SparX

The Vinod Khosla Episode 4:06 - He wrote OpenAI $50M in 2019, then apologised to his own investors for it 7:09 - Fusion. Hypersonic jets. Public transit. The bets everyone told him were stupid 9:07 - He's never once run a spreadsheet on a startup. He explains why they lie 10:41 - Lose 1x when he's wrong, make 100x when he's right. That asymmetry is his whole career 12:59 - Everyone's excited about data centres in space. He thinks they're pointless 37:32 - He gives fusion 80% odds in 5 years. Almost every expert disagrees 47:44 - "AI will ultimately free humanity to be human beings, not servitude to survival" - Khosla You'll find us as "SparX Podcast" on YouTube, Spotify or Apple Podcasts.

@vkhosla thank you for the conversation. Rare to hear someone this direct about what they actually believe.

The Vinod Khosla Episode 4:06 - He wrote OpenAI $50M in 2019, then apologised to his own investors for it 7:09 - Fusion. Hypersonic jets. Public transit. The bets everyone told him were stupid 9:07 - He's never once run a spreadsheet on a startup. He explains why they lie 10:41 - Lose 1x when he's wrong, make 100x when he's right. That asymmetry is his whole career 12:59 - Everyone's excited about data centres in space. He thinks they're pointless 37:32 - He gives fusion 80% odds in 5 years. Almost every expert disagrees 47:44 - "AI will ultimately free humanity to be human beings, not servitude to survival" - Khosla You'll find us as "SparX Podcast" on YouTube, Spotify or Apple Podcasts.

Excited to see the frontier technologies being built in India.

The whole nation awaits tomorrow. A remarkable day for the country. Wishing @PawanKChandana and the entire @SkyrootA team all the best. Jai Hind 🇮🇳

We sat down with the founders of Skyroot. Asked what India needs to lead globally in space, Pawan Chandana named 3 things: 1. Talent, first. India produces more engineers annually than any country on earth, and he wants more of them entering the sector. 2. Second, deeper collaboration between the government and Indian private companies, building missions together rather than in parallel. 3. Third, more joint missions between India and the rest of the world.

India attempts its first private orbital launch tomorrow morning. @SkyrootA @PawanKChandana

In 1963, India launched its first rocket from a fishing village called Thumba. The parts came to the pad on a bicycle. Vikram Sarabhai had almost nothing except the conviction that a poor country had every right to reach for space. Tomorrow at 11:30 AM IST, a rocket named after him attempts orbit from Sriharikota. It was built by a private company that was two people and an idea eight years ago, in a country that had no legal framework to allow it. They named the mission Aagaman which means Arrival. 63 years from a bicycle at Thumba to a private Indian orbital rocket on ISRO's historic First Launch Pad. Every generation of engineers in between built the thing that let the next one go further. Whatever happens in those nine minutes tomorrow, that lineage is the achievement. The rocket is just the proof 🇮🇳

Sarvam is offering startups 12 months of free API credits, 22 Indian languages, and priority engineering access. Link here - sarvam.ai/startup-program

"Made in India, which actually means made in India, including the components." QpiAI's 64-qubit Kaveri gets the headline. Whether the components underneath it are Indian is the part that decides if the phrase is true.

Why are Chinese LLMs open-source compared to U.S. models?

Chinese open models = 17% of global downloads, overtaking US

Humanity is split into Consumers & Creators.

Is computer science dead? CS graduates now sit at 6.1% unemployment.

@SarvamAI was picked from 67 companies to build India's sovereign LLM, got ₹99 crore in GPU subsidies, and open-sourced two models in February. Now it needs builders, because a national model without an ecosystem is an expensive research project. Free credits are distribution, not generosity. Still worth the look if India is your market.

The pitch underneath it is more interesting than the giveaway. Global tokenizers are built for English, so Indian scripts fragment into far more tokens than an equivalent English sentence. Sarvam's are reportedly 4x to 8x more efficient. Every startup running Hindi workloads on a global model has been paying a tax that never appeared on a pricing page. Which is also why cost per million tokens is a bad unit. A token is not a fixed quantity of meaning. It depends on the language.