routers eat the world; consilient software observer; crafting an alchemist's paradise. economic meta-learner. DMs open.
palo alto
Joined September 2020
- Tweets9.3K
- Following666
- Followers7.1K
- Likes9K
illiquid retweeted
Uma IA que monta o modelo de valuation em dois minutos não desintermedia o investidor, porque o modelo nunca foi a parte que determinava o retorno. O caminho tradicional na carreira ensina a planilha primeiro: back-office para demonstrar resiliência, investment banking ou sell-side para aprender a estruturar modelos de valuation e apresentações, e só depois o buy-side, onde se aprende a monetizar todo esse ferramental.
Eu comecei como estagiário da mesa proprietária de trading de um banco. Meu primeiro mentor não me ensinou a modelar. Ele me ensinou que existem no máximo três coisas numa ação que determinam se você ganha ou perde dinheiro.
Meu segundo mentor me desincentivou a fazer o CFA, mas me colocou para trabalhar numa biblioteca com mais de 4.000 livros, separados por tema. A cada semestre eu escolhia uma pilha sobre um tópico para me aprofundar, e conversava por horas com ele sobre o que estava lendo. Ele me ensinou a selecionar e a pensar criticamente. Um dos seus ensinamentos ele repetia sempre: "Capital não se aloca sozinho. Pessoas alocam recursos, sejam eles humanos ou financeiros."
Repare: uma ferramenta de IA que faz a parte braçal do modelo de valuation não desintermedia o investidor. Ela libera tempo. Se todo mundo tem o mesmo modelo em dois minutos, o modelo vira consenso e o preço já o carrega; o que sobra de diferencial é escolher os modelos mentais certos para pensar a empresa, isolar as três variáveis que importam e entender como elas se movem até a ideia virar dinheiro.
Abaixo, cinco livros que influenciaram minha formação. Nenhum deles ensina a montar uma planilha, e é por isso que continuam valendo depois que a IA passou a montá-la. Por que cada um, abaixo:
if you're into "financial models", pivot to ontologies. You need to think about firms as networks of complementary knowledge and know-how, not as a bunch of numbers on a spreadsheet. Accounting doesn't give you any view into the thing that actually matters (and will matter more and more going forward): how intangibles combine to create economic value.
Firms with economically valuable configurations of knowledge / know-how can deploy capital at attractive returns. Those with a differentiated ability to learn from deployment (accrue more economically valuable knowledge/know-how) absorb more capital over time.
Organisms / species with adaptive mutations absorb energy at high rates and use that energy more efficiently (become more selectively adaptive).
Are you getting it now?
illiquid retweeted
"The man who does not use computation cannot compete with the man who avails himself of its secrets. No hedge fund can price second order effects faster than a 200-megawatt Jane Street data center. No lone genius can produce a better language model than Anthropic."
@CooperVeit3 on the secretive, unborn market for compute:
arenamag.com/articles/counti…
illiquid retweeted
doing good on ur little excel model is not going to survive the activation of context complementarities
data sales is a beta game—alpha only exists in high entropy flows that you alone possess
why beta yourself when ai expands economies of scope on potential alpha?
from nber-
if you're into "financial models", pivot to ontologies. You need to think about firms as networks of complementary knowledge and know-how, not as a bunch of numbers on a spreadsheet. Accounting doesn't give you any view into the thing that actually matters (and will matter more and more going forward): how intangibles combine to create economic value.
Firms with economically valuable configurations of knowledge / know-how can deploy capital at attractive returns. Those with a differentiated ability to learn from deployment (accrue more economically valuable knowledge/know-how) absorb more capital over time.
Organisms / species with adaptive mutations absorb energy at high rates and use that energy more efficiently (become more selectively adaptive).
Are you getting it now?
illiquid retweeted
The future of capital allocation is about underwriting workflows, not just cashflows.
if you're into "financial models", pivot to ontologies. You need to think about firms as networks of complementary knowledge and know-how, not as a bunch of numbers on a spreadsheet. Accounting doesn't give you any view into the thing that actually matters (and will matter more and more going forward): how intangibles combine to create economic value.
