When will we have AGI?
>be epstein>do bonghit>drink wine cooler>finger 15-year old>say some dumb shit like "we should have agi">throw insignificant pocket change from your trillion dollars at it, already forgotten the next day[a few moments later]>childless schizo incel loser with infinite free time and zero awareness on internet: "EPSTEIN IS BEHIND THE ENTIRE AI PSYOP!!!!"
>>108997827never. AGI is a meaningless buzzword that exists purely to move goalposts forever. Nobody can define consciousness, intelligence, sentience, etc. so making these things goals just serves to make sure they can never be reached
>>108997847Your are a retard, the definition of each of these words are simple, not our fault if you fail to grasp them.Though we are decades away from agi at the very least.
>>108997827
>>108997963>Though we are decades away from agi at the very least.Decades or centuries or millenia? The fact you answered decades implies you think there's a current avenue of research that will actually achieve AGI, which is it?
>>108997827Not any time soon. And it won't come from current AI efforts. AGI requires the ability to learn and adapt. Current AI has a fixed model incapable of learning anything new.
One would expect that the main qualifying factor of AGI is the ability to learn and adapt to new tasks with as little information possible, unlike the current paradigm of overfitting massive models and then performing "in-context learning" or further fine-tuning.The way I see it, you can scope two levels of AGI: Virtual and Physical.Virtual AGI is kind of like the agentic assistants people were raving about the couple of previous months. They interact with UI, locate relevant task information, learn how to use programs or write their own when no optimal solution is found. LLMs in theory can already do this, but because they are not persistant and non-self-supervising they eventually fuck up or get stuck.Physical AGI on the other hand would have to be able to figure out how to process several different sources of information (visual, audio, force, LIDAR, RADAR, etc) which can add up to a lot of information very fast. You can fine-tune an LLM to do some of it, but trying to get them to operate at reasonable speed is a pain in the ass and requires compromise of some sort and immense levels of compute.Because of lower complexity, a Virtual AGI should be easier to create than a Physical one. But if the Virtual AGI you created is actually AGI, then it should be able to turn itself into a Physical AGI by learning and adapting, as also stated by >>109000426, which is what all the AI safety people fearmonger about. They, in their arrogance, don't realize they are crying wolf by claiming LLMs are it.I disagree about the statement that it won't come from current AI efforts. I agree insofar that it won't be an LLM that becomes AGI, but LLMs might contribute to stuff like generating complex decision trees or other intellectual busy work that becomes unmanageable for people after a certain level of complexity, and these decision trees (or some other form of algorithm) may contribute to the actual AGI model.Kind of like the interaction between Protein, RNA and DNA.
>>109000339I said at the very least, the idea that it can be done on computers is an assumption i'd not bet on.Though i'd count bio engineering as AI.
>>109000790I forgot to mention that Physical AGI is needed for robotics to interact with the world and therefor "Physical", though a "Virtual" AGI should also be able to take control of an already function robot control system, which shows how muddied this separation actually is.Anyone who has worked with computers simulators and real world hardware know that there is a major gap between the two and model being able to seamlessly overcome this gap would separate a true AGI that can create novel robotic solutions versus a very advanced virtual model than can control already created systems but not improve upon them, or rather it's attempts to improve them would run into calibration issues
as long as silicon valley keeps trying to optimize LLMs into magically becoming AGI, never
