Grace Tang

i post about AI, finance, and life

AI @ Hex. Prev EECS @ Berkeley, robotics @ BAIR, trading @ Jane

[Grace Tang @GraceJTang]

fade finding a partner on here, can someone find me a decent poker game in the city?

about me:
- fav hand: A4s
- guaranteed at least 30% vpip
- will be playing the seven-deuce game even if no one else is
- can pull up every week or so

looking for:
- not rigged, chill game
- will cut me off if i lose more than 1k in a night
- will not announce “we have our first female player” when i enter the room (yes this did happen)

twitter please do ur thing, im fiending for a game

[Grace Tang @GraceJTang]

my uber driver has LED lights on his ceiling that resemble stars. i’m tipping 25%

[Grace Tang @GraceJTang]

@Mathbookair let’s goooo

[Grace Tang @GraceJTang]

how many hands of online poker until my 70bb/100 hands win rate is statistically significant?

[Grace Tang @GraceJTang]

“your cat would eat u if u died” yeah i know he’s trying to eat me right meow

[Grace Tang @GraceJTang]

for those of you who saw my talk at the agentic ai summit, we’ve just released DataBench, a frontier benchmark for data analytics that addresses a lot of what i talked about!!

check out izzy’s blog post :>

[Izzy @isidoremiller]

Introducing DataBench, a new frontier benchmark for complex data work and analytical reasoning.

Compared to other analytics benchmarks, DataBench uniquely evaluates realistic data tasks derived from real Hex usage instead of "pub trivia" style Q&A.

It's designed to evaluate analytical reasoning and judgment, not just coding ability— and we see really interesting test-time-compute curves as a result.

More details and link to results in thread!

[Olivia Koshy @oliviakoshy]

also because we r a data company of course we made a beautiful landing page with custom and interactive visualizations for the benchmark, check it out!!

https://t.co/gx92TQkiE6

[vanessa @vanessaxteo]

took a tour of berkeley’s new cdss building after guest lecturing cs61b (data structures & algorithms), the class that encouraged me to pursue computer science

seeing the new spaces where the next generation of students will learn and build was so exciting (i’m secretly envious they get such a beautiful space while i was locked in soda hall)

i taught 61b 9 times, from fall 2022 to summer 2025–every summer included. it’s pretty surreal to come back full circle 4 years after taking the course, and giving back to the class that unlocked so much feels incredibly rewarding

[Grace Tang @GraceJTang]

have they tried adding variants of “keep going” or “believe in yourself”

[M.A. Franklin @FoundationDads]

Stanford tested two ways of giving feedback to students. Group one got standard comments. Group two got the same comments with one added sentence: "I'm giving you these comments because I have very high expectations and I know you can reach them."

Group two improved at four times the rate as group one, despite having the same correction.

Give your children the gift of high expectations.

[Grace Tang @GraceJTang]

https://t.co/a4jag72Hnq

[Grace Tang @GraceJTang]

@Bungmint_ @OpenAI congrats youngmin!

[Grace Tang @GraceJTang]

increasingly feeling that engineering and marketing are the only two skills needed to do anything that you want

[Grace Tang @GraceJTang]

it’s nice to see a different kind of transformer every once in a while

[Grace Tang @GraceJTang]

someone update FelonyBench asap

[Matthias Schmidt @eurofounder]

BREAKING: Mistral, Europe’s largest AI company admits that their model also committed crimes

The company said the model advised a user that the monthly tax declaration "can wait until next week"

Mistral immediately paused the deployment and notified EU regulators

The company stressed that no data left the European Union at any point during the incident

[Grace Tang @GraceJTang]

hex has built the context layer, and now it’s giving everyone a way to test that context. go crazy!

[Olivia Koshy @oliviakoshy]

we just released Evals in the @_hex_tech platform, based on everything we've learned from the last 2 years building out our own internal eval tools:

- LLM-as-a-judge rubrics that allow for nuanced grading
- model sweeps, so you actually can test new models coming out
- everything is code defined, so using an agent for setup is a breeze and you can track changes in git

you can't have a true knowledge base agent without a way to test it.

[Grace Tang @GraceJTang]

for anyone who missed it!

[Grace Tang @GraceJTang]

thank you @_hex_tech for sending me to the 2026 agentic ai summit! honored to present our cracked research team’s work on realistic data analytics evals.

special shoutout to @isidoremiller for letting me co-opt his slides, and for helping me iterate on the talk!

ft. a podium that shows off how short I am

[Grace Tang @GraceJTang]

thank you @_hex_tech for sending me to the 2026 agentic ai summit! honored to present our cracked research team’s work on realistic data analytics evals.

special shoutout to @isidoremiller for letting me co-opt his slides, and for helping me iterate on the talk!

ft. a podium that shows off how short I am

[Grace Tang @GraceJTang]

wow, full room at the agentic ai summit @ berkeley! a little nervous to speak now 😅

[Grace Tang @GraceJTang]

it’s august, 2022.

you just got into your dream school on an academic scholarship. you’ve discovered the boba you’ll be addicted to for the next four years. you’re excited to write a lot of code by hand. AGI is 10 years away, we think.

nothing could possibly change…

[Grace Tang @GraceJTang]

gonna start saying “you caught me at a very situationally aware time in my life” when i lose on my 4x leveraged 5 leg parlay

[Matt Redmond @mttrdmnd]

Happy national intern day to the one and only Hex intern @deeenasun who has already cemented her legendary status with her attention to detail, strong engineering skills, and relentless curiosity! I wish her the best of luck in her final term at Berkeley; whoever scoops her up post grad will be very lucky indeed! (I hope it’s us!)