Gary

Co-founder @ Perfectly (YC W26) | ex-MLE @tiktok_us and @Meta

[Gary @GaryLuo1994]

Spent an hour with @samdblond at Monaco HQ today.

Rare to meet a founder this sharp who's also this generous with his time.

[Alberto Rosas @albertorosasg]

Narrative violation: builders still want to build.

AI doesn’t kill headcount. It increases ambition.

We’re entering a new era of software factories, venture builders, and tiny teams creating massive companies.

The builder era is back.

[Zara Zhang @zarazhangrui]

I built a Chrome extension that turns your "read later" list into dedicated reading time on your calendar.

Save 5 articles → it auto-books a 30-min "reading block" on your Google calendar, links included. So you'll actually sit down and go through them.

No account, no server, everything local. Open source (link below)

[AI at Meta @AIatMeta]

We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2.

Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication.

We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating.

🧵👇

[Gary @GaryLuo1994]

@hthieblot Relationships

[Gary @GaryLuo1994]

Don’t ask LLMs to generate overall scores. Ask them to generate true or false scores on criteria. That’s how we build our AI sourcer.

[elvis @omarsar0]

If you use LLM-as-judge, this one is worth reading.

(bookmark it)

It's actually one of the most effective ways to use LLM-as-a-Judge for evals.

Holistic judge scores hide both their reasoning and their ceiling effects.

BINEVAL decomposes each evaluation criterion into atomic yes-or-no questions, answers each independently per output, then aggregates the verdicts into calibrated multi-dimensional scores.

Every question-level verdict is inspectable, so you can diagnose exactly why an output scored low, and the same verdicts feed straight back as targeted prompt-improvement signal.

Across SummEval, Topical-Chat, and QAGS, it matches or beats UniEval and G-Eval, training-free, with especially strong results on factual consistency.

Paper: https://t.co/oar6BZcasm

Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX

[Gary @GaryLuo1994]

Great video!

[Gabriel Jarrosson @GJarrosson]

I flew from SF to New York with one week to raise $50M for Fund II. The first yes, the no's, everything in between.

[Gary @GaryLuo1994]

@kushagrchitkar @rohansharma0509 @divitsheth007 @usealmanac Cong!!!

[Gary @GaryLuo1994]

Always the best!

[Bek @beknabdik]

Day 2 in YC

I’m realizing again, I’m in the best place I could be right now

[Gary @GaryLuo1994]

Welcome to our new office.

[Gary @GaryLuo1994]

Miss the days in batch, and also enjoy working with my best batchmates.

[Alberto Rosas @albertorosasg]

We share the office with 10 other @ycombinator W26 companies, and yesterday we ran our first group Office Hours.

The energy, ambition, and intensity in this room is hard to describe.

[Gary @GaryLuo1994]

@TTrimoreau Finding anything that AI builds.

[Zizheng Pan @zizhpan]

We’re expanding our team! Come and join us! You will be working in Beijing or Hangzhou. Fluent Chinese is required.

各个岗位热招，均可实习，欢迎投递🙌

[Paul Graham @paulg]

The users who complain about the flaws in your product may seem annoying, but they are on the whole probably your most valuable users. They complain because they care, and I doubt a startup could ever get really big without users who care a lot about the product.

[Gary @GaryLuo1994]

@zarazhangrui lack of the courage to be imperfect

[Yun-Ta Tsai @yunta_tsai]

Many people think any given ML project is 99% training.

In reality, it’s 50% evaluation, 40% data cleaning, 8% integration, and 2% training.

The first two set the noise floor for learning. No ML magic matters; the model cannot lower the noise floor, as that’s the optimal bound of Shannon encoding of your data.

Thus, not a single day goes by without me thinking about ontology. Even the old labels have to be constantly reviewed.

[Gary @GaryLuo1994]

Cong!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

[Gary @GaryLuo1994]

Be humble, man!

[Gary @GaryLuo1994]

Let’s go, Knicks!