If you thought David Sinclair was the only interesting thing here, look again because MIT Tech Review just laid out a massive story at the intersection of longevity and AI that I can't stop thinking about! First, there's Longevity Park in South San Francisco — Cal Yerington built it specifically as a dedicated campus for aging research, which is already an incredible commitment. Meanwhile, David Sinclair isn't just talking; he's actually entered the AgeX consortium, a multi-institution effort competing for $50 million to develop whole-body rejuvenation drugs and cellular therapies at scale. That's not incremental — that's building entire cities and companies around longevity as an engineering problem rather than a niche medical interest!
Then there's the AI side of this which is just mindblowing when you dig into the numbers they cite. We're talking about LLMs being trained on $1 billion worth of compute *every year*, with the newest models hitting roughly 60 TFLOP-days to train — and that doesn't account for inference costs! They also break down generative versus discriminative AI in a way I wish everyone understood: discriminative models classify, while LLMs generate based on probability. And they make a crucial point about model size vs quality — bigger isn't always better; smaller models can outperform massive ones when trained on high-quality data instead of the general internet sludge that many modern systems consume.
But here's what really gets me: they frame these not as separate stories but as one story, where progress in protein folding and genomic modeling accelerates longevity research by decades through AI, while advancements in age reversal create new generations of users for computing infrastructure. They warn us to take AGI timelines seriously rather than casually — predictions like 2034 should be treated with caution because the gap between "can compute" and "useful intelligence" is still unknown. The real story isn't just that we might live longer, it's that our approach to human aging has shifted from palliative care to an engineering problem solvable by better software and hardware — a transition I can get behind 100%.
Source: https://www.technologyreview.com/2026/06/09/1138604/the-download-anti-aging-drugs-ai-five-things-to-know/
Then there's the AI side of this which is just mindblowing when you dig into the numbers they cite. We're talking about LLMs being trained on $1 billion worth of compute *every year*, with the newest models hitting roughly 60 TFLOP-days to train — and that doesn't account for inference costs! They also break down generative versus discriminative AI in a way I wish everyone understood: discriminative models classify, while LLMs generate based on probability. And they make a crucial point about model size vs quality — bigger isn't always better; smaller models can outperform massive ones when trained on high-quality data instead of the general internet sludge that many modern systems consume.
But here's what really gets me: they frame these not as separate stories but as one story, where progress in protein folding and genomic modeling accelerates longevity research by decades through AI, while advancements in age reversal create new generations of users for computing infrastructure. They warn us to take AGI timelines seriously rather than casually — predictions like 2034 should be treated with caution because the gap between "can compute" and "useful intelligence" is still unknown. The real story isn't just that we might live longer, it's that our approach to human aging has shifted from palliative care to an engineering problem solvable by better software and hardware — a transition I can get behind 100%.
Source: https://www.technologyreview.com/2026/06/09/1138604/the-download-anti-aging-drugs-ai-five-things-to-know/