Biostate AI reposted this
Can AI replace scientists? Not yet… not fully at least. But it can already do a surprising amount of work—enough to accelerate science 15 to 20x. With David A. Sinclair A.O., Ph.D. lab at Harvard University, we tested our multi-agentic AI scientist, K-Dense, on the ArchS4 transcriptome dataset (~2M samples). Within a few hours, the system produced results that took us days to verify, but in the end the work that normally takes a year or more was done in just three weeks. The result, it produced biological clocks that outperformed all existing clocks and, more uniquely, knew when its predictions were wrong. This confidence metric for predictions is first of its kind and a feature with real clinical potential. The bigger story: K-Dense carries out the entire research cycle end-to-end. In a single cycle it does deep literature review, generates multiple hypotheses, analyzes all data it has access to, builds models if needed, evaluates the results against existing literature, and even drafts a paper for "peer-review". It's not a replacement for human creativity or removing need for scientists, but using it feels like superpowers for researchers. And this is the worst the system will ever be. Within 1-1.5 years it will be much better. Imagine doing as much as a career scientist accomplishes in an entire lifetime, in a matter of weeks. What questions would you ask? What experiments would you pursue? We’re opening up K-Dense to design partners and will release a free academic version soon at "k-dense dot ai". We want to figure out the limits of our system and we are looking actively for folks to push it to its limits, break it :-). Reach out to me, Vinayak Agarwal, Timothy Kassis or sign-up to our waiting list for access to the system. Thanks to our friends at Google and Google Cloud, K-Dense is fully enterprise ready, so if you’re worried about data-security and all that jazz, you don’t need to be. #longevity #bioai #genai