kitty

@smolotnikov Very relevant middle ground that's a key missing piece to tracking changes currently!

If AI is going to participate in our thinking, we need to build systems that keep humans in charge of the important choices. I built Priori, an interface for human-AI collaboration, toward that goal last year.

@christineist no but interesting people are

are all interesting job offers on twitter now

@devgerred its texas hold em (:

@kittyhychen i haven't played go fish in years

if you’re ever wondering what we do at lunch at pareto, its not rlvr or evals

@Bonecondor gonna be a banger

I’m in my happy place (less than 48 hours to event time, anticipation building, guest list management and line warfare starting)

expanding what’s deterministically verifiable for RLVR is a methodology bottleneck as much as a tooling one. most verifiers are script-based or llm-as-judge. the opportunity is building reward signals durable enough to open up the domains worth training on.

@VeryCoolGuy6 @realchillben agency > knowledge

@realchillben What dat mean

Information is only so valuable as those acted on

@guyfromhills dappled sunlight

big fan of whatever this is called

@belindmo @xai stacked panel

I'm starting a new startup! it's called Long Horizon Research. our first product is Sundial, an AI workspace for humans + agents to self-improve by creating skills together. We are hosting a hackathon tmrw with @AGIHouse @xai, come through

@sarahmsachs possessing the curiosity to dive into the weeds is a shockingly rare trait

A PI once told me "the best computer vision researchers are the ones who stop to look at the pictures". Graphs are great, but if you want to debug an issue, nothing beats putting everyone in a room and reading through some of your most problematic transcripts. 2x a week at @NotionHQ , we all sit in a room and spend 45 minutes going step-by-step thru our most token inefficient traces. One of our most productive meetings.

Von Neumann's insight in Game Theory was that when the "environment" contains other optimizing agents, the problem becomes irreducibly triadic. You can't just model your action against the world, because the world includes agents who are modeling you modeling them. The interaction requires a mediating structure — the game itself — that isn't reducible to either player's perspective. This is Thirdness in Peirce's precise sense: a relation that cannot be decomposed into dyadic components.

@eigenron the only real goal is to have fun

you'll reach a point in life where you'll have unmistakable belief that it is all a game. and the faster you reach that point, the better. because as you reach that point, you will start playing for the love of the game instead of futile motives.

on the ever evolving state of human data and the Pareto frontier ✨

there are boy offices and girl offices and this is an example of the former

@sodofi_ skill issue

sf social life is just attending networking events on luma

@ItakGol something something nominative determinism

OpenAI acquired OpenClaw FaceBook acquired MoltBook Choose your company name carefully…

@RhysSullivan aligned

Push your AGI timelines back a week

@iismallcombe @aurielws vagueposting queen

7pm sunset tomorrow, we are so back

@aurielws Such an underrated take! Attention to detail vis a vis actually digging through and following the model trajectories is critical to good data

@aurielws @henloitsjoyce @akshay_pachaar love this!

@henloitsjoyce I actually really like this post from @akshay_pachaar x.com/akshay_pachaar…