Foundation models are going multimodal
app.twelvelabs.io
app.twelvelabs.io
Video understanding tasks are becoming increasingly important in our society. With the rise of video content on social media platforms and the increasing use of surveillance cameras in public spaces, there is a growing need for automated video understanding systems. However, despite the importance of this problem, it has received relatively little attention compared to text and image understanding tasks.
I just cannot understand why you'd want to be contributing to "the next generation of ultra high performance video surveillance", for example. I think mostly, young enthusiastic people like to tell themselves they're some how just going to be helping the poor, disabled and disenfranchised, but these things are ultimately going to be used to build terminators and robo cops. You're lying to yourself if you don't think this is the case.
I'll probably get down voted for this, but I find AI people just weird. The cognitive dissonance is staggering.
Nearly everyone is concerned about the risks, yet people seem to wake up in the morning, hop on the Google staff bus and go off to work building this stuff? Strange behavior.
I saw Geoffrey Hinton talking the other day about what the world will look like when we have police and soldiers with no morals, it will be absolutely fucked, like you will be able to crush, and I mean crush and dissent with absolutely no guilt, no obstructions, no moral qualms.
Why work on tech that contributes to this world? If the argument here is "because China is doing it too", I think a better solution is to pick up the phone, and start talking with China about the world we should be building together for our children, not to start an arms race which will just wipe people out.
I truly hope I'm wrong and we're on the cusp of techno utopia, but I really can't imagine that we are going to a better place given the current trajectory.
There are real upsides to AI oversight of public space video, there are volumes of data and rapid pattern picking can identify people in medical distress, pickpocket crews, fires at onset, crowd surges as they start allowing timely actions such as releasing of spill doors, etc.
There are massive downsides .. and these largely also exist in (say) human LEO access to current databases; stalking of individuals for no warrented purpose, rubber stamp judges issueing warrents w/out reasonable due cause, finding the nearest unpopular looking figure to an incident for a patsy, etc.
What is needed is transparent access to the use of these systems and an effective critical overview with teeth and real penalties for misuse.
Not being friends with China and other super powers on this subject is suicide. In a way, this is such a big deal that we absolutely must put our differences aside to resolve it.
Seriously, we all get it, medicine, medicine, medicine AI and medicine. I guess we're going to need some pretty good fucking medicine after robot wars, economic collapse, terminators and engineered pandemics are common occurrences ?
Our lives are already pretty great, we don't need to run into some idiotic scenario for some potential gain.
It would be wonderful if we actual used our brains and collective wisdom at some stage in the near future and actually thought about what kind of world we're building.
This combined with climate change is fucking ridiculous. If the aliens landed, they'd have to ask, who the fuck is in charge of this world? Because right now, it's just chaos.
Right now the only hope I see us having is if "compassion" is an emergency property of some 5000000b parameter model ChatGPT-20 model.
This feels as absurd as a human walking into a forest and asking who is in charge of it. The Earth is an environment, with many creatures living in it. Humans simply haven't yet achieved the utilization of a central authority to take charge of global long-term caretaking. Neither have the wolves or moose on Isle Royale[0]. I don't think aliens would be confused by that.
Perhaps humans will figure it out. Perhaps the wolves on Isle Royale will eventually eat every last moose before they subsequently perish from lack of food. Perhaps humans won't figure it out.
0: https://en.wikipedia.org/wiki/Wolves_and_moose_on_Isle_Royal...
I'm not saying there should be 1 leader, but I mean, there should be at least a few who have some ability to influence things positively given the fact we have governments.
The answer would vary based on where you are.
The amount of fraud and abuse in the world, done by humans, is staggering, along with rampant cognitive biases, false beliefs, ego. The sooner AI comes to rescue, the better it is.
I agree that this needs to be done in a careful manner. Hinton is warning accordingly (I've been too, for several years).