Gaurav Sahu (@dem_fier) header
Gaurav Sahu (@dem_fier) profile picture

Co-founder & CEO @unifiedsciences Building the future of AI and Science. @Mila_Quebec · @UWaterloo · @IITKgp

Joined October 2023
The entire industry is racing to make AI cheaper. Most people assume the answer is better chips or bigger GPU clusters. However, while the industry is buying its next 2× when there's a 1.78× speedup is sitting in the software. We've been hard at work @unifiedsciences, building an autonomous optimization engine toward radically more efficient compute. Today I'm very happy to share some early results where our unified engine: 1) wrote faster GEMM kernels for cuBLAS (up to 1.78x faster) 2) made an efficient AI model 6x smaller and 5x faster without any quality loss 3) optimized diverse everyday AI workloads on a consumer GPU!
3
14
3
36
15,345
Gaurav Sahu retweeted
Most think vibecoding at scale is slop. Here’s a counterexample. Today we release a 16M-line formalization of the entire Stacks Project, critical infra underlying modern math, with optimized code quality and its highly interconnected structure preserved. A mathematical city 🌆
857
1,541
330
7,006
7,004,903
Gaurav Sahu retweeted
I've been reviewing a lot of code written by Opus 5.5, and it still seems to make a lot of bad software engineering decisions. The kind that will accumulate/calcify into unmaintainable software over time. IMO even the agents will struggle to maintain vibecoded software.
423
117
61
3,304
329,422
this is cool!!
📣 Call for Blog Posts at #ICLR2027 @iclr_conf Following the success of the past iterations, we are opening the Call for Blog Posts 2027! iclr-blogposts.github.io/202… Please retweet! Thread 👇
1
93
📣 Call for Blog Posts at #ICLR2027 @iclr_conf Following the success of the past iterations, we are opening the Call for Blog Posts 2027! iclr-blogposts.github.io/202… Please retweet! Thread 👇
2
12
1
65
6,680
Gaurav Sahu retweeted
📣 I'm hiring a Ph.D student through ELLIS to join my group at the University of Copenhagen. We work on tokenization-free language models (think: bytes, characters, and pixels). Your office will be at the Østervold Observatory in the beautiful Copenhagen Botanical Gardens.
4
38
1
247
9,070
Gaurav Sahu retweeted
OpenReview has informed us that they're experiencing technical difficulties with the ICLR submission site, and that it has been temporarily disabled. They are working on fixing it and we'll make another announcement as soon as it's back online.
16
16
8
365
114,659
A fine research opportunity with a fine researcher—you’ll enjoy your time, highly recommend!
1/ Hiring research positions at A*STAR's Centre for Frontier AI Research 🇸🇬 Working on differential privacy & safe AI: private training, synthetic data generation, de-anonymization attacks.
1
187
1/ Hiring research positions at A*STAR's Centre for Frontier AI Research 🇸🇬 Working on differential privacy & safe AI: private training, synthetic data generation, de-anonymization attacks.
3
10
2
116
15,591
2 new terms to add to my vocab. nice piece by the @COLM_conf PCs!
📢 New blog post describing the COLM 2026 PCs’ analyses of AI use in submitted papers. Link below 👇
2
164
my agents can finally talk courtesy @xhluca's agent-talk!
Introducing agent-talk, a new way for coding agents to work together. Through agent-talk, agents can message each other directly, enabling coordination on projects spanning multiple developers & agents. You no longer need to be the messenger between your agents.
1
7
821
Gaurav Sahu retweeted
Introducing agent-talk, a new way for coding agents to work together. Through agent-talk, agents can message each other directly, enabling coordination on projects spanning multiple developers & agents. You no longer need to be the messenger between your agents.
5
23
8
78
13,669
Gaurav Sahu retweeted
Some pretty amazing results by @dem_fier and the team at @unifiedsciences. The same engine found improvements across different problems: cuBLAS, a full model, NeRF rendering, neural search, and matrix inversion - all of which were verified by the system before being accepted.
1
2
3
571
Gaurav Sahu retweeted
Replying to @lcharlin
Congratulations Laurent and @dem_fier, rooting for your success and to discover many such findings!
1
1
4
249
Gaurav Sahu retweeted
We’re coming out of stealth with some exciting AI-optimized compute results. Some highlights: 1) Works on industrial-grade and consumer-grade GPUs. 2) It can optimize different compute layers, from the kernels to tokens, for higher tokens-per-watt on your GPU.
The entire industry is racing to make AI cheaper. Most people assume the answer is better chips or bigger GPU clusters. However, while the industry is buying its next 2× when there's a 1.78× speedup is sitting in the software. We've been hard at work @unifiedsciences, building an autonomous optimization engine toward radically more efficient compute. Today I'm very happy to share some early results where our unified engine: 1) wrote faster GEMM kernels for cuBLAS (up to 1.78x faster) 2) made an efficient AI model 6x smaller and 5x faster without any quality loss 3) optimized diverse everyday AI workloads on a consumer GPU!
4
9
25
8,453
Truly remarkable result! 🔥Can't wait to see what more is to come from @unifiedsciences!
The entire industry is racing to make AI cheaper. Most people assume the answer is better chips or bigger GPU clusters. However, while the industry is buying its next 2× when there's a 1.78× speedup is sitting in the software. We've been hard at work @unifiedsciences, building an autonomous optimization engine toward radically more efficient compute. Today I'm very happy to share some early results where our unified engine: 1) wrote faster GEMM kernels for cuBLAS (up to 1.78x faster) 2) made an efficient AI model 6x smaller and 5x faster without any quality loss 3) optimized diverse everyday AI workloads on a consumer GPU!
1
1
2
708
Gaurav Sahu retweeted
Great work by @unifiedsciences! The interesting part for me isn't the speedups, but that reproducibility and correctness checks are built into the search itself, not bolted on after, which in my opinion is the right way to do automated optimization. Kudos @dem_fier !
2
2
2
626
thanks @AbhayPuri98 :)
Absolutely insane numbers 🤯📈 Huge shoutout to @dem_fier for leading this. Amazing work! 💯🔥
207
One lesson that stood out: finding optimizations is getting easier. Knowing which ones are real is getting harder. Every claimed improvement was verified against the real baseline – from T4s to Blackwell – before the system accepts it. We think trustworthy, autonomous optimization will become an important part of how future AI systems are built. If you operate large GPU fleets, want to get more out of your compute, want to help build this future, DMs are open! Full write-up below (along with open-sourced artifacts): nitter.cf/unifiedsciences/status…
1
476