TickerTain — Make Analysis Fun Again

▶ 7:25 · $DEEPSEEK · Dwarkesh Patel: DeepSeek v3 requires 2400+ batch size for inference efficiency

now playing · $DEEPSEEK

Dwarkesh Patel: Continual learning will eliminate internal deployment gaps like Anthropic's four-month Mythos delay

Continual learning will force AI labs to deploy their smartest models immediately because competitors who ship earlier will accumulate real-world experience and surpass them, elim…

$DEEPSEEKneutral· mediumdwarkesh patel

Dwarkesh Patel: DeepSeek v3 requires 2400+ batch size for inference efficiency

Sparse models like DeepSeek v3 require batch sizes exceeding 2400 concurrent sequences for compute efficiency, creating massive economies of scale that favor large organizations s…

←

▶ 7:25 · $DEEPSEEK · Dwarkesh Patel: DeepSeek v3 requires 2400+ batch size for inference efficiency

now playing · $DEEPSEEK

$ANTHROPICneutral· mediumdwarkesh patel

Dwarkesh Patel: Continual learning will eliminate internal deployment gaps like Anthropic's four-month Mythos delay

Continual learning will force AI labs to deploy their smartest models immediately because competitors who ship earlier will accumulate real-world experience and surpass them, elim…

$DEEPSEEKneutral· mediumdwarkesh patel

Dwarkesh Patel: DeepSeek v3 requires 2400+ batch size for inference efficiency

Sparse models like DeepSeek v3 require batch sizes exceeding 2400 concurrent sequences for compute efficiency, creating massive economies of scale that favor large organizations s…