ML nerd. Experimenting with life. Currently vibing at https://xitter.cf/t.co/BqWEhjLINI

Present
Joined October 2020
Today I’m excited to introduce VibeEdit - vibeedit.app It lets you edit any content - image, video, audio, just using text prompts.
3
2
1
7
1,387
Do you remember when you joined X? I do! #MyXAnniversary
4
571
Balaji Varatharajan retweeted
avid readers = could be decent gpu programmers I post a lot of technical CS things but at Stanford, I was actually set on being an English major or something adjacent to that! so you can understand why I chose to be part of an intense reading humanities program freshman year. I spent probably 20 or more hours a week reading (it was 8 units so technically like 24 hours) from the old testament to zhuangzi. I think we were supposed to read the old testament in a week (I think I was like 80% there lol) reading became a default thing to do and it wasn't until I got into CS and more so reading docs for gpu programming that I realized how useful it is to be able to read things that you're not in the mood to read and also very poorly written (side effect of too much AI everywhere). people who read would be at least decent at gpu programming (if they actually tried). I don’t think it matters what they like reading (I mean what genre of literature prepares you to read nvidia docs ??), reading is a general skill that I’ve come to realize people don’t have anymore. this entire thought came to me as I was preparing to write another thread about new nvidia features for CuTe DSL kernels. it's surprising how much you remember if you read documentation fully yourself rather than putting it into chat or claude to summarize it for you. maybe there's a psychology study for this but it reminds me of how taking notes pen and paper is better for memory than typing it up or studying someone else's notes! there's also something about reading a good blog from someone who really knows what they're talking about where you end up mirroring their phrasing or their analogies as you work through problems later. sometimes I find when I don't do this, I can barely explain what I'm thinking and end up sounding incredibly funky 🤮 (anyways here's a picture of a part of the gigantic pile of books I now have)
12
4
192
7,535
Balaji Varatharajan retweeted
Everything that compounds is boring at the deposit. Applies to love, health and kindness.
3
9
431
Balaji Varatharajan retweeted
Hey there! I am looking for Product Intern #Hiring #Bengaluru #Hyderabad #Internship
2
2
2
264
Balaji Varatharajan retweeted
It is a hard and sad decision. I shared this message with folks at Thinky. Thank you all for the time together♥️ Just as the last sentence in my message: The future worth building is human.
379
228
176
9,119
2,184,768
Balaji Varatharajan retweeted
It's been 2 days in korea (1 in Seoul & 1 in Gangneung) and I'm already awestruck by the kindness of Korean people. 1/2 A bystander gifted me their portable charger unprompted, just because they saw me struggling to charge my phone at a public charging booth. 2/2 Another bystander went out of their way to help me navigate bus schedules, walked across the bus station (in 85 F + 12 pm sun) to talk with a bus driver for helping me.
1
2
1
27
1,880
Balaji Varatharajan retweeted
9-5 is stressful. So is unemployment. Being single is stressful. So is having one. Having 50 clients is stressful. So is having none. Life is full of trade-offs. Don’t romanticise options you don’t know.
21
124
4
793
24,257
Balaji Varatharajan retweeted
kinda surprising how obviously effective this is but nobody really does this consistently. Who is doing this ? ( im sorry im not doing this anymore because of obvious reasons...)
Don’t get degrees man. Just implement papers, tweet about it, write good blog posts, create nice repos. Biglabs will come and beg you to work for them.
13
8
1
398
48,474
Balaji Varatharajan retweeted
We taught a brand-new mini-series this year at @SCSatCMU on Modern GPU Programming for ML Systems, as part of the ML Systems course, touching on fun questions like what data layout swizzling is, how to use 3D TMA, and state-of-the-art Blackwell programming. We released a curated online book based on the materials: mlc.ai/modern-gpu-programmin… check it out
24
289
6
1,826
205,277
Balaji Varatharajan retweeted
A bit of news: After nearly 9 years, I have decided to leave Google DeepMind and join Anthropic (after taking some time to recharge). I am incredibly grateful for my time at GDM. @demishassabis took a real chance letting me lead the AlphaFold team just six months after finishing my PhD, and the entire GDM team taught me so much about how to do great science. GDM is a special place, and I’ll still be excited to hear about what amazing things they discover next.
