The headline was confusing and reading the article doesn’t really clear things up. I don’t think Gill is imagining the same sort of “pretend person” that I would want out of AGI. What I want is a personal assistant that knows me extremely well, is able to tirelessly work on my behalf, and has a personality tailored to my needs and interests. It should be general enough to understand me on a personal level and do a good job anticipating what I want.
knows me extremely well, is able to tirelessly work on my behalf, and has a personality tailored to my needs and interests.
Those may still be ANI applications.
Today’s LLM’s marketed as the future of AGI are more focused on knowing a little bit about everything. Including a little bit about how MRIs work and a summary of memes floating around a parody subreddit. I fail to see how LLM’s as they are trained today will know you extremely well and give you a personality tailored to your needs. I also think commercial interests of big tech are pitted against your desire for “tirelessly work[ing] on my behalf”.
What I want is a personal assistant that knows me extremely well, is able to tirelessly work on my behalf, and has a personality tailored to my needs and interests.
and you’re not concerned at all about this information being compromised and used against you?
Of course I’m concerned about it. That’s why I would take measures to ensure the information is well protected. I already run local LLMs and image generators for most of the stuff I use AI for, both to ensure that I have control over what sort of outputs they generate and to keep any inputs I run through them private. An AGI assistant like what I’m describing is something I would want to run on my own hardware too.
Yes, I do. Perhaps not the current generation of hardware, but the chip manufacturers are currently throwing hundreds of billions of dollars into designing the next generation of AI-specialized hardware so I expect the next generation to be very impressive. The software has also been getting more efficient, making better use of the hardware that already exists. I’ve been experimenting a lot with it.
The headline was confusing and reading the article doesn’t really clear things up. I don’t think Gill is imagining the same sort of “pretend person” that I would want out of AGI. What I want is a personal assistant that knows me extremely well, is able to tirelessly work on my behalf, and has a personality tailored to my needs and interests. It should be general enough to understand me on a personal level and do a good job anticipating what I want.
That would not at all be a waste of energy to me.
Those may still be ANI applications.
Today’s LLM’s marketed as the future of AGI are more focused on knowing a little bit about everything. Including a little bit about how MRIs work and a summary of memes floating around a parody subreddit. I fail to see how LLM’s as they are trained today will know you extremely well and give you a personality tailored to your needs. I also think commercial interests of big tech are pitted against your desire for “tirelessly work[ing] on my behalf”.
Depends on how much energy it takes. If it takes more resources than it frees, then I’d say it is not worth it.
I am quite sure it’ll cost less than it would to hire a human for the job.
I’m talking about the energy and resources to actually create and provide this service.
So am I.
and you’re not concerned at all about this information being compromised and used against you?
phew…
Of course I’m concerned about it. That’s why I would take measures to ensure the information is well protected. I already run local LLMs and image generators for most of the stuff I use AI for, both to ensure that I have control over what sort of outputs they generate and to keep any inputs I run through them private. An AGI assistant like what I’m describing is something I would want to run on my own hardware too.
Do you really think you’ll be able to run a full fledged, real-feeling AGI on home hardware?
Perhaps an assistant, maybe…
but good on you for forethought.
Yes, I do. Perhaps not the current generation of hardware, but the chip manufacturers are currently throwing hundreds of billions of dollars into designing the next generation of AI-specialized hardware so I expect the next generation to be very impressive. The software has also been getting more efficient, making better use of the hardware that already exists. I’ve been experimenting a lot with it.