Tu Trinh (Alina)

Aka Alina Trinh. ML research engineer @scale_AI @ScaleAILabs | Previously EECS @UCBerkeley @CHAI_Berkeley @berkeley_ai

[Slater Stich @slaterstich]

wow so much drama about navier stokes. had no idea things could blow up so infinitely in such a finite time.

[𝘼𝙡𝙚𝙭 @Alex1here]

Day 1 of helping GPT -6 Astra unlock AGI.

[grovy @grovymango]

Being Vietnamese kinda makes you a nepo baby . Like. u were born into eating Vietnamese food

[Scale Labs @ScaleAILabs]

📣Call for contributions + co-authorship!

RSI Bench is our ongoing effort to evaluate whether AI agents can develop the capabilities required to advance AI R&D.

Got a frontier AI research problem on your mind? Submit it to RSI Bench, and set the standard the whole industry uses to measure progress!

Contributors receive:
→ Co-authorship on the RSI Bench research paper
→ $2,000 per accepted task for the initial 50 tasks
→ Modal compute credits to build + iterate
→ Access to the RSI Bench research community

Register below for full requirements.

[Tu Trinh (Alina) @thetututrain]

@nabla_theta lol was this your first interaction with her? So funny that that's all it took

[Tu Trinh (Alina) @thetututrain]

@vipul_1011 What about X?

[Tu Trinh (Alina) @thetututrain]

@Jackyhuang the best vibes 🥹

[Tu Trinh (Alina) @thetututrain]

@DZhang50 So sleek!

[Tu Trinh (Alina) @thetututrain]

@Jackyhuang 🙋‍♀️ i have many side quests

[Scale Labs @ScaleAILabs]

Today we're releasing HiL-Dynamics, the first open-source tool that measures how production agents actually collaborate with humans under uncertainty. Not just whether they got the answer.

Now you can measure exactly when your agent asks for help, when it makes assumptions, and when it'll confidently ship the wrong answer.

Our findings 🧵

[Tu Trinh (Alina) @thetututrain]

@jennsun one keyboard per hand

[Tu Trinh (Alina) @thetututrain]

Locked in and just managed to sneak in at the end there 💪🏻

[jenny ☀️ @jennsun]

All the winners were startup founders 🤔

Top 3:
1. John Q @johnlqian
WPM: 169
Blindfolded: 78
Twin: 44

2. Alex H https://t.co/nxPlDGdp27
WPM: 151
Blindfolded: 84
Twin: 44

3. Arman K @ksw_arman
WPM: 154
Blindfolded: 88
Twin: 32

Top 10 based on WPM:
1. 179 Kevin L @kliu10
2. 173 Jay J simplifyinterviews
3. 169 John Q @johnlqian
4. 161 Anntaylor
5. 156 Ben S
6. 154 Arman K @ksw_arman
7. 151 Alex H https://t.co/nxPlDGdp27
8. 148 Kevin L @linguinelabs
9. 142 Wonyoung D
10. 140 Alina

[Tu Trinh (Alina) @thetututrain]

@druv_pai @thinkymachines Congrats on the launch!! Thinky is lucky to have you

[Thinking Machines @thinkymachines]

People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.

We share our approach, early results, and a quick look at our model in action.

https://t.co/AFJZ5kH7Ku

[Tu Trinh (Alina) @thetututrain]

@DeepIssueMassaj excited to see how it does on HiL-Bench after 👀

[Tu Trinh (Alina) @thetututrain]

@jehrjd45963 @ScaleAILabs working on adding the latest and greatest! stay tuned ✌️

[Tu Trinh (Alina) @thetututrain]

@FazlBarez @icmlconf Wow insane, congratulations!

[Tu Trinh (Alina) @thetututrain]

@philhchen @tkkong Wooo! something amazing is cooking 🍲

[Tu Trinh (Alina) @thetututrain]

I strongly believe this ability is a huge missing puzzle piece when it comes to fully integrating agents into our workflows. Get them better at selective escalation, and we get one step closer to more reliable human-AI collaboration.

[Tu Trinh (Alina) @thetututrain]

So thrilled to share HiL-Bench, a benchmark to measure how well agents can identify when they need to ask for help in SWE and SQL tasks. Spoiler alert: not too well yet!

[Scale AI @scale_AI]

New @ScaleAILabs Research: Your AI agent just gave you an answer but did it actually solve the problem, get lucky, or just sound right?

Today’s benchmarks can’t tell.

We built HiL-Bench (Human-in-Loop Benchmark) to test a critical skill: does your agent know what it’s missing and when to ask for clarification? 🧵