Compressing drug design time from years to hours The drug development that used to take months or even years to complete now only requires 24 hours of calculation and a few weeks of laboratory testing. In the future, we may be able to customize the drugs we each need like customizing a suit. In a recent in-depth interview, Josh Meyer, co-founder of Chai and former OpenAI employee, shared their latest progress. Josh Meyer realized early on that since big language models can understand English and French, they should also be able to understand * real natural language - DNA and protein sequences. This prompted him to devote himself to AI pharmaceuticals With the improvement of model capabilities, the speed and quality of AI designed drugs are exponentially increasing. Just a year ago, the success rate of computer designed antibodies in the laboratory was only 0.1% (one in a thousand or one in ten thousand) And their latest Chai 3 not only has a success rate close to 30%, but more importantly, it has achieved a leap in molecular weight: in the past, AI generated molecules still required a lot of manual optimization, while Chai 3 generated molecules have already fully loaded various attributes and can even enter the subsequent stages without modification. Meyer referred to this as "Zero shot" drug screening. At the microscopic level, Chai's model is as precise as an atomic level microscope. Compared to AlphaFold's accuracy of only about 3% in predicting antibody antigen complexes a few years ago, Chai's model has improved by an order of magnitude. Under cryo electron microscopy, the error between the structure predicted by AI and the actual physical structure is less than the diameter of an atom. Pharmaceutical giant Eli Lilly conducted extreme stress testing on Chai's model internally (applied in a large number of real projects). The authentic feedback provided by Eli Lilly (what is effective and what is not) greatly helped Chai obtain the data needed for the model and establish a future research and development roadmap. When asked which disease Chai would prioritize conquering, Meyer replied: At the molecular level, AI models do not care whether it is treating cancer or immune diseases. As long as there is a clear target, the model can develop countless projects in parallel, greatly reducing the time to find the corresponding drug. In addition, due to AI designed drugs having more precise and stronger therapeutic effects (with larger amounts of effect), this may even make subsequent lengthy and expensive clinical trials easier, as drugs with significant effects are more likely to be statistically proven effective. Chai's business model is very clear and identical to top AI labs: * * the better the technology ->the more customers use it, the greater the value created ->earning more funds to purchase computing power ->training the next generation of stronger models * *. In the pharmaceutical industry, which is extremely large and willing to pay for "success rate and speed", even being one year ahead of competitors can translate into billions of dollars in value. Therefore, in this field, people place more emphasis on the "upper limit" of AI (how good drugs can be made), rather than simply "cost reduction" (how much money can be saved). The rise of Chai once again confirms that biology has officially become a branch of computer science. When 24-hour calculations can replace years of trial and error, humans may only be a few iterations away from conquering incurable diseases.