Eric Schmidt’s Post

Of course it is a race. As in every race safety matters. We invented biotechnological ways to shorten plant breeding period, and loose the opportunity to test them carefully. And now we can afraid of gene drifting and interactions. We raised the stakes even further - and AI helps us. However every model has hallucination rate: 10-30 percent. First we accepted we know everything about genes, however it was false, and now we accept AI is exact, however it is not. Bias x bias = ? In this increased racing situation I hope you can take your time for more careful testings.

Eric Schmidt have you seen PyLabRobot?

The dominant friction is cross-institutional verification cost. Too many cylinders generate. Too few cylinders compose. The danger is that variation outruns the system’s ability to verify, translate, and retain it across key institutions and market execution. That means the epistemic debt accumulates whenever discovery is celebrated before reproducible transfer is secured. Epistemic debt = the gap between local breakthrough and system-wide verified adoption. Composition is the constraint. Verified transfer is scarce. The system fails when it cannot reconcile what it generates. The dominant friction is coordination cost under verification. Industrial leadership is retained structure. Where variation exceeds cross-institutional verification, debt accumulates invisibly. When debt accumulates, each new program increases friction faster than capability. The strategic question is who can close the loop with less epistemic loss. America’s risk is flood without closure.

The patent lead and VC depth are real advantages, but the critical variable is time-to-deployment. China's integrated state-commercial pipeline compresses that cycle. The US innovation engine is stronger—the coordination infrastructure around it still isn't.

High-minded good intentions - Like all good screamplays start - https://lnkd.in/gcVJzRsy

Based on what I've written previously, I'm confident that AI isn't yet capable of fully merging with biotechnology. What seems like a huge achievement to us today will seem insignificant in five years. This doesn't happen in nature. The technological advances that nature is built on don't lose their properties or value. Nature's biotechnology is too intelligent, versatile, and efficient. Nature has achieved 100% energy consumption, while we experience losses of over 60-70% when using energy.

Eric, AI is progressing thanks to the slow progress of HBM for GPUs, which can't make significant progress because the MEMORY WALL PROBLEM for DRAM (HBM) hasn't yet been solved. Over the past 30 years, DRAM (Trc) progress has ranged from 60ns then to 45ns today. According to ITRS Roadmap from 2003, today's DRAM (Trc) should be 1-3ns. ITRS then thought that CPU frequency would reach 53.2 GHz in 2018. According to this logic, by 2025, it should have reached 70 GHz. Over the past 30 years, all DRAM progress has been in increasing frequency and capacity, but today we've reached the limit. Future problems with the impossibility of developing HBM4 using existing methods will prove this. All future speedups lie here. The Chinese are already at this point in solving this problem. If they solve this MEMORY WALL PROBLEM, they won't need EUV for CPU/GPU manufacturing. But it seems to me that their entire current quest to achieve EUV properties is connected to this concealment of the fact that they are currently moving in a different direction toward solving the MEMORY WALL PROBLEM. I don't see US investment capital actively attempting to solve this problem. It's being tackled by enthusiasts who are struggling to find funding for R&D.

I’m not a doctor, but through deep experimentation with AI, I’ve designed a device that addresses diabetes challenges. Current AI evaluations and simulations are looking incredibly positive