Sarah Guo

hi, it’s me, social choice nerd calling in!!! For anyone following the NYC dem primary: ranked-choice voting (RCV) lets voters rank candidates in order of preference, redistributing votes until one candidate achieves a majority. It’s used statewide in Maine and Alaska, and in over 50 U.S. cities (including SF!) NYC adopted it for the dem primary to reflect broader support, reduce negative campaigning, and reflect diverse voter preferences. There are some criticisms! Mainly, complexity. There are also some failure modes. For example the “center squeeze” is when a moderate candidate who could beat either extreme candidate individually is eli…

Congrats Tarek and team! A triumph of execution and long-term commitment to a vision (and some super smart folks).

@jasonth0 no, these are companies with tens of millions of ARR and real users and lots of product investment

Congratulations team @harvey__ai! AI progresses at a different pace

actually you can get funded with no TikTok following, but instead, happy and committed customers (Though public proof of user love aka “customer marketing” is one of the most powerful distribution weapons.)

Running an AI startup today using a SaaS-era playbook or even previously “best-in-class” speed is like gaming on dial-up. By the time your GTM feels comfortably dialed-in or your strategic bets seem sure, the match is over. Markets are moving (and being won) at meme speed. hesitation equals irrelevance. Seeing strong traditional execs caught off guard, challenged to adapt.

@dchaplot @miramurati @sritreee @BrendanFoody @deviparikh @DhruvBatraDB

a good way to judge investors: 1. do they actually have relationships worth something to you? earned trust with customers, hires, partners, investors? 2. how will they treat you, vs those relationships with people more “important” today? will they make a call that costs them something for you? It could be just an uncomfortable one, one where they go out on a limb, where they put their credibility on the line to help you hire, or something that costs them social capital I’m grateful every day @mvernal is this type of (secure) person. most people aren’t

We should celebrate hacking, not cheating. Cheating quietly pockets extra Monopoly money (insider trading, plagiarism, inflated metrics). Hacking openly flips the game board, exposing rigged rules and inventing a better game for everyone: Wikipedia replaced Britannica, Airbnb monetized spare rooms, Linux openly challenged Windows. Both break the rules, but hacking doesn’t compromise integrity. it reveals the old game wasn’t worth playing.

Some dreams are not for sale

The “AI leapfrog effect” is quietly transforming sectors historically slow to adopt technology—legal, healthcare, logistics. They’re not merely catching up. they’re vaulting ahead. Traditionally, automation in these industries was blocked by high barriers: expensive software, difficult integrations, ambiguous ROI. The business math rarely worked. Large language models, code generation, MCP and computer-use agents have fundamentally shifted the economics. Suddenly automation is affordable, fast, and obviously valuable. Examples are emerging everywhere: @harvey__ai in legal, @latent_health and @openevidence in healthcare, @SolaAI_ in logisti…

the power/battery of today’s smartphones is just not gonna work if we really begin to use local models all the time (or even just constant input, eg voice processing)

@ssankar @boztank @kevinweil @bobmcgrewai 🇺🇸

.@AnthropicAI founder @8enmann on whether Claude Code competes with Anthropic’s own codegen model API customers: https://t.co/IZ7lvrKBu6 full @NoPriorsPod interview https://t.co/mZXsWc9HP8

Anthropic, of the all the top labs, spends the least on and believes the least in human data collection (“it’s not scalable path”). Interesting research result if you push on without it: (another interesting research direction is how we learn more from the pretraining data we already have)

guys GCP / the internet may be down but @baseten gotchu https://t.co/iRv9233j0G

VC firms are nothing without their people and the GOATs who worked there a decade+ ago or lost touch with the market moving into management, do not count. you’re only as good as your last investment

super cool glimpse into agents taking actions in the wild

Apple transcription is still so, so bad compared to what’s available on market, they should be embarrassed the use case for transcription: https://t.co/hx3cCU0cqS

is it weird that I think of nature as pre-training and nurture as post-training? (we have 3 kids)

@gradypb too bad I can also agonize about a piece of writing for three weeks because I’m not clear on what I think (yet)

guys if you missed @aiDotEngineer yesterday and want to see my accidentally no-speaker notes keynote on the AI application opportunity, complete with: - a full five minutes of stalling through AV struggle - several Steve Ballmer impressions - a recipe for building “cursor for X” https://t.co/9x0bGV21lq https://t.co/8LXW23Rrru timestamp 59:08 thanks to @swyx for the invite and the amazing community he’s built

if anyone has some bad news to bury might I recommend right now

new @NoPriorsPod with the legendary @drfeifei! topics: why she’s building world labs missing gaps in the AI future robotics and physical intelligence greatest challenges of 3D her PhD, ImageNet advising sensei Karpathy importance of human-centered AI https://t.co/z4Ok7t74Ly link to full interview: https://t.co/v5gh8O7V70

guys Jamey is a savior work with her

I had ChatGPT do the comparison for you! I am trying different styles. I usually write now when I get the feels — like I have something to communicate to a specific founder(s), or when there’s something really interesting happening in the world. I think the ChatGPT critique is okay. This essay was too long. I told @GuptaRK22 whose writing I admire I thought it should have been half. Sermon-like? Youch. Taste in food is an obvious metaphor here. The idea that this is more “manifesto-like” is fair, but I have a stronger point of view on a lot of things than 5 years ago. There were a few anecdotes I forgot to put in — like Ivan being friends w…

Taste

Stripe returns errors in plain English: "That card number doesn't look right." Not ERROR_INVALID_PARAMETER. Because developers debug at 2am. Spotify's shuffle isn't random. It avoids playing the same artist twice in five songs. True random feels broken. Engineered random feels right. Notion's drag handle appears only on hover. Six dots arranged in two columns. Not three lines, not always visible. Because permanence is clutter and six dots whisper "grab me" while three lines shout "I'm a menu." This is taste. The relentless, almost painful ability to know what should exist, what shouldn't, and where quality matters. It's the difference betw…

reasoning + tool use is so good

join us next week 6/9 6pm at @conviction in SF for the latest edition of Mixture-of-Experts, our talent event highlighting breakout AI startups that are hiring! Feat the founders of: @openevidence @CorridorSecure @decagonai @ListenLabs and Collate. https://t.co/eBpkYe5mzX engineers, researchers, builders: register here https://t.co/hLDMlfgCaN cc @jessemzhang @SurbhiSarnaSF @zackmziegler @jackhcable @itsalfredw

guys we have 🤌 lineup coming up