@hinadixit

Founder & CEO @DecomputeAI , Ex - Investment Partner @Microsoft & @Samsung, SWE AI Leader @Apple, AI @stanford

 Cupertino, CA
decompute.run
Joined June 2009
  • Tweets1.3K
  • Following773
  • Followers2.8K
424Media
BlackBird, you beauty! Comparison side to side: BlackBird vs Perplexity, where BlackBird is completely local and uses much smaller model. Still in Beta but you can request access here: lnkd.in/gXBTtWnX Follow @DecomputeAI for more updates.
322
1,047
37
7,096
32,411,821
By mid-2027, I expect consolidation among undifferentiated inference providers with unhedged compute costs, even as token demand grows. I’d underwrite revenue and compute cost per paid GPU-hour together, not token growth in isolation.
a16z
@a16zOct 24
Zooming out, GPUs are still renting ahead of a standard 6-year depreciation schedule Rental price vs. schedule: - H200: +143% - B200: +111% - RTX PRO 6000: +73% - H100: +54% - A100: +9% Most of the gap opened up after April, when token demand took off Charts of the Week: a16z.news/p/charts-of-the-we…
98
My pov is simple. We need more compute, but at an affordable price, where we don’t make AI a bubble. Excited to introduce our heterogeneous inference’s V0.
Hina Dixit@hinadixitOct 25
I recall being told by a VC, while pitching Decompute, that compute is not a problem anymore, and it has been already solved. 😅
1
1
2
182
I recall being told by a VC, while pitching Decompute, that compute is not a problem anymore, and it has been already solved. 😅
Yuchen Jin
@Yuchenj_UWOct 24
Heard from a neolab friend: “Even if we fail to train the best model, we locked in GPU capacity early. H100/B200/B300 rental prices have since gone up 2–3x. We can just resell the compute and print money.” OK, so that’s the business model.
1
1
339
NVIDIA can sell compute into robotics before humanoids deliver attractive labor economics. I’d separate that opportunity from the much harder business of making a robot pay for itself on a customer’s floor.
NVIDIA Robotics
@NVIDIARoboticsOct 22
We’re excited to announce the official release of NVIDIA Isaac Lab 3.0. 🎉 Isaac Lab 3.0 brings a new modular and lightweight architecture along with major new capabilities across the robot learning workflow: multi-physics and multi-rendering support, training at scale, and streamlined deployment on real hardware. With Newton integration, the new OVRTX renderer, RL post-training, and LEAPP deployment, there’s a lot to explore. Here’s a look at what’s new 🧵 (1/7)
157
I’d value a robotaxi operation on productive vehicle-hours, not the autonomy stack alone. A self-driving car that spends most time being maintained earns nothing while idle. Imagine a fleet operator that improves utilization across driving systems could capture meaningful value.
Tickerade
@tickeradeOct 23
🚗 Hertz $HTZ is heating up! Short interest reportedly hits 72% as borrowing fees surge from the mid-teens to triple digits. Shares have jumped 23% in five days, fueling speculation of a potential short squeeze. 📈 Trading between $1.78–$2.18, Hertz faces a critical catalyst ahead: its Nov. 5 earnings report. An $80M investment in Oro Mobility adds another key factor as short sellers face rising borrowing costs. Will bulls squeeze higher? ⚡
1
1
104
Hina Dixit retweeted
Hina Dixit@hinadixitOct 8
Introducing Kiran-10M: 84.0% average accuracy on BABILong qa1–qa5 at 10M tokens — and 85.6% at 0k. Same frozen 4B-class open-weights LLM, same memory, every length in between inside 84.0–86.4. Ours is the flat one. 1/7
3
2
1
3
581
If inference is 10% of a robot’s operating cost, making it 30% cheaper saves 3% overall. I’d put research dollars into reducing interventions when human support dominates the bill. The operating cost breakdown should shape the model roadmap. 1/2
1
97
Interpretability belongs in the infrastructure budget. This paper uses causal analysis to identify which connections drive robot actions, cutting MolmoBot’s inference compute ~33% with similar simulated performance. 2/2 Paper: arxiv.org/abs/2610.20750

Causal Analysis of Robot Foundation Models Enables Efficiency Gains

Vision-language-action (VLA) models are a promising approach to general purpose robot manipulation. While the internal computation of vision-language models, which compose the backbone of VLAs,...

arxiv.org
34
Margaret was an icon and inspiration! 💔
Massachusetts Institute of Technology (MIT)
@MITOct 7
We’ve lost an icon. Margaret Hamilton, a computer science pioneer best known for her work at MIT leading development of the flight software for the Apollo missions to the moon, has died at 90. Read about her legacy via MIT News: news.mit.edu/2026/margaret-h…
Margaret Hamilton is reclined inside a NASA spacecraft module, surrounded by control panels and switches.

ALT Margaret Hamilton is reclined inside a NASA spacecraft module, surrounded by control panels and switches.

178
AI memory that doesn't decay, doesn't hallucinate its sources, and runs on hardware you already own. Introducing Kiran-10M that can convert any model architecture to beat RAG and baselines on longest context possible with just $30 compute cost! No need to burn tokens anymore. 👀
120
Oh, and one more thing. It costed me only $39.
Hina Dixit@hinadixitOct 8
Introducing Kiran-10M: 84.0% average accuracy on BABILong qa1–qa5 at 10M tokens — and 85.6% at 0k. Same frozen 4B-class open-weights LLM, same memory, every length in between inside 84.0–86.4. Ours is the flat one. 1/7
1
1
244
Introducing Kiran-10M: 84.0% average accuracy on BABILong qa1–qa5 at 10M tokens — and 85.6% at 0k. Same frozen 4B-class open-weights LLM, same memory, every length in between inside 84.0–86.4. Ours is the flat one. 1/7
3
2
1
3
581
more replies
Not claiming today: the exact base checkpoint, speed/memory/hardware figures, anything about the mechanism beyond the diagram. Patented; technical report to follow. 6/7
1
55
To my mum, Kiran Dixit ❤️ 7/7
51
Thank you for covering BF1, Chad! Appreciate it! 🫡 kimbodo.com/how-recent-ai-re…

How Recent AI Research Changes What Product Teams Should Build: Faster, Safer, and More Trustworthy...

A dense wave of papers from academic labs and industry groups reports practical advances across four clusters that matter for production AI: (1) long‑context

kimbodo.com
1
1
100
Thank you for reviewing BF1, James! Checkout our results for Qwen3.8-27B at 1M context as well. james.trappett.org/blog/bf1-…
1
1
109
Load more