432 Photos and videos
weekend project: tiktok but every feed is a small policy generated at runtime, rewarded by how long its scrolled quickquack.sdan.io
the best policies get served more with some randomness; kept optimizing for scroll time against a learned embedding of the user
right now it serves videos from wikimedia, the inspiration was akin to how grok llm retrieves and ranks a feed per user-- but what if we can do this with REINFORCE and improve the feed forever
2
41
3,395
i made a realtime analytics platform to visualize everyone going to my web apps
using cobe.js and durable objects
pingpong.sdan.io
1
40
3,495
this project started almost six years ago on a quest to replace google analytics.
a few months ago i revived it to add the realtime globe you see when you click the green dot in sdan.io at the top
grateful for all the cf pops sdan.io/projects/rapid-analy…
7
931
we have something reminiscing superintelligence yet interpreting diffusion of agi across the global economy is pretty hard
my hope is we can answer this simply by a function of: total token input to accelerated labor output
Jun 1
I don’t know where to come out on it. AI researchers have access to models I don’t and a much greater appreciation for the trajectory of future models. But, many also have less exposure to the physical, regulatory and institutional constraints that shape much of the economy.
19/n
1
21
4,424
well if waymo and fsd are so good, why aren’t all cars self-driving?
in high school i interned at a self-driving car co in the south bay refusing to get a drivers license bc diffusion of such tech was 1-2 years away... directionally right but wrong since the limiting factor is actually simulating other humans with an OOM of nines at 30hz(which ASI still can't effectively do)
realizing that most knowledge work doesnt even hit one nine, runs at 0.01hz on a desktop interface where the input lets you transmit at best 10 bits per second via keyboard mouse should paint an extremely optimistic view of a world of self-driving computers that arrive on a magnitude of months
importantly, safe salient superdiffusion of self-driving computers doesnt require retrofitting, upskilling, or disruption of digital labor; similar to tesla shadow mode where its fsd policy grades itself against human actions steering throttle, an ai lab or the ai rollup that runs models beside operators doing economically relevant work, learning from how "off-policy" it is across mouse keyboard, turning knowledge work from a single-threaded human loop into massively parallelized computer work directed by humans operators seems like an extremely optimistic mission to work towards
kargarisaac.medium.com/activ…
2
6
1,660
Surya retweeted
In just over 2 weeks - I head back and hope to see a mom and her 3 cubs survive another winter in Alaska 🤞
13
103
1,159
11,979
Surya retweeted
May 27
Introducing AttuneBench!
We built this benchmark on a simple premise: for self-improving AI to reach its full usefulness to humanity, it needs high EQ.
We decomposed EQ into distinct skills and evaluated 11 frontier models across 50 real-life topics, from relationships and marriage to school and job stress, using 50,000 first-person annotations.
4
11
58
18,303
“study Long Lake”
May 14
Solving the science of asset selection in a future (or indeed the present) where every company is a "Context Acquisition Company" is the real frontier.
I love that everyone is getting around to the idea that the secrets (scarce context) currently illegible to/hidden from computers (human or machine) are everything. Now the next leap for people to make is that the science of sourcing, selecting, and monopolizing that context (really THE ASSETS that produce it) is everything. If AI progress is a function of compute and data (most algorithmic progress is really just data progress; h/t @BerenMillidge, @_kevinlu, @mentalgeorge, @GarrettLord, etc.), then every company is going to have a context desk just like they will (or already do) have a compute desk. The difference is, CONTEXT IS NOT FUNGIBLE. Most context (both that exists right now and that will be created in the future) will be completely commodity beta. Winning will be about getting to and instrumenting the right asset (context production factory) first. And yes, there are right and wrong answers. To do this kind of asset selection well requires an extremely scarce meta-capability: the ability to coordinate the right kind of access and the right kind capital at the right time.
These assets (and the secrets within them) are structurally difficult to access, evaluate and instrument. They are not floating around in banked processes, to be frictionlessly purchased on listed exchanges, or willingly coming through Mercor or Handshake's expert portal. (Yes, a context production asset can be (very often is) a single person or collection of people.)
When @WillManidis talks about a Deal Guy Yuga, what he means is that there are people who have deeply internalized the fact that at the limit, in a world of infinite intelligence, access to/monopoly on the right permissioned data streams is all that matters. Getting yourself to a position (meta-access, meta-capital) where you have the ROFR on those permissioned data streams, means being a generational Deal Guy. This is a very different and specific kind of "Deal Guy" though.
Knowing which asset(s) are going to give you the right context to create, compound, and commercialize the best vertical world model now and into the future is the new form of security analysis. But the triple-exceptional combination of domain expertise, meta-access, and technical ability that’s required to execute this new security analysis effectively is scarcer than the talent at quant firms, YC combined, and dare I say, the labs, combined. Palantir understood this and it's why they focused on getting root-access (or something close) to the "highest-status" institutions, and the data streams they produce, first. If you have the talent that can get access to and create value within those institutions, everything else should be a forgone conclusion.
