Skills are now a valuable form of software, like Emil's world-class design eng work. Projects like 𝚙𝚐𝚋𝚘𝚝, 𝚔𝚗𝚒𝚙, 𝚞𝚗𝚕𝚒𝚐𝚑𝚝𝚑𝚘𝚞𝚜𝚎 create amazing AI engineering loops.
everybody on your team who is making architectural decisions, those people must know performance and they must make decisions that will allow the other people downstream of them to use an architecture which can be optimized later. If you don't do that, you're just rolling the dice.
A big-enough random reversible circuit is plausibly a secure cryptographic permutation, a big-enough random irreversible circuit degenerates into having only a few possible outputs.
But, simply just having your loops build everything without having some guardrails around the blast radius without having guardrails around how you think about quality, I think is a recipe for disaster.
there's so many people that are not willing to fight for the best idea because they're going to upset somebody else. And if you don't fight for the best idea, then what idea are you fighting for? You're fighting for the mediocre idea, the politically expedient idea
we find that the performance on held out tasks decreases dramatically. Whereas if we um just take out a random 20% of the data that's less diverse than the most diverse subset, the performance um only decreases a little bit. And so this suggests that actually having really diverse data plays an important role in enabling it to generalize to new tasks.
So if you design call it loop, call it graph, call it workflow, it's kind of all the same. If you design something that gets a trigger or gets an input and does something for you and maybe there's a decision in the way, boom, there's your graph.
if you are using these systems right you can just grow more quickly you can build more things you can take on more stuff so you know I would be surprised actually if we ended up with like net fewer
And I think there's a lot of room in a lot of domains for much faster validation models, possibly learned valu validation models that can uh you know get you a a approximation to the true answer much much more rapidly. And that changes how those experimental loops can be thought of and how quickly you can go around those loops.
It was instead a very strongly held opinion that that you get to excellence by giving people a lot of agency and accountability. By pushing decisions as deep in the organization as possible, hiring great people who can be trusted to have good judgment and make good decisions.
if you want to do loops engineering, you should build one loop at a time and you should keep them small and contained. Basically, I think everything except stop reading the code is really good advice.
You'll notice what I said was not use loops to ship the features that users want. We use loops to actually improve the codebase quality and we read all the code because we care about how it's architected and we care not just about the system architecture but what I would call the program design
So I think the the thing I'm most excited is actually like what we call like iterated loops or like slow loops where we basically have a cron job. We have the loop the the the structure of the loop is really easy. It's like run this llinter fix one thing commit and push and then we run that every night in our GitHub actions and we wake up every morning to one PR that makes the codebase a little bit better.
Paradoxically, the organization that demands trust can get less of it, while the organization that shows their work - thereby not asking for trust - builds up more of it.
I think a lot of the things that people might say that like oh, it was a waste of time to learn this subject cuz I didn't actually use like those details on the job. I think that's a very wrong way to think about it. And I think that's what a lot of people are doing now with AI. Like hey, what what if I'm not going to be writing for a loops a couple years from now. Um I don't think those things are a waste of time.
And then you look at the actual trial records, and maybe one in a hundred convictions for that crime actually ends in a capital sentence. And almost all of the other ones end in a fine or a public flogging, but not in the sentence that's on the books.
So, I think there is a world where it is concentrated, in which case it's going to be really hard to index AGI. There is another world where it is not It's electricity, then like basically every company has access to AGI. So, you just buy you use buy the index. So, like, you know, Nigeria just needs to buy the index.
we went from infrastructure is code to infrastructure is data and infrastructure is code is like if this do that um bring in this module for loops all this stuff and Kubernetes is like no no no you have to specify exactly the containers you want how much memory that they need and then we have the status field to tell you if they were running or not
know your goal or suffer a death by a thousand compromises. Because what I had done my whole career and most of us do is compromise to get that next engineer, CTO, investor. You put a jerk on your board because you impressed with their firm name and the valuation and all your friends are going to be impressed and it's going to be so much easier. We make all these compromises and contort ourselves and eventually we wake up and it's company we don't want to work at.
