Are groups introducing new molecules registering their experiments and methods?
Your cell-free methods can have positive impacts. But, the Internet is full of escaped chemicals, molecules, clusters and organisms from the past. Do you have global protocols, that can be audited, for your new things when many species, including humans, insects, plants, bacteria and viruses are exposed? What about colonies in aerosols in air?
Most of the serious issues on the Internet now are taking 20 or more iterations to find 99% of the variations. It is not just one or a handful of factors, parameters, independent events or entities, but dozens, with millions of instances. Those LLMs with only a few rigid attention layers are not performing well because the underlying processes themselves have many more degrees of freedom. The LLM makers are using “one size fits all”, and beginning groups in every “new” area are re-inventing the basics.
Best to start at global level, and use global distributed gathering, verification, indexing, global open tokens, archiving, and lossless sharing. When you begin where all countries, groups, humans and AIs can collaborate without error, at least your records, coding, communication and storage will be stable and verifiable.
Even writing about it is hard, as the terms vary from group to group, even when they are all talking about how they handle rare events. And “rare events” or factors, when new groups begin, are often at parts per hundred or parts per thousand level. A part per million error, these day, can mean 8100 deaths or extended health problems. A part per thousand error can bring down a plane or rocket. Covid was one, or a few human groups, but using off the shelf components and a “let’s see what happens when we do this” attitude, with no monitoring or protocols can cascade.
I do not have a simple name for it, because analyzing the propagation of information and its physical products and impacts is something hundreds of millions of groups all face, and all invent their own ways. None of the LLM or search engine groups, and almost none of the research and collaboration facilitator groups are sharing their methods and experiences in global open form. So if they make mistakes, it will not be easily traceable. Tracing “covid” years after the fact is hard, to impossible. I am checking plastics, chemicals and nuclear materials. LLMs introduce mistakes simply by using out of date training data, highly biased data, unverified and untraceable data and methods. When future courts place blame, they might simply award damages when there is sufficient evidence to clear overloaded courts.
I think it is better if the industries and disciplines self regulate and keep lossless records. The governments and industries and user communities are not well enough informed to not make mistakes. The systems are poorly designed and allow mistakes at parts per hundred level, even if pieces are ppm ppb or even ppt. Every case I investigate, it usually comes down to “many groups working on something” but not working together globally. Thousands of small universities and corporations and individuals can work together, and the collaboration sites can help monitor, record, and look for new issues and new opportunities.
Using human deaths and hospital cases is NOT a good way for society to track mistakes in vaccines, in contaminants, in teaching people without background in watching for impacts, “new-to-nature molecule” making methods.
Filed as (Are groups introducing new molecules registering their experiments and methods?)
Richard Collins, The Internet Foundation