600
962
738
14,387
6,192,331
Balaji Varatharajan retweeted
Here is the technical report on SubQ 1.1 Small. subq.ai/subq-1-1-small-techn… This is the second iteration on our Subquadratic Sparse Attention (SSA) model, and the first to be deployed with design partners in the coming weeks. The results are compelling and verified by @AppenResearch. - Near-perfect long-context retrieval up to 12M tokens on the needle-in-a-haystack test, with up to nearly 1,000x attention compute reduction. - A balance of long-context optimization and general reasoning ability, with strong performance retained across knowledge, coding, and non-coding enterprise agent benchmarks. - At 1M tokens, SubQ 1.1 Small requires 64.5x less compute than dense attention and runs 56x faster than FlashAttention-2. These results highlight a significant scaling advantage thanks to the efficiency gains from the SSA architecture. We included some details and learnings from the development process which may be helpful to the community. Comment with questions, I’ll try to respond!
67
70
46
540
154,340
Balaji Varatharajan retweeted
We're thrilled to announce that we have raised $234M in the first close of our $300M Series B at a $1.5B valuation. @HCLTech and @BessemerVP have joined us in this round, alongside continued support from @khoslaventures and @peakxvpartners For countries and companies, sovereign control on the AI stack is no longer an optionality. Sarvam will be the partner of choice for this aspiration. The capital allows us to accelerate our momentum towards this full stack of models, compute, and deployments. A huge thank you to our customers, partners, investors, and the Sarvam team for your trust and belief in what we are building. We’re just getting started. Read more: sarvam.ai/announcing-series-…
648
1,485
416
9,873
1,110,084
Balaji Varatharajan retweeted
Fable 5 is out now but before that, it had its model card updated: stampr-ai.com/model-details.… The doc's changelog is mostly accurate this time but eg missed mentioning the removal of this footnote: "This threshold maps to the High-stakes sabotage opportunities threat model in our current Responsible Scaling Policy."
1
1
2
16,909
Balaji Varatharajan retweeted
one of the quotes i find most inspiring on a hard day: "Whatever your hand finds to do, do it with all your might, for in the realm of the dead, where you are going, there is neither working nor planning nor knowledge nor wisdom" Ecclesiastes 9:10
1,215
2,624
372
19,222
1,839,999
I got my first customer for @vibeeditapp about a month ago 🤗
that first annual plan sale, when was it for you?
1
2
101
Balaji Varatharajan retweeted
Introducing DiffusionBlocks: Block-wise Neural Network Training via Diffusion Interpretation pub.sakana.ai/diffusionblock… What if we didn’t have to hold an entire neural network in memory to train it? Standard neural net training optimizes all parameters jointly. As a result, the memory required during training grows linearly with the depth of the network. In our #ICLR2026 paper, we propose DiffusionBlocks, a principled framework to train networks one block at a time, drastically reducing memory requirements while matching end-to-end performance. With DiffusionBlocks, we split the network into blocks and train them one at a time, so you only need memory for a single block. How? We explicitly assign each block a role: to move the representation a little closer to the target than the block before it did. That role turns out to be precisely what a diffusion model does, step by step. Each block only needs to optimize its own objective and can be trained independently. We validated this across five different architectures: • ViT • DiT • Masked diffusion • Autoregressive transformers • Recurrent-depth transformers In each case, performance is competitive with end-to-end training while using a fraction of the memory. This perspective also extends naturally to recurrent-depth (Looped) transformers, which apply the same network iteratively and normally require expensive backpropagation through time (BPTT). Viewed through DiffusionBlocks, we can replace those multiple iterations with a single forward pass during training. Read our paper and code, to learn more. Paper: arxiv.org/abs/2506.14202 GitHub: github.com/SakanaAI/Diffusio… 🐟
56
349
100
2,297
882,917
Balaji Varatharajan retweeted
Supersonic. Mach 1.21. Quarterhorse Mk 2.1 is now the world’s first privately developed, unmanned supersonic jet and the fastest unmanned aircraft flying today. This flight makes Hermeus the fastest company in aviation history to go from founding to supersonic flight - exactly 364 days after the maiden flight of our first aircraft. Now, we fly faster. A special thanks to @DIU_x, Director @OwenWest91, Maj. Gen. Joe "Solo" Kunkel, and Deputy Director Kyle Norman.
66
194
50
1,555
784,664
Balaji Varatharajan retweeted
3
1
10
788
Balaji Varatharajan retweeted
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
7,877
10,921
7,548
149,097
28,074,863