The way I think of the abyss is it's this place that we go to as founders and entrepreneurs after our thing, after it dies or it's bought or it's over for whatever reason. It's this amorphous place that we are in our life that has no structure.
If AI-driven labor displacement ends up being large in magnitude and permanently drives down the demand for labor, it will likely be necessary to go beyond mere incentive programs to long-term income support for a significant fraction of the labor force.
my sense is that these complex traits have not pushed in one direction because there's advantage there's spectrums where there's advantages to both ends of the spectrum
Um and then they can kind of suggest random things and it it but it it it um often I find that trying to chase them down and make them work and find they don't work, it wastes more time than it saves.
We show that equilibrium generically occurs at neither the Harberger nor Glaeser-Luttmer benchmark. Cost-minimizing suppliers drive allocations to vertices, not interiors. Corners are not an assumption but an outcome about what cost-minimizing suppliers choose. The correct benchmark is corners, not random, and corners generate qualitatively different welfare properties: losses far larger than either efficient or random distributions, and discontinuous jumps from small parameter perturbations.
but ice sort of forms through a process of random nucleation and then extension. And this is cool because you can uh modulate the like sort of nucleation the rate of nucleation and extension to then modulate the probability of ice formation.
And so what ends up happening is the emails that the AI write are pretty good. Okay? If you're getting terrible emails, it's a poorly trained product from a bad vendor.
I just don't know where this is going. I don't know what the end is. When I don't know what the end is, I don't know what the beginning is. And it's that simple.
If you're a paper towel demon and feel guilty about it, I can't recommend these enough. Perfect for cleaning counters and random floor messes (talking to parents).
Anna Karenina by Leo Tolstoy: Made the random choice to listen to this and loved the experience (especially for getting through some of the boring parts, sorry). Shout out Maggie Gyllenhaal.
PMs are no longer saying to the designer, hey, can you draw this thing out for me? That frees up designer time to go explore more deeply the stuff they need to go into and it allows anyone to kind of add to that first conversation of where should we go and look further and wider and broader at the option space.
If you do too much exploration, you can have your team feel a little bit too scattershot, just trying a 100 different random ideas. What's the through line? What's the strategy? How do you pattern match, you know, successes across them? And if you do too much in exploitation, which is often the MMO of growth teams, it can lead to this like saturation and stagnation where you're just locally maximizing a thing.
Tweets from 1765-1799 by a 4’9” hunchback physicist, friends with Goethe and Kant, admired by Nietzsche, Schopenhauer, etc. Such wonderful random thoughts, beautiful perspectives on thinking for yourself, observing nature, language, freedom, philosophy, religion, and more. Hundreds of initial insights, especially inspiring because they’re undeveloped.
My model is that research requires a mix of skills. The day-to-day coding and execution is crucial. But there's also a set of harder-to-learn conceptual skills, collectively called research taste. These skills take a long time to gain because they have poor feedback loops, but they take very little time to use.
This means that working on neural networks is NOT getting us closer to AGI, except indirectly.
Their words now
It’s become very difficult for me to maintain the belief in the stupidity of ChatGPT when every time I laugh at it, it ends up ridiculing me 6 months later.
In other words, the singularity ends up in their future and they can no more avoid the singularity than they can avoid time coming their way. There’s no shenanigans you can do once you’re inside the black hole to try to skirt it.
We'd constantly be improving the tools and just the iterative process and the speed at which that improves products is the critical element to success in games. The slower the iteration cycle, if you make a build every week and you prove, you go through one iteration every week, you're going to be way way way worse by the end of your project than a game company that makes new stuff every day.
until there are feedback loops of open source AI, it seems like mostly an ideological mission. People like Mark Zuckerberg, which is like America needs this and I agree with him, but in the time where the motivation ideologically is high, we need to capitalize and build this ecosystem around, what benefits do you get from seeing the language model data?
