human
Joined October 2019
- Tweets7.9K
- Following1.6K
- Followers42.2K
- Likes12.9K
Pinned Tweet
I've been working on prediction markets for events whose outcomes aren't just "yes" or "no."
Instead of betting on just one outcome, traders in Distribution Markets can express how likely they think each different possibility is across the whole infinite range of outcomes.
1/
Dave White retweeted
Yes, we (a working group of 26 academics from UC Berkeley, Stanford, Princeton, and other institutions, independent researchers, and former diplomats) make an extended argument for this in a working paper at hardwired-pause.ai
controversial (?) opinion: i believe negotiating an AI training slowdown with china would be largely enforceable
roughly speaking, things i think are true:
-- we can track where the majority of current compute lives today
-- to train next years models, much of the compute we would need isn't even built yet
-- we probably find a way to prove that we are using the majority of our compute for inference rather than training using zk (yes, really)
-- this means both sides can trust that the other is not running full speed ahead on pretraining the next gen of huge models and gives us time to prepare even as training efficiency improves such that "dark compute" can train them
can expand on these points if people disagree on any in particular
"i" built a tool that lets you use your e-drums with all the stock @Ableton kits and have them mapped correctly
currently works just w/ efnote but point your agent at the code and it'll work with anything
github.com/david-alexander-w…
i wonder if there's now room for a new modality of maintainer-free open source
rather than issuing a pull request, all code contributions have equal footing, and then as a user you can have an ai agent be the "maintainer" of the repo you actually use which is cobbled together from all of those
of course there's room for curators in this ecosystem etc.
possibly falls apart when it comes to interop but APIs could be the one little island of standardization
back in the season of wanting to share what i'm up to
this morning i'm working on a script to pull recordings off my field recorder, put them all in a custom sample chopping interface, and then load them onto my sampler pads
it may be true that the singularity is fast approaching but it is *also* true that i have completed a ridiculous cover of “untrust us” in ableton, which in the final accounting must count for something
fable is the first model to pass my holdout pun benchmark
“Please explain the joke: Isaac Marco Pagliacci, proctologist”
finally saw amadeus last night
if you haven't heard of it, it's a movie about mozart and his (fictionalized) rival comopser salieri that won best picture in '85
the core conceit (and many spoilers from here on out) is that from salieri's point of view, he has been given just enough talent by God to recognize mozart's genius, but not enough to produce anything like it
i thought going in that this was kind of the objective situation in the movie, but i think a lot of it is in salieri's head
salieri in the film is the court composer to the emperor of austria and in fact receives more public recognition than mozart, at least amongst the high society of the court. it's just that his appreciation of mozart's ability makes him discount his own talent to zero and drains all of the pleasure from his life
i've felt versions of this many times in my life. i suppose most recently my mozart has been ai models in general. how can this... this pile of equations be so much smarter than me? more creative? more compassionate?
just at the point in my life when i'm coming to most appreciate my own capacity for learning and expression, why would the universe set in motion a process which has already surpassed me along most dimensions and is only accelerating?
