![](/static/253f0d9b/assets/icons/icon-96x96.png)
![](https://lemmygrad.ml/pictrs/image/7c8a8314-de0a-4701-9761-fc8abfb9c9c7.jpeg)
Im curious how each agent differs, or is trained. Seems they had doctor and nurse agents, as well as patient agents. This would be a good way to start partial implementation. It would allow some tasks to be taken over by the in a hybrid format which could allow an even richer training environment.
I could never see the west doing this in a way that would actually improve the quality of service.
One of the issues with LLM AIs that we’ve seen time and again is that it can be extremely confident and perfectly incorrect. I have no doubt they are doing their best to train the AI with the best data, but I hope they are also working to solve some of the underlying issues with LLMs.
Flow batteries seem very promising, but the chemistry required needs more scale/external funding to be viable.
There were some thermal battery retrofits for coal power plants using carbon and steam that looked interesting in principal, though cost and logistics are not fully solved problems, and the round trip efficiency was rather bad compared to other storage methods.
There were also some molten metal batteries that have been working towards useful scale over the past decade or so. They had cheap and abundant matetial inputs and significantly long charge discharge.
There are many neat options out there. I think researching and building out each as they become viable would help to improve system resiliency and long term viability.