Sure it makes a few things easier to execute when you have an agent running locally. But many people fear what such an agent might do under a wrong or misunderstood prompt to your local system. There is also mistrust of how it may access your data locally compared to a more controlled scenario where you specifically choose which files it has access to. So, no, “everyone” “should” not necessarily be using more CC. It depends on the tasks in hand and risks associated
For a lot of people, such options are not possible, e.g., elderly, sick, bed-ridden, socially challenged or so on. You underestimate the need and impact of such technology.
If the elderly, sick and bedridden can talk to a computer they can also talk to real people through the computer. For the socially challenged I get where you're coming from, but the question is whether it would help them to develop their social skills or whether they would just become dependent on the chatbot and withdraw even further from the real world.
> If the elderly, sick and bedridden can talk to a computer they can also talk to real people through the computer.
If this is so dependably true, then why does this population still suffer loneliness? How many hours of your day do you devote to zoom chats with lonely old people?
I ultimately support your side of the debate - that digital simulation is not a true medicine - but without being honest about humans and human nature, you create empty arguments.
Some in this thread call pornography useless, call cigarettes useless, call drugs useless. However those products are used endlessly and with great satisfaction by billions of humans throughout thousands of years. It's simply not true - these things are useful to humans, quite obviously.
What causes this puritanical disposition regarding human needs and satisfaction while also literally arguing for humanism vs synthetic encroachment? Without acknowledgement of what humans actually are we won't find a real reason that AI chatbots shouldn't replace them.
You speak as if the "real world" for a bedridden and lonely individual is something that they should just endure and enjoy, but you would never choose to switch places with such a person.
I was once contacted by FinalSpark where they offered free early remote access to use their biocomputing platform. The platform is accessible remotely and allows experiments on neurospheres made from living cells, sitting in an incubator, in their lab, in Vevey, Switzerland. A neurosphere is a round structure build out of approximately 10’000 neurons, connected to electrodes in different places. The platform uses python scripts to communicate with the neuron allowing for various functionalities, such as: Stimulate living neurons, Read data from neurons, Log all the data in a database, and Display graphically the results of experiments for further analysis.
I was too busy to come up with a clear project idea that could beat alreadty existing stuff such as neurons playing Doom [0] (not related to FinalSpark). Still waiting for someone to show something cool using this platform.
As someone actually in this space, does the "rental" concept give you any concerns about the quality of research this can support? Like, if the previous customer's use of the organoids will have stateful impacts that impact what you observe? It strikes me that with conventional computers in the cloud we have pretty straight-forward assurances that each customer gets the instance in a fresh state.
How hard is it to do something like this on your platform [0]? Are there any other real-world examples where people are using your platform to do something similar?
If you talk about learning to play doom, I would say it is unrealistic at this point and still a topic of research, the purpose of the neuroplatform is precisely to find reliable ways to train neurons to perform a specific task.
Well, we already have systems with 32 electrodes, and we are looking at alternative to increase this by several orders of magnitude. I hope we have something next year.
I'd imagine its easy to set up independent platforms for different users. Organoids are pretty easy to develop. Large costs come from Multi Electrode Array recording devices that can be >30K.
That's a great question I hadn't thought of. Neurons definitely have state in vitro (internally and inter-neuron, e.g. synapses and tunneling nanotubes).
>So other than training ever-larger models on the same internet data, how can they make better LLMs?
Training a multi-modal model that can integrate audio, visual, text and all sorts of data modalities to human level capabilities still remains an clear challenge. The bottleneck here is not the lack of data imo.
Always amazes me to see that society as a whole has been and continues to be willing to use public (or private) funding to support natural sciences. In the short term, one often faces the argument about the meaningless of doing things just for the sake of knowing or understanding nature. But in the long run, attracting scientists and engineers to work on such problems must (insert my optimism) add a significant value to our society as a whole.
reply