Firms with economically valuable configurations of knowledge / know-how can deploy capital at attractive returns. Those with a differentiated ability to learn from deployment (accrue more economically valuable knowledge/know-how) absorb more capital over time.
Organisms / species with adaptive mutations absorb energy at high rates and use that energy more efficiently (become more selectively adaptive).
Are you getting it now?
illiquid retweeted
Real ones know.
if you're into "financial models", pivot to ontologies. You need to think about firms as networks of complementary knowledge and know-how, not as a bunch of numbers on a spreadsheet. Accounting doesn't give you any view into the thing that actually matters (and will matter more and more going forward): how intangibles combine to create economic value.
Firms with economically valuable configurations of knowledge / know-how can deploy capital at attractive returns. Those with a differentiated ability to learn from deployment (accrue more economically valuable knowledge/know-how) absorb more capital over time.
Organisms / species with adaptive mutations absorb energy at high rates and use that energy more efficiently (become more selectively adaptive).
Are you getting it now?
illiquid retweeted
Join me and @lefttailguy at 1p et xitter.cf/i/spaces/1pKkOXgnlQVKj
illiquid retweeted
The labs of course are on a rabid search for their next revenue S-curve. Which is why they need to brute force / accelerate deployment into high value verticals and "pace the frontier". Coding took them to $70B in ARR (OpenAI saying they missed estimates), but it is not an iron law that "better" coding / AI research models will put them on the trajectory to $1T ARR, and that's what they need to 10x+ their equity from here. To continue to retain (and attract) the best talent, the equity needs to plausibly 10x. For that, they need the meta-harness / learner for all economically valuable knowledge work, but there's a number of technical, political, and economic barriers that will make this far from easy.
Is Pacing the Frontier just a euphemism for Accelerating Vertical Deployment?
One of the big limitations labs have is a large opportunity cost on their compute. They have to constantly weigh dedicating capacity to inference and/or domain specific post-training vs. training the next best general frontier (coding) model.
If that impetus for large model training is reduced, can they start doing multiple vertical specific post-training runs in parallel and make it very difficult for vertical specific providers to compete?
So: to many parallel domain-specific post-training / distillation / inference loops.
I think this is more bearish software / VC application layer than it is compute in some sense. If the labs can focus on / dedicate more resources (compute, researchers, etc.) to vertical post-training (can even do this under the guise of aligning models better to certain domains), they should be able to serve domain specific models at much better cost-performance than application layer competitors (they have lower cost of capital, more scale, etc.).
There are still trust issues that will impede them from perhaps getting the best domain specific traces, but I think that extends to application layer businesses as well: the best end customers are actually looking to post train their own models (see: Kirkland and Ellis, Latham, Eli Lilly, Roche). So you have this tough game theory where the best customers work with bespoke providers (e.g, Palantir) and the long tail is totally ok using lab harnesses. So the application layer is left with the messy middle.
illiquid retweeted
edge routers eat the world meets intelligence agencies
we are all fighting for pole position
early but correct @lefttailguy @hypersoren @curious_vii
We discovered a sophisticated recording device that was hidden inside one of our offices. We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier. We referred the matter to the FBI, and we’re grateful to the agents and prosecutors who moved quickly to make an arrest. The FBI’s investigation is ongoing and we will continue to support their work.
At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data. The employee was terminated.
We take protecting our technology, data, and people seriously, and as a matter of national importance. Our security program is designed to protect against a wide range of threat actors, from insider risks to cybercriminals to nation states, and this case is an example of how AI can help address these sophisticated security threats.
illiquid retweeted
most ai applications today
-ai customer service
-legal contracts
-classification
look like control problems
> you imperfectly observe the environment, take an action, and get a reward
do you always need frontier AI to operate these controllers?
