Sandeep Chinchali (@SPChinchali) / X

Sandeep Chinchali

127 posts

Sandeep Chinchali

@SPChinchali

UT Austin Prof. Chief Scientist at Poseidon AI. Work on GenAI and Networked Intelligence. Stanford CS PhD. All views my own.

Austin, TX

ece.utexas.edu/people/faculty…

Joined October 2021

Sandeep Chinchali
@SPChinchali
Jul 22, 2025
From Stanford to UT Austin, I’ve spent my career chasing one question: how do we gather and value the right data to make AI work in the real world? There are three battlegrounds in AI: Compute – Solved by Nvidia. Models – Rapidly commoditizing. Data – The last, unpriced
Poseidon
@psdnai
Jul 22, 2025
AI is moving beyond the browser and into the real world. The bottleneck? Data. Today we’re announcing a $15M seed round led by @a16zcrypto to build infra that collects, curates, and licenses high-quality data for physical AI. Incubated by and built on @StoryProtocol.
00:00
20K
Sandeep Chinchali
@SPChinchali
Jul 17, 2025
I’ve spent my career chasing one question: How do we gather the right data to make AI work in the real world? From Stanford labs to UT Austin classrooms, I searched everywhere. The answer isn’t another AI lab, but a blockchain built to treat data as IP. That’s why I am joining
43K
Sandeep Chinchali
@SPChinchali
Mar 24, 2024
Curious about the value proposition of #DecentralizedAI? Let's go over a popular project #bittensor and describe why it’s crucial to have external demand and real utility for AI services.
48K
Sandeep Chinchali
@SPChinchali
Jul 25, 2025
The last generation of AI was trained on what was easy to scrape. The next generation will be trained on what’s hard to coordinate. → First-person stereo video → Multilingual voice in real-world environments → Edge-case navigation footage → Biometric signals from diverse
Poseidon
@psdnai
Jul 25, 2025
Scraped internet data got us this far, but the next generation of AI – robots, AVs, voice-to-text – needs something else entirely: Real-world, long-tail, multimodal data. No one has been able to solve the problem of how to source this data at scale, until now.
11K
Sandeep Chinchali
@SPChinchali
Jul 28, 2025
Right now language models can write exams, but embodied AI still fails in kitchens, warehouses, and hospitals. The adjective “robotic” brings a vision to mind that isn’t particularly smooth or capable. The issue isn’t how advanced the model is, but the quality of the data it’s
4.5K
Sandeep Chinchali
@SPChinchali
Jul 24, 2025
The leading AI teams don’t just need high-quality data. They need data that’s IP-cleared and ready for commercial use. That’s what @psdnai makes possible by building on @StoryProtocol – every dataset is ready for commercial use and structured for integration into training
16K
Sandeep Chinchali
@SPChinchali
Jul 29, 2025
With Nvidia, OpenAI, Anthropic, xAI, and co in the lead, the race to solve compute and model architectures is largely over. Everyone has open source model weights and GPU budgets. The real challenge now is coordinating high-signal, IP-cleared data from the physical world: •
Poseidon
@psdnai
Jul 29, 2025
While building robotics models at Stanford, @SPChinchali saw a pattern: It wasn’t the 99% of data that dramatically improved performance. It was the rare 1% – edge cases and long-tail data. Poseidon is built to make that data usable for AI teams. More on the blog:
4.8K
Sandeep Chinchali
@SPChinchali
Aug 12, 2025
Last week, I had the chance to meet with Kim In-yeop from @kedglobal for a great discussion about what we’re working on at @psdnai. Here’s a translated quote from the piece: While companies like Nvidia and AMD are proving prominent in the AI infrastructure market, and OpenAI
2.5K
Sandeep Chinchali
@SPChinchali
Jul 11, 2025
Super sharp move by the Story team. Data and labels are quickly emerging as the most valuable form of IP in AI, as models become open source and architectures commoditize.
S.Y. Lee
@storysylee
Jul 10, 2025
<Story’s Chapter 2: AI-Native Infrastructure for the $70T IP Economy> Today, we’re sharing what’s next for Story. We call it Chapter 2, which includes expanding verticals of IP tokenization, flipping the licensing model, and an evolution of our infrastructure to meet the most
26K
Sandeep Chinchali
@SPChinchali
Jul 17, 2025
Replying to @SPChinchali
Full article:
Why I Joined Story as Chief AI Officer
From sandeepchinchali.medium.com
904
Sandeep Chinchali
@SPChinchali
Jun 18, 2024
Excited to share our new paper in @icmlconf on #synthetic time series generation using #diffusion models!
3.3K
Sandeep Chinchali
@SPChinchali
Jul 11, 2025
Replying to @SPChinchali
Story’s blockchain offers the rails to easily register and tokenize that data while rewarding the people who generate it to align incentives for a world increasingly shaped by AI. Big vision from a team I was lucky to meet when they brought me out to speak in Palo Alto.
681
Sandeep Chinchali
@SPChinchali
Jul 24, 2025
Replying to @SPChinchali
You don’t just mitigate risk, you unlock new, functional markets for POV video in homes across worldwide, voice data in low-resource dialects, and so on. These markets haven’t existed, not because the data isn’t valuable, but because there was no way to structure and license
337
Sandeep Chinchali
@SPChinchali
Jul 25, 2025
Replying to @SPChinchali
This data can’t be pulled from public datasets. It has to be collected with IP-clearance and licensing. That’s why we built @psdnai on top of @StoryProtocol. Via Story, licenses are machine-readable, programmable, and legally enforceable.
248