Avinash (@avinashj_) header
Avinash (@avinashj_) profile picture
New York
Joined July 2011
Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)Photo by Avinash (@avinashj_)
(3/4 releases) Today we share RareDiseaseBench, a series of 35 diagnostic workups reconstructed from published rare disease cases. In each task, an agent starts with a patient’s presentation, decides which tests to request and why, analyzes the data it receives, and uses the results to propose a diagnosis and management plan over one or more visits. Gemini 4 Argon does the best on this benchmark (43.8% attempt pass rate), with Opus 5.5 and Astra following. Paper: latch.bio/raredisease Results: benchmarks.bio/raredisease More details in thread:
2
13
2
55
4,095
Avinash retweeted
Really excited to share some of what I’ve been working on lately! Language models are remarkably general robot policies, but on their own are far too slow and unreliable to deploy. This work asks the question: can we extract a superior task-specific policy from a generalist language model? With enough effort, the answer is yes! Compared with the base LLM, our policies double success on three of four tasks and finish tasks 4-8x faster, even before counting LLM inference latency. Details and post in the thread below (1/n)...
9
16
2
116
7,556
Today, we are announcing an agreement to acquire ADT. Home security has the qualities we look for in the businesses we own: an essential service built on specialist expertise and earned trust. We see an opportunity to build on that foundation for decades, investing in its people and technology to deliver the best outcomes for customers. Our ambition is to help ADT define the next era of trusted home security. thriveholdings.com/thrive-ho…
37
38
64
972
331,886
yesterday was a surreal day for @hexanalytics! 1. we raised 100 million dollars 2. we hosted our first ever user conference 3. we launched Hex 4.0 during our product keynote I got to share more about 4.0: a crazy ambitious rewrite of our entire product from scratch to be truly agent native, fully reviewable, and extremely fast
22
4
1
85
4,140
Choosing the right AI model is an operating decision. Today, we’re sharing our first set of benchmarks drawn from work across the 75+ accounting and IT services firms that make up Thrive Holdings. We believe the field needs more benchmarks tied to economically valuable work, so builders can understand where models are useful in real-world environments. We use these tests to help us compare model configurations and choose which models our systems run on, with the goal of delivering better service to our customers. We’ll add to the benchmarks as we build new products and expand to additional industries. evals.thriveholdings.com/
3
19
11
141
26,849
In the early 1990s, the accounting firm RRBB owned four computers for a staff of around 30. They called them “lunch boxes”: 30-pound machines that auditors lugged through airports and into client offices. The New Jersey firm was founded more than 65 years ago and has adapted through each new wave of technological transformation, from writing tax returns on green ledger paper to asking software to auto-sort numbers. In 2024, they became the first firm to join our accounting platform, Current. The firm’s history is less a story of computers replacing practitioners than of practitioners repeatedly learning what a computer can’t do on its own. This is the first in a series about the people and firms rethinking professional services around AI. Here is RRBB’s story: thriveholdings.com/the-pione…
4
17
9
176
117,561
We’ve raised over $2B in additional capital from investors including D1 Capital Partners and Altimeter Capital. Holdings now owns and operates 70+ businesses across accounting and IT services, where our engineers build and deploy AI products used every day to serve tens of thousands of customers. We’re also launching a new platform focused on the technical and regulatory work behind physical assets. We want to make it easier to modernize and build America’s infrastructure. thriveholdings.com/thrive-ho…
11
24
15
469
480,885
Avinash retweeted
introducing Laurie: turn your laptop keyboard into an instrument compose ideas, strum with your trackpad, and play straight into your DAW of choice. all in the browser live now at: laurie.fm ◇ × ✧ *
27
15
10
209
54,791
Avinash retweeted
Personal update, I started a company earlier this year with @rolandtshen! It’s been an amazing ride at @tryramp, joining as a founding member of the Workflows team when we were still in a WeWork, to seeing the company grow to $44B, 75,000 customers, and a team of 2000. Grateful for the lifelong friendships and for all the lessons I’ll take forward from seeing a generational company built on the inside: the bias towards constantly shipping, building an incredibly talent-dense team, and always scaling your ambition. We’re tackling one of the largest verticals out there and are already helping early partners rebuild operational workflows in an AI-native way. Our team today is ex-founders and staff engineers. We’re growing and looking for folks that thrive in an environment that optimizes for velocity. If this sounds exciting to you or a friend, please reach out!
95
24
7
228
34,352
Why you should never bet on yourself. Tomo, launching june 25th
140
57
23
652
99,099
Avinash retweeted
I’m an AI researcher turned brain tumor patient, and recently I used the models to crack my mystery fatigue faster than my PCP could. I believe everyone can do the same with their own symptoms. Here’s how:
128
323
86
3,438
847,797
1/ I’m excited to share that I’ll be taking on an additional role as Head of Research for Thrive Holdings, alongside my work at OpenAI.
52
29
12
1,178
149,530
in markets, to go long on something is to bet it grows more valuable over time. much of the conversation today is short on humans, wagering that ai makes people redundant. we believe the opposite is true for the industries @ThriveHoldings operates in. we are long humans. thriveholdings.com/long-huma…
100
128
86
2,102
628,049
Avinash retweeted
At @ThriveHoldings, we built a product with @OpenAI to automate tax prep for the 30+ accounting firms we own across the country. This season, it processed 7k+ returns. But what I think is more interesting is that the product meaningfully self-improved as accountants used it.
34
51
17
1,078
348,837
Avinash retweeted
New preprint with @J_Pablo_Bonilla! We built a neural network that decodes quantum errors — and discovered that fault-tolerant quantum computing might need far fewer qubits than we thought. 📄 Paper: arxiv.org/abs/2604.08358 Website: andigu.github.io/projects/ca…
3
23
1
144
7,125
Reliving my recent japan trip through MinecraftLM (github.com/mattzh72/minecraf…), our world building agent. Some examples in 🧵 Prompt: "A traditional Japanese onsen as an isometric diorama. A small wooden bathhouse building with tiled roof, noren curtains at the entrance, and shoe lockers visible inside. Behind it, an outdoor hot spring pool surrounded by natural rocks, steam rising from the water. Wooden deck connecting the bathhouse to the pool with a small changing area, towel racks, and wooden buckets. Bamboo fence enclosing the space, stone lanterns along a path, maple trees with red leaves overhead. Snow on the ground and rocks for the classic winter onsen look. The bathhouse glows warm from inside, contrasting with the cold outside."
MinecraftLM is one of the most impressive agents I've ever seen... and we've open sourced all our work. LLMs aren’t known for deep spatial reasoning, but here we have one iteratively building a pirate ship into an impressive structure. Find other demos and our repository below.
2
3
1
10
3,918
1/ All the structures you see here were generated by an LLM from a single text prompt (notably, Gemini-3 👀). Over the past week, we’ve been researching the spatial understanding capabilities of agents and how they can build large scale, playable environments (particularly Minecraft to start). We found they do extremely well despite what many think. We're open-sourcing our work for others to build on top of github.com/mattzh72/minecraf…
3
3
1
18
1,440
1/ I’m excited to share recent work on inferring cell-cell interactions using attention by @justjhong and me, supervised by @elhamazizi. Open the thread 🧵for a brief tweetorial of our method. bioRxiv link: biorxiv.org/content/10.1101/…
2
30
4
188
25,002