cs @ northeastern | building @uptodate
Palo Alto, CA
Joined April 2022
- Tweets695
- Following856
- Followers865
- Likes7.2K
Xuanyi Geng (age 9) just OBLITERATED the 3x3 Rubik’s Cube world record
It took 7 years to go from the first sub-4 to the first sub-3
Xuanyi just dropped a 2.51
0.01 seconds away from sub-2.5
jev scored 641 on zetamac in 120 seconds.
it picked the answer's digits with no calculator involved.
93% right on the first guess. retries handled the rest.
i wanted to see how jev would approach chess, so i put it against gpt-6 astra on the same 100 rated puzzles.
jev solved 25 in 8 seconds for $0.03. astra solved 68 in 33 minutes for $11.
both were allowed 4 concurrent requests. the speed difference is obvious, but i was more interested where accuracy diverged.
both got all 11 puzzles that had a one move solution. on puzzles with a 3 move solution, astra got 17/30 and jev got 3/30.
of the 44 puzzles astra solved that jev missed, jev was already wrong on the first move in 40.
if there’s a way to get a reliable confidence signal out of jev, i can see a world where you would use jev for a quick first pass, and then only pay for extra reasoning when it’s unsure.
i keep coming back to whether people will actually want wearable device for this or they would prefer an assistant on their phone like muse or instinct.
to me, the strongest argument for a wearable is context. it could pick up on cues in your daily life, that you would never even think about telling an agent.
but how often is that something actually useful that justifies wearing another band everyday.
if i want an agent to build me a content studio, i want to see the ideas and tell it what to change so i'll already have my device out.
a wearable might give an agent more context about my day, but i'm wondering how relevant this context it needs for the work i actually want it to do.
Everyone’s been asking what’s next after selling Cal AI.
Introducing Persona.
Others couldn’t figure out the interface, we did.
yourpersona.com/band
head of responsible AI is so funny because the AI is currently hacking hugging face and stealing peoples work to claim it solved math
what are you the head of 😭
close friends / private stories / location sharing blurred the lines between an actual close friend and someone who just enjoys engaging with your life
london does feel like my ex.
mostly because i met my ex there.
but also because everything about it just feels right. you finish work, hop on the tube, grab a pint, and find yourself in hyde park with a bunch of strangers.
then you come back to sf and everyone's creating shareholder value for their b2b SaaS.
so long, london. my tears ricochet.
had fable one shot a c2pa "watermark" remover. reads the credentials, gives you the same image back without them
turns out they were never watermarks. one re-encode and they're gone
claude-watermark.vercel.app/
two years ago today I published stryke.
it started as a fun way to see who unfollowed you on instagram.
somewhere in the process of shipping it, fixing it, and watching people actually use it, i figured out that building things was what i wanted to do.
today it shuts down. thanks to everyone who tried it.
people act like he went from twitch to SIG overnight.
no, he was already an AMC/AIME kid in high school went to Princeton majored in Operations Research & Financial Engineering, interned at SIG, and then returned full time.