illiquid retweeted
AI x Finance
@GavinSBaker: AI Jesus
@JaredSleeper: Avenir, sharp on AI and software investing
@khemaridh: AI for the buy-side
@guangyurobert: Shortcut, AI agent for Excel
@aleksizy: ex-head of data science at Coatue
@FredaDuan: AI business models and economics
@buccocapital: tech investing and AI market takes
@patrick_oshag: Invest Like the Best
@TSOH_Investing: AI winners and losers
@didier_lopes: OpenBB, open-source finance plus AI
@AlphaSenseInc: AI market research
@CharlieZvible: ex-HF analyst at AlphaSense
@ShanuMathew93: AI power and data centers
@AgustinLebron3: ex-Jane Street, trading and decisions
@quantian1: quant markets, AI trades
@TMTLongShort: pod L/S view on AI and tech
@obsidiancap1: tech investor
@lefttailguy: software and AI infra
@thedatagap: AI, investing and data
@S_curvecap: tech L/S PM
@JaredKubin: AI and markets
@GrantStenger: ex-Jane Street and Numerai
@P_Remarks: earnings call takeaways
@pitdesi: fintech VC
@jbahrdestefano: fintech investor and writer
@mishhman: builds financial agents
@uncovered_alpha: ex-sell side and buy side on AI and stocks
illiquid retweeted
i will elaborate
economy trading below token value is a consequence of market failure in information markets
holders of valuable tokens maximize profits by vertically integrating / selling to one buyer not broadly selling tokens
token values fall as its distribution scales!
illiquid retweeted
YES! Absolutely. We have been working with societies and expert groups (such as @JACCJournals ) to create a virtuous cycle of knowledge gap identification --> knowledge creation. This pulse on the clinical knowledge requirements, paired with the evidence gaps uncovered by our engine is truly one of the projects @IAmSamFin and I have been most passionate about working on
Rough back of the envelope math: if 40% of US physicians are OpenEvidence DAUs (60%+ are registered users) and there's ~22 workdays a month, 42m queries in August implies ~5 queries per workday day per physician (Google is ~4.2 searches per day and ChatGPT is ~2.5 queries per day).
This kind of pulse on real-time clinical uncertainty is remarkable. Imagine you could now categorize those queries to create a map of the frontier of knowledge / know-how in medicine, and what's missing. By doing so, OpenEvidence is creating the closest thing to a universal ledger of medical problems and solutions.
OpenEvidence can see for which flavors of clinical uncertainty existing knowledge / know-how / solutions are most scant (or don't exist at all). They can then find and plug in that knowledge / know-how, create it by funding studies / initiatives, and eventually even discern the best heuristics / simulation environments to train the judgement of the next generation of human capital in the life sciences. You start to see how OpenEvidence's flywheel will start to spin a lot faster than the labs in terms of medical capabilities.
The general lesson here is that it's all about your product sensing and enumerating demand in some valuable subset of economic reality faster than competitors. In this case, OpenEvidence enumerates physician demand for medical knowledge / know-how that can help solve patient problems. They do this faster and with higher fidelity than anyone else. This is incredibly valuable.
illiquid retweeted
Bingo. 100%
AI types don’t understand this. They want to push investors out of Excel because they don’t fundamentally understand investors. Eventually that may happen.
But, for now, a quantitative surface is necessary.
Reliable Excel read/write changes everything, and by my evals Opus 5.5 first to hit threshold
I wonder what's next. I suspect one interesting endpoint is that the excel model becomes the quantitative surface of a much deeper research graph / self-constructing ontology. I could imagine something like every quantitative revenue key driver being hyperlinked to a graph of the intermediate qualitative assumptions that underpin it. Maybe analysts use some kind of ambient capture during their desktop and field research process, and then there's an agent-mediated live evidence stream (e.g, expert calls, filings, transcripts, podcasts, articles, tweets, etc.) that supports or contradicts the analyst's assumptions in real time. And based on that you could calibrate your confidence about various qualitative assumptions at any given point (as new evidence comes in) and the model could compute how that flows through to quantitative KDs.
Overtime, this surface / system would learn how new evidence shifts your priors and how accurately your updates actually end up representing economic reality. It could then start to understand how to weight different pieces of qualitative evidence!
