Sarah Guo

this is the funniest darn quote https://t.co/kaZbKVrceJ no SoundCloud, but if you’re the founder of an ai company and want some pocket money from Satya and friends to get started, apply to Embed https://t.co/gdxho7JVX2

this is electricity usage. don’t confuse it with intelligence https://t.co/xxUIyQS3wI

America is stronger than we think https://t.co/si1vcNCmgk

1/3 apps are open for Embed v4, due 2/14! Embed is a Schelling Point for the highest quality early-stage founders in AI: offering cash, compute, community and a catalyst follow @conviction for updates, links below https://t.co/96R8gjMg6Q 2/3 appl & program details https://t.co/plhpXwe4l1 3/3 why we founded the program https://t.co/jzQqm7MC5m

helping someone become the best version of themselves obviously benefits from both feminine and masculine energy vc firms expressing only alpha male energy feels like a mistake. great founders have innate two sigma aggression, aren’t looking to be dominated, but complemented having empathy doesn’t make you weak anything human needs to be nurtured

favorite example of generative UI today?

Clear demonstration of how AI will be most usefully applied in systems, rather than in a single model call 👏 @douwekiela and team

using a @nikitabier app, even one built of spite, is product school for consumer founders

have kids, have them early there is no “readiness” or good time for it

.@SurbhiSarnaSF is an unstoppable force. I asked if we could invest one sentence into our convo about her starting another company. Incredibly excited to see Collate (and the mission-driven applied AI team led by her + @nateps) accelerate the pace of innovation in biotech. more from @alexrkonrad: https://t.co/SnEppEDKGe

If Amazon had a pm with taste somewhere, it would have released a books version of “Spotify wrapped” sometime in the last 20 years

okay I realize everyone has recently gone all Bryan Johnson disciples with their “98 oura sleep score” but we only have so much time on earth and something has to give it’s not gonna be work, fitness, family or friends also I’ve been a shitty sleeper since kid 1 😮‍💨

1/ rStar-Math is super cool, underrated. It demonstrates that small language models (7B!) can match larger models (like o1) at math reasoning through self-improvement: 2/ - self-generating high-quality math training data (for a policy SLM) using MCTS and code verification - a process preference model, based on predicted reward for each intermediate reasoning step - iteratively improving both w/multiple rounds of MCTS and retraining on validated data 3/ Self-improvement strategies work in much smaller systems!

can we stop pretending now it’s not arson of wealthy neighborhoods in LA

Bill is a friend and tech founder who (as a real nerd does) has spent the last five years learning about wildfires, including becoming a firefighter and starting a fund to invest in wildfire prevention tech Pay attention 👇

what is this anti-AI propaganda being fed to the kids https://t.co/hz01FwHj0H

The quality of English writing in international academic papers feels like it has dramatically improved over the past two years (I’d assume assisted by LLMs). Seems like a big unlock in global scientific collaboration / learning

MANIFEST DESTINY

if you refer me someone we hire, I have a vacation home in Tahoe that’s yours for a weekend

requirements: 1/ hacker, hustler 2/ ideally handier than me or @pranavreddy https://t.co/qRLu9e6ui3

do any other organizations (besides hedge funds) have something like the “Discretionary Game” at Bridgewater? A game mechanic that allows individual decision making to factor into comp?

this reminds me of Edward Burger’s (the mathematician) book on “effective thinking” is this kind of prompting broadly effective? if so, must be very short term phenom before baked in cc: @alexalbert__ @lilianweng https://t.co/l6wNxvQhZF

despite the tech world’s general enjoyment of shitting on X/Twitter’s EPD team, many great innovations — including “community notes” over a moderator team enforcing “truth” policy 👏 https://t.co/HdPryLxryU

Still not priced in SOTA AI model for <$10M in spend

there are only two modes for startups - it’s not working - we are always slammed and everything is breaking

has anyone programmed a Unitree g1 before, pls dm

@karpathy @zheanxu ?

startup founders, what did you spend your last $6M on because apparently you could have trained a frontier model

.@deepseek_ai highest impact innovations can be summarized as “communications efficiency.” AI is a distributed systems problem!

2.788M H800 GPU hours < $10M