
2) We are barely scratching the surface in code. Enterprises will have tens of thousands of apps, dynamically generated apps, per user/per customer apps, etc.
3) Enterprise adoption is still nascent. Leading companies are spending 10x+ as much on tokens as laggard companies.
4) Video/image understanding/generation will be huge drivers of token spend- as will anything to do with operating IRL/physics.
Mini-
1) Many use cases (consumer chat, various enterprise "agent" use cases) are close to intelligence-saturated. Intelligence saturation will render many major use cases today dirt cheap in a few years. See @StatueofIBBertY's posts on this topic.
2) Enterprise spend rationalization is real and imposed spend ceilings may be durable. As costs decrease/efficiency increases token volumes will go up, but budgets may not.
3) The global compute "shortage" is driven by capacity hoarding, not inference volumes per se (yes individual companies are inference compute constrained, to be clear). "Enterprise" adoption probably won't be fast enough to keep up with capex, we need prosumer-y use cases like chat/code that are very responsive to model quality increases. If we go 8-12 months without those, funding for training cools even slightly and hyperscalers like Meta pivot to being neoclouds, we could easily find ourselves in a compute glut.
4) No one knows what % of frontier model API spending is from cleverly hidden distillation attacks (or partly motivated thereby). Many, many deep-pocketed actors are motivated to do this at scale.
In general, I don't know if there's ever been a cycle where there are so many considerations on both sides that have order-of-magnitude-sized impacts on how supply/demand work out for tokens, infra, etc.
What a time to be alive.