If you want examples of the teams that (I believe) actually understand this new science of asset selection and long term value capture in a world of infinite intelligence, study Long Lake and @formationbio. They know and have known that it's all about being able to get the right asset (context), in the right market, with the right team (machine and human) first. These two companies are very far ahead on the scientific frontier of context acquisition.
GC backed Long Lake last year. Do you think it’s a coincidence that Long Lake chose to work with General Catalyst? My bet is that Long Lake knew they wanted to acquire Amex GBT before they partnered with GC, and that they partnered with GC because Ken Chenault (the ex-CEO of Amex) is General Catalyst’s Chairman. That gave them the right access at the right time to a very valuable context asset (Amex Global Business Travel)
A superhuman vertical-specific Elon operating every company means market leading monopolies in every single slice of the unstructured economy. The thing is you have to build this superhuman Elon while flying the plane. You can't build this superhuman Elon without the very specific context that operating specific assets in the real world gives you. In fact, there's only one stream of context that was able to produce human Elon! Knowing which context stream is likely to do the same a priori is so extremely difficult, but probably possible.
I’ll let you intuit why Amex GBT is both most likely to be the market leading monopoly if it were operated by the superhuman Elon of business travel and why it’s also the most likely to produce the context to build that superhuman Elon.
The labs of course are very large acquirers of context at present and I think they will continue to play and improve their capabilities here. Through their deplyoment companies, they have already chosen the PE funds that they deem to be the best Context Acquisition Funds. Through in-house deployment focus on Life Sciences they have chosen the vertical they see as containing the most valuable context producing assets. They will acquire very seemingly unrelated companies and will acquihire very interesting people just to get tokens, they will create a Context Acquisition Fund of Funds. But it's not a foregone conclusion that they become the best performing context acquisition companies. Or that they even view it this way. And that presents an opportunity for anyone that does.
3
1
89
22,417
frontier intelligence requires being out on the frontier jobs.ashbyhq.com/long-lake
10
3,184
trillions of tokens
May 12
the AI roll-up is only getting started
Long Lake CEO @alextaubman has bought 30 real-economy, non-tech companies and just took the largest corporate travel platform private (Amex GBT, $6.3B)
its a bet that AI can better transform legacy industries and that someone has to turn capex into real economic growth
"You see the hundreds of billions of CapEx the labs are investing. Somebody's got to take that and turn it into GDP growth. That's what Long Lake was formed to do."
2
1
59
16,522
long lake management holdings
May 13
AI is one of America’s greatest achievements, and yet 99% of U.S. services businesses lack the resources to deploy it. We’ve built Long Lake to bring AI to the real world, as a true partner, at scale. It’s a hard problem that can only be solved by a purpose-built team of world-class AI Engineers, Operators, and M&A professionals working together.
We’re hiring - join us.
llmh.com
5
82
20,272
the industral revolution made goods abundant by turning labor into capex and similarly ai is doing the same to cognition: everything software-shaped is marginal inference cost
yet historically America is not software-shaped, its not tool-shaped, its a country of services bundled in to limited liability organisms that trade trust for distribution and the right to be accountable for other humans
what the labs have built -- tool-shaped, inscrutable weight files that echo consciousness, locked behind a series of API calls -- provides ever-growing leverage on digital labor. yet the rest of the economy is not inscrutable; it's owned, operated, liable, naturally anthropogenic of how people manage bureaucracy and drive capitalism.
"You can outsource your thinking, but you can't outsource your understanding"
the core understanding of how a firm runs with its situational judgement, awareness for a customer under economic pressure, its duty for that customer, historical context on how to serve customers better -- was not previously trainable (we're now seeing data co's buy slack logs, emails, etc). synthetically generating this data even by paying someone to roleplay as an analyst or physician; making them feel like their service is being converted into training exhaust breaks the core trust an institution has fought so hard to earn.
Understanding itself is what makes a firm an institution.
in this tweet i also want to poke at the "super stack" -- we have superintelligence, superalignment, yet superdiffusion: intelligence saliently diffused across the economy, doing inter-planetary backpropogation on claims resolved, minutes saved, margin expanded, where the policy is rewarded with USD is a core third pillar humanity needs to safely monitor; focusing on accelerating human capability and expanding human agency with superaligned, superintelligent models that are economically-aligned.
imagine backpropagating across a country.
the only place to run the full super stack is inside an AI rollup. AI rollups inherit trust, build the surfaces of work, and route intelligence to seek exponentially harder economically-relevant reward signals -- accelerating human operators, expanding human agency, and enabling recursive diffusion across the firm's economy. AI rollups are the full-super-stack institutions that turns tokens into labor.
recursive self-improvement will bring us a country of geniuses in a datacenter. but for America -- who will birth the country of services, if not an institution?
May 3
the industrial revolution made goods abundant. ai will do the same for services
7
3
111
15,685
this took me longer than expected to write, appreciate @yunyu_l @khushkhushkhush for reading and reviewing
if you want to work on superdiffusion -- reach out
2
5
1,878
6
1,193
how it works: sdan.io/projects/backscroll
7
932