Podbit · Moonshots with Peter Diamandis
The Keller-Jordan NanoGPT speedrun has reduced GPT-2 training costs by 99%. Until KIMI K3, nobody knew if that efficiency would scale to frontier models. Now it's proven. A 10-trillion-parameter model that cost Elon Musk billions can theoretically be replicated for 1% of the price — and that realization changes the entire economics of the AI industry.
AI's brand collapse isn't Sam Altman's fault — it's Washington's. When governments don't regulate, don't protect jobs, and let billionaires crowd the White House, ordinary people conclude the game is rigged — and they're right.
AI companies ingested the world's written work without consent, monetized it, and are now selling it back to people about to lose their jobs to the very same technology. There's no PR spin that fixes that story.
Two-thirds of people globally use AI regularly, but willingness to trust it has fallen from 52% to 43% since 2022. Only 10% of US adults are more excited than concerned. The gap between what people use and what they trust is widening fast.
Nuclear power's support fell to 49% in 2019 but climbed back to 61% by 2025 — not through a PR campaign, but because tech giants needed it for AI data centers. AI's redemption arc, if it comes, will follow the same path: necessity first, understanding second.
Most of the world's valuable data sits locked inside enterprise applications, never touching a general model's training set. Fireworks AI was built on the conviction that activating this private data through specialised models is the real frontier of AI.
The AGI believers assume one model will solve everything. Lin Qiao thinks that's both technically wrong and philosophically depressing. The future is millions of specialised models — one per application, per use case, per company.
When Fireworks was founded, open models were in their infancy. Betting on them was a huge gamble. The PyTorch roots gave the team conviction in open ecosystems, and the payoff came as open models crossed quality thresholds that now rival closed models for the vast majority of enterprise use cases.
Fireworks CTO Dima embedded at Cursor for months to build a distributed reinforcement learning infrastructure that decouples the trainer from RL rollout across six global data centre regions. This let a capital-constrained startup run training jobs that previously required 100,000 interconnected chips at a hyperscaler.
Fireworks processes more than 40 trillion tokens a day today. Lin Qiao projects that number could be 20x to 100x higher by end of next year. At those volumes, worries about a CapEx bubble look completely backwards.
We use essential and analytics cookies to run Vuci. To understand how the site is used: Privacy Policy.
Install Vuci on your phone
Add it to your home screen for a faster, app-like experience.
Install Vuci on your phone
Tap the Share button, then “Add to Home Screen”.
A new version is available
Reload to get the latest Vuci.