We're all project-managers now. The trust required of the LLM is no different than that that required when a PM trusts a dev-team to implement correctly. As a PM, you need to understand testability and write that into the requirements.
Nice point. I agree with this, and it implies that it's useless to apply the category of creativity to LLMs (at least wrt interpolation and extrapolation), and even more useless to suggest that we'll cross the creativity threshold once the models have been sufficiently embiggened.
The more I read these think-pieces on AI the clearer it is to see that they have very little to do with LLM technology, and are really just talking about the collapse of the set of values constructed during the enlightenment. The sociocognitive atrophy that you describe is inherent to the idea that newness and progress are the ultimate goal that we should organize society around. The problem is that newness and progress can only be defined relative to a status-quo, and hence are incoherent goals in themselves. When pursued far enough, they become their own strange, monstrous status-quo that betrays the original intentions of the people who pursued these values. Hence "we mistake the flattening for progress". This was the case before AI, it's just that AI makes it much harder to ignore, and in many ways encapsulates the problem.
The point w/ electronic calculators is the same point made by Plato regarding books. It used to be easy to laugh off these concerns, not so much today. Imo, this is the real progress: people are now asking meaty questions regarding the ultimate human purpose of books, calculators and technology.
wow - so you pmuch baked this kind of testing into your day to day work. Might not be a bad startup idea to create a harness that can do this automagically.
How do you go about comparing models? Personally, I've found that the quality of output is overwhelmingly dictated by the quality of the input I give it - and that the quality of input differs massively depending on the task at hand, how much documentation exists, what kind of mood I'm in, etc.
Hence, I'm beginning to believe that unless you're doing very specialized work - the model itself has become a commodity, incremental quality is irrelevant and that economics and privacy are what actually matter now.
> I think the only way to find out what data is meaningful is to collect and analyze more of it
So the idea is to just muck around with data, then ???, then make people healthier? To a hammer, every problem looks like a nail I suppose.
I don't work in healthcare, but it seems to me that the main problems in the field are:
1) a focus on addressing symptoms, not causes
2) pathologization of normal processes
3) normalization of pathological processes
4) financialization of care + doctor evaluations
5) regulatory capture by care providers
1, 2 and 3 are inherently philosophical problems, and there's no amount of data that you can toss at these problems to solve them. Thinking that data can solve these problems is itself part of the problem.
All I want is an AI that can take in basic information about my demographics, lifestyle, family history, religious beliefs, symptoms and vital signs - and then provide me information on tests I should run and drugs I should take - and then most importantly : tell me how to obtain those tests and drugs without ever dealing with some doctor who's 200k in debt from medical school and needs to appease their administrator by recommending x-many surgical procedures a quarter.
The incentives are bad - not the data or lack thereof.