Sarah Guo

“Maybe the AIs didn’t surpass us; maybe they just became a mirror — showing what we value most.” https://t.co/btjk5DvHq4

2025, everyone https://t.co/ST51yt9siY

the most fun NeurIPS submission

that feeling when you believe but you know you’re also crazy for believing

this is the year of real time voice! Congrats to the @Cartesia team. Their architectural creativity, obsession with performance, and taste finally makes AI conversation feel human (sub-200ms latency, multilingual consistency, and natural emotional range).

when the pre-game is as good as the game! At $NVDA GTC in DC to chat AI, chips, power, startups, and what we need to continue the cycle of innovation and the American compute build-out. 🇺🇸 Livestream tomorrow at 8.30am ET: https://t.co/2u3uWxTqXZ https://t.co/pttq7blrXr

.@cognition RL-trained models...making dev experiences better in the wild

we do not equip young people for their most important choices: what to do, who to be with, where to live really glad Bill wrote this

“real engineers won’t use agents to code” said the naysayers

a customer friend gave me a problem statement for hardware design/manufacturing ai software that I think is promising, please lmk if you are interested in this

this is super cool and everyone should have exactly the software they want, without the cruft

.@JaredKushner is an extraordinary diplomat — normalization, peace in Gaza and release of hostages are all unequivocal goods. Jared is a learner, relationship-builder and positive sum thinker. In business we judge people by their results, rather than their identities.

New @NoPriorsPod w/ @ericzelikman (former Stanford/@xai) about his new startup: humans& we talk about training models with EQ that collaborative with humans to unlock their full potential, the challenge of understanding human objectives, long-term memory & his views on abundance https://t.co/KzXNYvFkOV Full interview on YT (or any podcast platform) https://t.co/Xgejnfen2I

The models don’t know how a private equity firm like KKR does a particular process, and that specificity can be very important to KKR. It’s not a reasoning or IQ problem, and that’s where process data and post training can be important to our customers. -@winstonweinberg CEO @harvey

if you could interview anyone dead, who would you interview?

big consultancies are feeding their clients AI slop. using consumer-grade tools this way is going to erode trust https://t.co/3BPZU7fTPw

stopping for 30’ to celebrate @conviction turning 3. thankful to our team, founders and community (back to work) https://t.co/5Tr8ozclKU

“AI will dramatically alter consumption of information. Why do people love ChatGPT? I want to click on a link…said no one. The opportunity for $SNOW in AI is about unlocking the data in and available to us, how users want to consume.” - @RamaswmySridhar in convo with @gradypb https://t.co/te6kivtwUa (was and is the right guy as CEO)

Coming up on @NoPriorsPod: The GOAT, Jensen Huang Intel CEO Lip-Bu Tan ⭐️ researcher Eric Zelikman Magnetar CEO Neil Tiwari And some other surprises… Who else do we need to have?

as someone who once wrote 12 essays about Dante over the course of a semester, the experience of “closely reading” something substantive, and then thinking hard about it, feels generally lost now. and hard to recapture (Particularly in education!)

custom thinking for everyone!

1. What 2. Are you looking for a job https://t.co/5pEEsXaouI

GET AFTER IT

@o_mega___ shut up u clanker

2/ Step one sounds simple: log expert actions. Every redline, override, or escalation is a datapoint. In theory, you can clone a policy from history. In practice, logs don’t capture why. A clause might be struck for legal risk, client preference, or sheer habit. 3/ Next, train a reward model from feedback. Collect comparisons: “this draft is safer than that one.” Works great in RLHF papers. In high-stakes domains, expert time is scarce and their judgments often conflict. Sparse, costly, and messy signal. 4/ Even when you have outcomes, credit assignment seems kinda brutal? A case outcome depends on years of motions; a patient’s recovery on …

2/ Quinn Emanuel strikes fear. It didn’t get there with clever clauses alone. It’s decades of scar tissue: judge-specific heuristics, forum strategies, discovery playbooks, damages theories tested in high-stakes trials. That accumulated edge is institutional capital. 3/ A product that really helps Quinn amplify its edge isn’t just “run contracts through GPT.” It encodes fallback hierarchies, privilege rules, sanctions risk, motion odds, firm-specific research strategies. Generic models won’t surface that out of the box. 4/ UCSF saves lives at scale, not just because it has brilliant individual doctors, but because it runs care as a system: …

terminal is all you need

Undying fan of @LauraDeming and team @untillabs - happy to be a repeat investor/supporter of their cryopreservation mission. This a group of scientists and engineers at the intersection of brilliant, mad, and intensely practical. Amazing team to join.

one of my favorite experiences with @HeyGen_Official remains video translation! seeing immigrant founders communicate in their native tongue (translated to English) with all their expression, passion and gestures is really cool

Big Kaz believer. The last ceo I can think of that took a massive milestone-based equity package was Nikesh Arora. Worked out for him (and the company). Trust.