And this is important, so yes, you should expect legs or jointed sticks, how many joints they're going to be? Anybody's guess. Do you expect humanoids, things with a sensing apparatus on top of a shoulder with two arms and two legs? That's probably a pretty random set of occurrences that led to that.
In fact, there’s a handful of dates that tell us that the fourth creation does continue farther on, that that baktun place should have 20 baktuns in it, like their counting system would dictate, not 13.
Physics hardly even acknowledges that the universe is random at its base. We like to think we live in a deterministic universe and everything’s deterministic. But I think that’s probably an artifact of the way that we’ve written down laws of physics since Newton invented modern physics and his conception of motion and gravity, which he formulated laws that had initial conditions and fixed dynamical laws.
It’s the fact that you’re a deterministic structure living in a random background. And also, all of that selection bundled in you allows you to select on possible futures. So that’s where your will comes from.
And what we learned from his declassification is that no matter how nuclear war starts, there was a bunch of different scenarios, with NATO involved, without NATO, all different scenarios. No matter how nuclear war starts, it ends in Armageddon. It ends with everyone dead.
The more addicts rely on these stimuli, the less pleasure they receive. At a certain point, this cycle creates anhedonia—the complete absence of enjoyment in an experience supposedly pursued for pleasure.
But we run the simulations to see what happens, not if we knew what happens, we wouldn’t run the simulation. So whoever created this existence, they’re running it because they don’t know what’s going to happen, not because they do.
Friedman’s focus on the money supply has not held up, as Samuelson suggested, but the alternative Keynesian macro models recommended by Samuelson in the same interview have not done better and they were not outperforming simple random walk models of predicting the macroeconomic future.
We’re actually going to build that, I think, but it’s not going to be one static tape. I think the human brain is too complex to be stuck in one static tape like that. If you look at ant brains, maybe they can be stuck on a static tape, but we’re going to build that using generative models. We’re going to build the TikTok that you actually can’t look away from.
tech barons are ordinary mediocrities, no better and no worse than the monopolists that preceded them, and any differences come down to affordances in technology and regulation, not an especial wicked brilliance.
Mind candy horror, continuing Hand's interest in beautiful places where people just go missing (and come to bad ends). It's exciting and creepy, and also one of the better fictional depictions I've found of the strangeness of 2020.
Programming, of course, is forgetting, but we need to at least try to be aware of the costs of the abstractions we choose and consider who it is that ends up being forgotten.
Like nuclear physics, with its potential for global catastrophe when put to destructive ends, the proliferation of pandemic biology ought to be considered a matter of international security.
Now you have a carbon fuel burning aircraft that is more complex and heavier than the plane that it replaces, and if you do your sums carefully, you find that there are no fuel efficiency gains and, therefore, no environmental gains.
Control System Design by Bernard Friedland. Control theory rocks! I've been meaning to learn it for years. The math is easy, but the idea of setting up feedback loops to stabilize systems is cool, and the techniques to do it optimally are really fun.
For assembling research for academic purposes, I like Scrivener. It keeps together a lot of odds and ends of information and makes it easier to go from a pile of quotes and references to a finished document.
A korrent is a belief a person has stated in their own words: one
sentence stating the claim, backed by a quote and a source, kept at
korrents.com.
Under a name here, the quoted block is what they actually said.
The korrent beneath it is the claim those words support, in
korrents' wording — tap it to see the record, its source, and who
else holds it.
Nobody here wrote their own korrents. They are compiled from public
statements, and a person can change their mind, which is recorded too.
About the English under a post
Some people here publish in a language other than English. Where they
do, this site shows a machine translation beneath the post, in
this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the
serif above is exactly what the person published, and it is what to
quote them on. A translation can be wrong in ways that matter,
especially about tone.
Only the post's own words are translated. A quoted post, a linked
article and a belief on korrents.com
are left in their original language.