Parth Asawa (@pgasawa) / X

Parth Asawa

179 posts

Parth Asawa

@pgasawa

CS PhD student @Berkeley_EECS

Berkeley, CA

Joined July 2020

Pinned
Parth Asawa
@pgasawa
Jun 15
The AI community seems to increasingly be heading towards a polarized world when discussing safety and consolidated power. I see this discourse as a false dichotomy, so @profjoeyg and I wrote an essay on how we need to change the conversation (link below).
75K
Parth Asawa
@pgasawa
Oct 6, 2025
Training our advisors was too hard, so we tried to train black-box models like GPT-5 instead. Check out our work: Advisor Models, a training framework that adapts frontier models behind an API to your specific environment, users, or tasks using a smaller, advisor model (1/n)!
104K
Parth Asawa
@pgasawa
Feb 5, 2025
Instruct-tuned models are getting better at following instructions and ‘reasoning’ every day, but they’re shockingly poor at generating diverse responses. Diversity is crucial to many tasks like synthetic data generation. We tackle this with a new approach, BARE 🐻! (1/n)
49K
Parth Asawa
@pgasawa
Aug 31, 2025
Shreya’s work is awesome and she’s an amazing research mentor! Any university would be lucky to have her as faculty :)
Shreya Shankar
@sh_reya
Aug 31, 2025
On my way to VLDB! 🇬🇧 I am on the job market this year, seeking tenure-track CS faculty positions. I will be giving a talk on DocETL and on a panel titled “Where Does Academic Database Research Go From Here?” I would love to meet folks; please reach out if you’re also attending!
7K
Parth Asawa
@pgasawa
Feb 5, 2025
Replying to @pgasawa
🤝 This paper is a fun collaboration with @aczhu1326, along with @jaredq_, @ChenLingjiao, @BorisHanin, Ion Stoica, @profjoeyg, and @matei_zaharia. 📜 Paper: arxiv.org/pdf/2502.01697 💻 Repo: github.com/pgasawa/BARE (code soon) (2/n)
1.5K
Parth Asawa
@pgasawa
Oct 6, 2025
Replying to @pgasawa
📜 Paper: arxiv.org/pdf/2510.02453 💻 Code: github.com/az1326/advisor… This project was co-led with @aczhu1326 and advised by @matei_zaharia, @AlexGDimakis, and @profjoeyg. Reach out to @aczhu1326 and me if you want to chat about interesting applications! (8/n)
2.7K
Parth Asawa
@pgasawa
Nov 19, 2024
Super excited for the world to start using Rox!! Getting AI applications to stick in the real world is hard, but the Rox's drive and principled approach has driven insane traction. Loved my time there this summer and stoked to keep seeing them push agents that empower users. 🚀
Rox
@rox_ai
Nov 19, 2024
We built a B2B SaaS sales company and here’s what it taught us about B2B SaaS sales 🧵👇 (but actually) Today we’re launching Rox, the first publicly available AI agent swarm for the top sales teams, and in the private beta it already helped reps grow their books 30%. 2025 is
00:00
905
Parth Asawa
@pgasawa
Feb 5, 2025
Replying to @pgasawa
💡TL;DR: Instruct models → Higher quality, lower diversity. Base models → Higher diversity, lower quality. BARE → Best of both worlds (or, as @matei_zaharia might say, the BARE necessities). 📧 Feel free to reach out to Alan and myself. Go bares 🐻! (10/n)
586
Parth Asawa
@pgasawa
Feb 5, 2025
I wonder if MAA problem writers know that the 30 problems they wrote for this week and the next are not just for USA(J)MO qualification but will benchmark AI progress for the next year (“AIME25”) 🙃
1.1K
Parth Asawa
@pgasawa
Jan 16, 2024
Check out some of our recent work on automated systems to help make LLMs work in production settings!
Shreya Shankar
@sh_reya
Jan 16, 2024
We all know LLMs make mistakes. One simply cannot deploy LLM pipelines without assertions, yet writing good assertions is tedious & hard. So, we built SPADE, a system that analyzes prompts & auto-generates custom assertions in low-data settings: arxiv.org/abs/2401.03038
5.1K
Parth Asawa
@pgasawa
Mar 5, 2025
If you’re an undergrad at Berkeley, this is one of the best communities of inspiring, kind, and generally awesome people you can join!
Amit Kumar
@amitku
Mar 5, 2025
A simple idea to build the @UCBerkeley startup alumni network has grown beyond my wildest dreams into #AccelScholars, a tight-knit community of the most ambitious, talented, kind-hearted people, whose individual stories we’ve been fortunate to support for the past eight years
1.5K
Parth Asawa
@pgasawa
Feb 5, 2025
Replying to @pgasawa
🐻Base-Refine (BARE) leverages the diversity of base model generations and refines them with more capable instruct-tuned models. Our insight is that by combining these models, we improve the diversity of a dataset while controlling the quality of individual data entries. (5/n)
645
Parth Asawa
@pgasawa
Sep 16, 2025
Super exciting to see how much the @rox_ai team has accomplished!! Engineering, acting, revenue ops -- they seem to do it all 😄
Rox
@rox_ai
Sep 16, 2025
6 months, 25 million revenue agents & 3 trillion tokens later... Rox is now globally available 🌎 Just as coding agents 10x’d engineering, revenue agents 10x customer work. With Rox, humans are evolving to orchestrators while agents manage the end-to-end customer lifecycle.
00:00
1.7K
Parth Asawa
@pgasawa
Feb 5, 2025
Replying to @pgasawa
🎯By fine-tuning on generated data for multiple domains, we find: - RAFT: up to 18.4% improvement over SOTA generation method. - GSM8K: 101% improvement over instruct-only data. - LCB (Test Output Completion): comparable to SOTA models of similar size w/ just 1,000 samples (8/n)
520