Results for “scaling”

1,628 results

Transcript mentions 753

See all ›

State of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI | Lex Fridman Podcast #490

Lex Fridman133 views
17:08extended thinking. Extended thinking and inference-time scaling is just a way to make the models marginally smarter. And I will always48:06- I guess the big question here is, we talked quite a bit about the architecture behind the pre-training. Are the scaling laws holding strong across pre-training,48:16post-training, inference, context size, data, synthetic data? - I'd like to start with the technical definition of a scaling law—

Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494

Lex Fridman56 views
1:56not just scaling up the problem, but you're distributing the problem, then everything gets in the way. This is the Amdahl's Law3:09scale up based on the capabilities of Moore's Law, which has largely slowed because Dennard scaling has slowed.22:39- Yeah. So one of the things you've been a believer for a long time is scaling laws, broadly defined. So are you still a believer in the scaling laws?

Topics & chapters 594

See all ›

AI summaries 281

See all ›

Confronting Andrei Jikh | Can Bitcoin Reach $1 Million

The Iced Coffee Hour54 views
The hosts discuss cryptocurrency investment strategies, emphasizing the importance of self-custody over brokerages and explaining Bitcoin's scaling solutions like the Lightning Network.summaryBitcoin Scaling Solutions: Bitcoin's base layer is slow due to 1MB block limits and 10-minute confirmation times. The primary solution is 'off-chain' scaling via the Lightning Network, which allows instant, cheap transactakeawayBitcoin Cash vs. Bitcoin Core: The split occurred over scaling philosophy. Bitcoin Cash chose 'on-chain' scaling (larger blocks), while Bitcoin Core chose 'off-chain' scaling (Lightning Network) to preserve decentralizattakeaway

Why Top Founders Are Racing Into AI Infrastructure

a16z7 views
Key insights highlight that AI scaling has shifted from an engineering problem to a resource limitation, necessitating bespoke hardware solutions like ASICs and new data center architectures, while noting a trend toward summaryShift in Scaling Laws: AI scaling has moved from an 'engineering problem' (where adding engineers hits diminishing returns) to a 'resource limitation' where pouring more money and compute directly yields better results, takeaway

JUST RECORDED: Elon Musk Announces SPACEX Plans

Brighter with Herbert24 views
The plan involves scaling mass-to-orbit capabilities to millions of tons annually, manufacturing specialized AI satellites using existing Starlink technology, and eventually establishing a lunar mass driver to produce insummaryCompute Scaling Roadmap: The target is 1 GW/year of space AI compute by end of next year, scaling to 10 GW in 2.5 years, 100 GW in 3.5 years, and ultimately 1 TW/year via the 'Terafab' chip factory.takeaway

Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494

Lex Fridman56 views
Extreme Co-Design is Necessary: Scaling AI requires optimizing the entire stack (GPU, CPU, memory, networking, power, cooling) because distributed computing introduces Amdahl's Law bottlenecks. NVIDIA designs systems liktakeawayAgentic Scaling Law: Beyond pre-training and inference scaling, the next phase is 'agentic scaling' where AI agents spawn sub-agents to perform tasks. This multiplies compute requirements and creates a feedback loop whertakeaway

Elon Musk latest interview (AI, Robots, Space Data Center, DOGE)

iWealthyfox39 views
Elon Musk argues that scaling AI compute on Earth is physically constrained by electricity generation and permitting, making space-based solar power the only viable path for terawatt-scale AI.summaryStarship Scale: To achieve 100GW of solar/AI capacity in space annually requires ~10,000 Starship launches per year (one every hour). This requires scaling to 10,000-30,000 launches annually using a small fleet of 20-30 takeawayOptimus Production: Optimus robots are designed from first principles with custom actuators. Production will follow an S-curve, starting with simple tasks and scaling to millions per year. They will outperform human corptakeaway