Even if they are able to put some control on commercial offerings, open source or available to the public models will be around.
See the behavior around Llama when it first launched. I expect an underground movement around LLM’s similar in vibes to the cracker/hacker scene of the 90’s.
Existing models will persist, but I do not see a timeline where R&D and deployment of extremely powerful (GPT-5/6+) models do not receive the same comprehensive scrutiny as the R&D and deployment of fusion weapon systems.
Creating these models require tens or hundreds of billions of dollars and buildings full of high end hardware that is only produced in one or two locations in the world. You can't hide that from world governments like 90s nerds hiding their pirated Playboy scans.
This the same NTT with the following goal on their website for 2030:
> One major feature of Digital Twin Computing is the undertaking to digitally express humans, especially the inner state of individuals. By representing not just the outer state of humans but their inner state as well, it should be possible to achieve advanced interactivity even from a social perspective, such as human mobility and communication.
> Moreover, representing the personality of every person should make it possible to achieve interactivity based on diversity and individual features as opposed to interactivity between individual digital twins without personality that are statistically rounded out as average values.
> We argue that these features will enable the creation of a virtual society in which a variety of things and humans interact with each other in advanced and sophisticated ways beyond the limitations of the real world.
People are losing their mind about this stuff. I have been trying to use ChatGPT and generative products meaningfully in my work for some time now and (for my work at least) I'm seeing it more and more as a gimmick no better than Eliza https://en.wikipedia.org/wiki/ELIZA.
At its best it can perform better than a regular search engine, but at its worst, it is like speaking to an idiot exhibiting advanced Dunning-Kruger - it is always 100% confident, regardless of it having hallucinated key parts of its response.
5 comments
[ 4.7 ms ] story [ 27.0 ms ] threadEven if they are able to put some control on commercial offerings, open source or available to the public models will be around.
See the behavior around Llama when it first launched. I expect an underground movement around LLM’s similar in vibes to the cracker/hacker scene of the 90’s.
Creating these models require tens or hundreds of billions of dollars and buildings full of high end hardware that is only produced in one or two locations in the world. You can't hide that from world governments like 90s nerds hiding their pirated Playboy scans.
> One major feature of Digital Twin Computing is the undertaking to digitally express humans, especially the inner state of individuals. By representing not just the outer state of humans but their inner state as well, it should be possible to achieve advanced interactivity even from a social perspective, such as human mobility and communication.
> Moreover, representing the personality of every person should make it possible to achieve interactivity based on diversity and individual features as opposed to interactivity between individual digital twins without personality that are statistically rounded out as average values.
> We argue that these features will enable the creation of a virtual society in which a variety of things and humans interact with each other in advanced and sophisticated ways beyond the limitations of the real world.
https://www.rd.ntt/e/iown/0003.html
Sounds a bit like this is really about "generative AI for me but not for thee."
At its best it can perform better than a regular search engine, but at its worst, it is like speaking to an idiot exhibiting advanced Dunning-Kruger - it is always 100% confident, regardless of it having hallucinated key parts of its response.