Lots to think about.
illiquid retweeted
yes, it is—and @curious_vii is a thoroughly good and brilliant dude. If you're trying to figure out AI at an organizational level, I really can't think of anyone better to call. Always way ahead of the curve.
We met up for a beer in DC in 2023, and a few simple but very practical words from him at the time about about how to use superwhisper effectively has probably saved me 80 hours of my life from unnecessary work, so I can focus more on the things that matter.
illiquid retweeted
Concise read on:
Why @clavicular is right about Looks-Maxxing. I think if started a company akin to, or a sub-brand brand under @bryan_johnson's Blueprint, it would print money deservedly
If at the limit everything becomes about the right to access and the right to actuate on valuable economic opportunities, attractiveness confers a big advantage: I am quite literally 5x more likely to respond to and enumerate my economic problems to a pleasant Stacy all else equal. I am also much more likely to become a customer, because I'd rather have quarterly catch up calls with the pleasant Stacy. Therefore, by being pleasant and attractive, the pleasant Stacy grants the company she works for differentiated access to my demand and right to actuate on it downstream.
This is a thielian "very important truth that very few would like to publicly admit". This ability / class (~access generating human capital) will be massively repriced.
Has anyone scraped company Linkedin profile photos, asked model to classify attractiveness level, and correlated that with some measure of product differentiation? I know scraping Linkedin is notoriously hard, but request for research.
illiquid retweeted
"But to really understand Palantir, you have to take a step back and realize that it is, first and foremost, a deeply intellectual institution.”
arenamag.com/articles/forwar…
illiquid retweeted
Today is a big day - we're live right now and covering some developing stories:
- Alibaba's latest ai strategy roadmap
- Analyzing updates from the UN general assembly week as top ai leaders meet with heads of government
- China continues investigating DeepSeek and Moonshot for alleged leaks to Anthropic
Following this up with a really awesome slate of interviews:
- @jon_stokes author and CTO at symbolic.ai, discussing why everyone should maybe be skeptical of AI benchmarks
- @eddylazzarin GP at a16z crypto, talking about AI safety discourse and how productive conversations around open vs closed ai development is getting pushed to the side
- @Houda_nait Research Lead at OpenAI, on why she believes in p(bloom) > p(doom), and why personal agents are so exciting right now
- @gaganbiyani co-founder of Udemy & CEO of Maven, and @eriktorenberg GP at a16z, on the launch of the Horowitz Andreessen Academy
- @ByrneHobart writer of The Diff & GP at Anomaly Fund, and @lefttailguy will be joining us to speak on their Arena Mag piece “Forward Deployed”
- @ElliotGlazer lead Mathematician at Epoch AI the on Hodge Conjecture rumors and OpenAI’s claim of resolving 100+ longstanding open math problems
- @ajratner co-founder & CEO of Snorkel AI, talking about the >$300M raise and why expert data is the scarce input for agentic enterprise AI
- Andrei Georgescu CEO & co-founder of Vivodyne, on training world models of human biology on lab-grown tissue at datacenter scale
- Nikhil Aggarwal, co-founder and CEO of Corridor, on the company's $25 million raise and how they're leveraging AI to change the healthcare industry
UN MEETS ON AI | TRUMP SUPER INTELLIGENCE | ALIBABA AI STRATEGY xitter.cf/i/broadcasts/1vJpPNWEq…
illiquid retweeted
it seems the labs do not intuit hayek or
>what hayek didn’t write about
the incentives of his Man on the Spot to
-protect the information rights of
-and economize actuation
exclusive to owning and operating at the edge
Excited to hear @lefttailguy at 1p et!!
I will be returning to @MTSlive at 10am PT today with @ByrneHobart to discuss our @arenamag piece on Palantir, the Forward Deployed Model, institutional fitness in an age of accelerating machine intelligence, the bottleneck in converting tokens into economic value, and lab positioning in this world. Looking forward to it!
illiquid retweeted
Required reading from @arenamag on @PalantirTech. Join us in the battle against entropy and apathy.
arenamag.com/articles/forwar…