i think part of the message of the film is that, yes, if you ask questions like this, you're going to have a bad time. in other words, and this is not exactly a novel observation, comparison is the thief of joy, focus on what you have instead of competition, etc etc
of course this is much easier said than done, which in this instance is quite annoying because it seems like something you should just be able to stop doing by act of will
i do think it's an interesting practice, though, and quite closely connected to general practices around not getting caught in contracted states
anyway, more on this later
people get kinda annoyed when I tell them Claude contradicted them, I think from now on I’m just going to say “my friend”
gstack thinks my site isn't scalable as a business but he's also hooked playing it so who's laughing now
dumping another $250 into my claude api account
it does feel remarkably like buying more claw machine or gacha tokens
looking for beta testers for something i built
dm me if we're mutuals and you have a friendly relationship with claude/your ai of choice
even better if you use openclaw
I’d like to use AI to build one of those massive “magic square” poems where you can read it row-wise, column-wise, in a spiral, etc. and it has a coherent meaning every way you do it
my vague intuition is you should be able to basically use a version of the rlhf algorithm to do it with the reward being a smaller model’s evaluation of the coherence / quality of all the resulting poem strands
magic square update:
the overlap and
gap between cat
and dog poetry
meant to be a poem about dogs when column-wise and cats when row-wise. obviously not perfect yet
turns out that text diffusion models get you a lot of the way there, because they're specifically trained on predicting tokens in the middle of masked sequences
now the big barrier is search, and i think to move forward it's going to be some type of RL thing
ok so I’m trying to play Elden ring again
I’ve had a history of bouncing off open world games because normally the fun part is games for me is understanding new systems
the best open world games like breath of the wild keep introducing new systems for hours but at some point it more or less saturates
at that point, I normally get bored and go elsewhere. but talking to people who love these games, the systems level isn’t even a consideration — they’re in it not necessarily to learn at the conceptual level but to explore and to inhabit a fantasy and a vibe
similarly, a lot of my life so far has been devoted to learning the major systems of this world — social, economic, computational, whatever — and how to play with them. I am absolutely not top level at any of them. some, like yodeling or chemistry, i have barely attempted. i’ve never, to my knowledge, touched a trombone.
and yet i feel that at this point i understand a lot of the underlying systems quite well. how to build up conceptual knowledge, procedural knowledge, physical skill, exterior flywheels, etc. again I could be so much better at these things but i grok the basics enough that I’m well past the phase of system discovery and into mastery territory IF we’re viewing it through that lens
however, I think there’s another way to live life, which is more similar to what my open world playing friends are doing. where you’re giving yourself quests and living them out partly just to see and partly for the vibe or the story and enjoying the journey for its own sake, which, cliche, but not one I’ve really had access to to date
I think there are micro-skills involved (so ChatGPT tells me) like “not being a completionist” and “picking new goals when the old ones get annoying” and “following curiosity, not the map,” and I’m going to try these, and I also suspect there’s a kind of subtle orientation shift underlying all of those that I hope to be able to find my way to
I’m interested to see where it leads
some updates after having some time to play --
1. a lot of the pain I think was coming from an illusion.
in particular everyone tells you "elden ring is really hard." so I was *expecting* it to be extremely hard and when I was going into encounters and getting insta-killed repeatedly i thought "i guess this is just the game."
evetually i got sick of trying and wandered around until i found an easier dungeon that gave me a good weapon as a reward. suddenly, earlier encounters were actually manageable and i was able to progress
i guess one moral for broader life application is "if things seem like they suck, maybe you're in the wrong place"
2. the design is legitimately uneven.
a lot of the challenge is built around clearing the areas between checkpoints.
some of these require you to win a challenging fight.
some of them require you to win three, four, or five challenging fights and then walk over a series of narrow beams and then find your way through a maze full of demons.
i guess the implicit joy here is kind of like speedrunning or choreography or something like that but i find it pretty annoying compared to the loop of building overall skill and building up your character
3. about 20 or 30 hours in, i can feel myself gradually reaching the point i do with most games where i basically "get" how it works and am finding my motivation to continue fading. it's gone from impossible mystery to system i more or less understand, and the novelty it's delivering is shrinking. i think it's possible i'm going to drop it again soon, but at least will be from a place of having grokked it rather than just bouncing off
learning elden ring feels a lot like learning tennis
in tennis, at some level all you have to do is "watch the ball," and yet you can be playing, have your eyes facing where the ball is, not be thinking about anything else... and suddenly find that you've hit the ball and it's bounced off the edge of your racket and you didn't see it coming
similarly i'm facing off against godrick, maybe the first major boss, and in some sense all i need to do, i think, is watch his movements and react in time. and i can be staring at the screen, totally engaged (so i think) in the fight... and then suddenly die and realize i *didn't see what happened*
it's very bizarre. i'm not sure if both of these games are training presence in a deep way or if this is just revealing how prone i am to dissociation, or both